TW201306793A - Method and apparatus for sleep scoring - Google Patents

Abstract

A sleep scoring method is provided. An electroencephalogram or an electrooculogram of a person measured in a testing duration is first analyzed with a multi-scale entropy analysis, so as to generate plural entropies. Based on the entropies, the sleep status of the person in the testing duration is evaluated.

Description

Method for judging sleep state and judging device

The present invention relates to a sleep state determination mechanism and, in particular, to a method and apparatus for automatically determining a sleep state.

In most people's lives, sleep occupies nearly a third of the time. However, many people suffer from long-term sleep-related problems (such as insomnia or obstructive sleep apnea), resulting in a compromise in overall quality of life. How to correctly assess the sleep state of the subject, and then seek a program that can effectively improve the quality of sleep, has long been a topic of concern in the field of medical engineering.

Most of the existing sleep state judgment methods measure the electroencephalogram (EEG), electrooculogram (EOG) and electromyogram (EMG) after the subject falls asleep, and will record it. The result is submitted to the expert for manual identification evaluation, and the physiological state of the subject during the measurement period is judged according to the physiological signal obtained during each measurement period (for example, in units of 30 seconds). According to the Rechtschaffen & Kales rule, the sleep state can be divided into: the awake phase, the non-rapid eye movement phase (including the first phase to the fourth phase), and the rapid eye movement (REM) phase. In recent years, the third and fourth phases have been collectively referred to as the slow wave sleep (SWS) phase.

Since it is quite time-consuming and labor-intensive to manually evaluate the sleep state by experts, there is also a method for automatically determining the sleep state, and the instrument is based on the characteristics of the waveform, spectrum, amplitude, etc. of the alpha wave, the spindle wave or the slow wave in the physiological signal. Rechtschaffen & Kales rules to determine the sleep state of the subject. However, most of these automatic judgment methods need to collect a plurality of physiological signals at the same time as a basis for judging, so it is necessary to install a plurality of detecting devices on the subject. As a result, in addition to the preparation of the measurement is very cumbersome, for the subject, it is nothing more than adding extra interference, but it leads to a decline in sleep quality.

In addition, whether the sleep state of the subject is judged based on one or more physiological signals, there is a problem in the prior art that the first stage and the fast eye movement stage cannot be clearly distinguished.

In order to solve the above problems, the present invention proposes a method and apparatus for determining a sleep state. According to the judgment method and the judging device of the present invention, the multi-scale entropy (MSE) is used to analyze the brain wave signal or the eye movement signal of the subject, and the single-channel physiological signal can be measured to automatically determine the sleep state. And it has been proved by experiments that it can achieve quite high correctness. In addition, by combining the multi-scale entropy analysis and the autoregressive model operation, the judging method and the judging device according to the present invention can effectively distinguish the first stage and the fast moving eye stage.

According to an embodiment of the present invention, a sleep state judging method includes the following steps: (a) obtaining a brain wave signal or an eye movement signal of a subject during a measurement period; (b) performing a physiological signal for the physiological signal Multi-scale entropy analysis to generate a plurality of entropy values; and (c) determining a sleep state of the subject during the measurement period based on the plurality of entropy values.

According to another embodiment of the present invention, a sleep state determining apparatus includes a collecting module, an analyzing module, and a determining module. The collection module is configured to obtain the physiological signal (brain wave signal or eye movement signal) of the subject during a measurement period. The analysis module is configured to perform a multi-scale entropy analysis on the physiological signal to generate a plurality of entropy values. The determining module is configured to determine, according to the plurality of entropy values, a sleep state of the subject during the measurement period.

Since the judging method and the judging device according to the present invention can be designed to be fully automated, a large amount of time and effort for manually interpreting the signal can be omitted. In addition, the judging method and the judging device according to the present invention can generate the judgment result according to the signal of the single channel, and can reduce the interference of the detecting device on the subject. The advantages and spirit of the present invention will be further understood from the following detailed description of the invention.

Referring to FIG. 1, FIG. 1 is a flowchart of a sleep state determination method according to an embodiment of the present invention. Step S12 is first performed to obtain a physiological signal of the subject during a measurement period. The physiological signal can be a brain wave signal or an eye movement signal. In practice, a multiple polysomnographic (PSG) instrument can be used to measure and record one or more physiological signals required in step S12, such as F3-A2 brain wave signal, F4-A1 brain wave signal, C3-A2. Brain wave signal, C4-A1 brain wave signal, P3-A2 brain wave signal, P4-A1 brain wave signal, left eye eye movement signal, right eye eye movement signal, and the like. Because sleep belongs to brain activity, the closer the physiological signal is to the brain, the more faithful it is to the sleep state; in general, the brain wave signal is better than the eye movement signal, the eye movement signal is better than the motor current, and the motor current is excellent. In blood oxygen capacity.

Step S14 is to perform multi-scale entropy analysis on the physiological signal to obtain a plurality of entropy values. The multi-scale entropy analysis evaluates the complexity of the time series based on the entropy values corresponding to a plurality of different time scales τ. The sample entropy is used as an example to explain how to perform multi-scale entropy analysis on physiological signals.

Suppose a brain wave signal is sampled to become the time series of length N _{{x 1, ..., x N} }. For a scale τ, this time series is divided into a plurality of non-overlapping windows each having a length of τ. After averaging the data in each window, the scale τ corresponds to a coarse-grained sequence y _τ (j) of length N/τ, and the elements in the consecutive coarse sequence are in each window. Average value of the data. Each element can be expressed as:

The consecutive coarse sequence Y _τ corresponding to the scale τ can be expressed as:

After obtaining the elements in the consecutive coarse sequence, the sample entropy of the consecutive coarse sequence can be calculated accordingly, as explained below.

A sequence vector of m dimensions is defined as follows:

μ ^{( m )} ( i )={ y _τ ( i ), y _τ ( i +1),..., y _τ ( i + m -1)}. (Formula 3)

The distance between two vectors m ^{( m )} (i) and μ ^(m) (j) each having a length of m is defined as:

The parameter r is defined as an acceptable similar tolerance. If d(i,j)≦r , the two vectors μ ^(m) (i) and μ ^(m) (j) are judged to be similar. The similar probability of these two vectors can be expressed as follows:

Where d(i,j)≦r , ω _j =1, otherwise ω _j =0.

In Equation 5, j ranges from 1 to (N/τ) between integers and is not equal to i .

The probability that there are m points in the two sequences is calculated according to the following equation:

The odds of having m + 1 points in the two sequences are:

For the sample entropy, in the calculation of Equations 6 and 7, the range of i and j is between 1 and (N/τ). The sample entropy corresponding to the scale τ is:

When performing the above calculation, m = 2 can be selected and r = 0.15 × SD ( SD stands for the standard deviation of the original time series).

In multi-scale entropy analysis, the sample entropy corresponding to each scale τ is calculated. The multi-scale entropy analysis performed in step S14 may cover the scale range 1~20, but is not limited thereto.

Subsequently, step S16 is performed to determine the sleep state of the subject during the measurement period based on the plurality of entropy values. Figure 2 (A) is a multi-scale entropy analysis example of a subject actually obtained by the above method. The horizontal axis is the number of the measurement period (in chronological order), and the vertical axis is the scale. The different colors in the figure represent the sample. The size of the entropy. For example, if the point corresponding to the 100th measurement period and the scale τ is equal to 3, the figure is bright blue, indicating that the scale 3 is obtained according to the multi-scale entropy analysis of the physiological signal measured during the 100th measurement period. The sample entropy is equal to 0.8. In this example, the scale of the multi-scale entropy analysis performed in step S14 ranges from 1 to 20, so each measurement period corresponds to 20 sample entropy. Figure 2 (B) is a legend for further arranging the analysis results of Figure 2 (A). The horizontal axis is also the number of the measurement period, and the vertical axis is the average value of the 20 sample entropies corresponding to each measurement period. .

Figure 2 (C) is the result of an expert's manual analysis of the same physiological signal according to the Rechtschaffen & Kales rule. The horizontal axis is the number during the measurement period, and the vertical axis is the five sleep states, from top to bottom (also from shallow to Deep) is Wake in the awake phase, REM in the rapid eye movement phase, the first phase S1 in the non-rapid eye movement, the second phase S2 in the non-rapid eye movement, and the slow wave sleep phase SWS. Figure 2 (C) is used as the control group of Figure 2 (B). Comparing the two graphs, it can be seen that the average value of the sample entropy during each measurement period has a high correlation with the depth of the sleep stage, and the correlation coefficient Up to 0.7628. It can be seen from this that step S16 can include determining the sleep state based on the average of the plurality of entropy values.

Figure 3 is a statistical legend of the analysis results from 20 subjects, with the horizontal axis as the scale and the vertical axis as the average entropy. For example, after all the average entropy values corresponding to the scale 5 and determined to be the SWS stage according to the judgment method of the present invention are averaged again, the points corresponding to the scale 5 on the lowest SWS curve in the graph are obtained; The range indicated by the I-shape above and below indicates the standard deviation. It can be seen from Figure 3: (1) The entropy values of various sleep stages will increase with the increase of the scale; (2) For each scale, the entropy value and the depth of sleep are negatively correlated, and Wake from the awake stage The entropy value of the SWS to the sleeping stage is monotonically reduced; (3) the curve overlap of the S1 stage and the REM stage is very high; (4) after the scale is greater than 13, the curves of each stage tend to be gentle. It can also be seen from Figure 3 that in scales 1-8 and scale 20, except for S1 and REM, the entropy values are quite different between the two phases; between scales 9-19, it is between all phases. The entropy values are all very different. According to the above observations, the multi-scale entropy analysis performed in step S14 can be limited to only cover the scale range 1~13, but not limited thereto.

It should be noted that the experimental results presented in Figure 2 (A), Figure 2 (B) and Figure 3 are based on the single-channel C3-A2 brain wave signal of the subject. It can be seen from the above that with the judging method according to the present invention, a relatively good judgment effect can be obtained only when a single physiological signal is measured. Similar experimental results can be obtained by replacing the physiological signal obtained in step S12 with the eye movement signal.

In another embodiment, the determining method further includes an autoregressive model operation program to improve the discrimination of the S1 phase and the REM phase. Referring to FIG. 4, step S12 and step S14 in this embodiment are the same as those shown in FIG. 1, and therefore are not described again. As shown in FIG. 4, while performing step S14, step S18 may be performed to perform an autoregressive model operation procedure on the same physiological signal to generate a plurality of autoregressive coefficients. An autoregressive model is a parametric model used to describe a steady-state time series whose model parameters can be used to determine brainwave states. In the autoregressive model, the existing signal x(t) is represented as the sum of the weights of its previous value x(ti) and an uncorrelated error ε (t) , expressed as follows:

Where a(i) is the autoregressive coefficient and p is the number of autoregressive models. In practice, step S18 may be used to generate eight autoregressive coefficients when the autoregressive model operation program is performed. For a detailed description of the autoregressive model, refer to the paper "Dynamics of human sleep EEG." published by Olbrich et al. in Neurocomputing, 2003, and the paper "Advances in quantitative electroencephalogram" by Thakor et al., Annu. Rev. Biomed. Eng. Analysis methods", and in 2007, Berthomier et al., "Automatic analysis of single-channel sleep EEG: validation in healthy individuals".

Step S20 is to determine the sleep state based on the plurality of entropy values obtained in step S14 and the plurality of autoregressive coefficients obtained in step S18. In this embodiment, step S20 is based on the eight auto-regressive coefficients generated in step S18 and the thirteen entropy values generated in step S14 (corresponding to scales 1-13) as the basis for determining the sleep state, but not limited thereto. . In other words, each of the measurement periods has twenty-one features for determining the sleep state.

In an embodiment of the present invention, step S20 is to analyze the isentropic value and the autoregressive coefficient to determine a sleep state by using a linear discriminant analysis (LDA) program, that is, select from the foregoing five sleep stages. The classification that best fits the characteristics of a measurement period. For a detailed description of the linear discriminant analysis, refer to the paper "Nonparametric weighted feature extraction for classification" published by Kuo et al. in IEEE Trans. Geoscience and Remote Sensing in 2004 and the paper published by Lin et al. in EURASIP Journal on Applied Signal Processing in 2008. Nonparametric single-trial EEG feature extraction and classification of driver's cognitive responses".

Fig. 5 is a diagram showing another detailed implementation flow of the sleep state judging method according to the present invention. Compared to FIG. 4, the method presented in FIG. 5 further includes a reduced sampling step (S22), two filtering steps (S24, S26), and a smoothing step (S28), which are described below.

Step S22 is a down sampling process for the physiological signal to adjust the amount of data of the subsequent operation according to requirements. For example, the sampling frequency can be 256 Hz, but not limited thereto.

Step S24 is for filtering high frequency noise in the physiological signal before multi-scale entropy analysis. For example, step S24 may include a band pass filter, and the band pass band may be approximately 0.5 Hz to 30 Hz, but is not limited thereto.

In step S26, the signal generated in step S24 is filtered again to generate a signal suitable for the autoregressive model operation. For example, step S26 may also include a band pass filter, and the band pass band may be approximately 4 Hz to 8 Hz, but is not limited thereto.

Changes from shallow to deep or deep to shallow in the sleep phase are usually periodic and continuous. Therefore, after using linear discriminant analysis to determine the sleep state during each measurement, this characteristic and the Rechtschaffen & Kales rule can be used to correct some of the misclassified analysis results. Step S28 is to selectively correct the sleep state obtained in step S20 according to a before and after sleep state. For example, if the step S20 determines that the sleep states during the three consecutive measurement periods are the second phase, the fast eye movement phase, and the second phase, respectively, the rapid eye movement phase is likely to be a false positive, so in step S28 Correct this rapid eye movement phase to the second phase.

The confusion matrix shown in Fig. 6 is used to compare the experimental results and the manual interpretation results obtained by the above automatic judgment method. A total of 8480 measurement periods were evaluated in this experiment, and the autoregressive coefficients were not considered, and only the entropy values generated by multi-scale entropy were used as the basis for judgment (scales 1-13). It can be seen from Fig. 6 that the judgment results of the two methods are the same or different measurement period numbers. For example, the automatic judgment method and the expert both estimate that the number of measurement periods in the second stage S2 is 3278; the number of measurement periods that are evaluated by the automatic judgment method as the second stage S2 and evaluated by the expert as the awake stage Wake Then there are four. On average, the consistency of the judgment results of the two methods is 76.91%. For the Wake phase, the slow wave sleep phase SWS and the rapid eye movement phase REM, the consistency of the judgment results of both methods is higher than 84%.

Figure 7 shows the average consistency between the automatic judgment method and the manual interpretation for each sleep stage when multi-scale entropy analysis with different scales is used. The horizontal axis of these six graphs is the number of scales used, and the vertical axis is the average consistency (%). As can be seen from Fig. 7, when the number of scales used in the multi-scale entropy analysis in the judging method of the present invention reaches 13 (i.e., considering scales 1-13), the average consistency of the two modes tends to be stable. It can also be seen from Fig. 7 that compared with the single-scale entropy analysis, the multi-scale entropy is used to analyze the sleep state to obtain a higher average consistency, that is, the effect is better.

Figure 8 is another confusion matrix. In this experiment, the judgment method according to the present invention simultaneously considers the entropy value and the autoregressive coefficient generated by the multi-scale entropy. More specifically, this experiment uses thirteen entropy values and eight autoregressive coefficients as input signals for linear discriminant analysis for each measurement period. It can be seen from FIG. 8 that after combining the multi-scale entropy and the auto-regressive model operation program, the average consistency of the judgment method according to the present invention and the judgment result of the manual judgment is improved to 85.38%, and in addition to the first stage S1, The consistency of other stages is higher than 84%.

Figure 9 shows another confusion matrix. In this experiment, the judging method according to the present invention simultaneously considers the entropy value and the autoregressive coefficient generated by the multi-scale entropy, and adds the correction procedure as in step S28. It can be seen from FIG. 9 that after adding the correction procedure, the average consistency of the judgment method according to the present invention and the judgment result of the manual judgment is again raised to 88.11%, and the consistency of the other phases except the first phase S1 Both are higher than 86%.

According to another embodiment of the present invention, the sleep state determining apparatus 60 shown in FIG. 10(A) includes a collecting module 62, an analyzing module 64, and a determining module 66. The collection module 62 is configured to obtain the physiological signal (brain wave signal or eye movement signal) of the subject during a measurement period. The analysis module 64 is configured to perform a multi-scale entropy analysis on the physiological signal to generate a plurality of entropy values. The determining module 66 is configured to determine the sleep state of the subject during the measurement period according to the plurality of entropy values. The method of judging in the first embodiment of FIG. 1 can be implemented in the sleep state judging device 60, and the implementation details thereof can be referred to the previous description and will not be described again.

FIG. 10(B) shows a detailed implementation example in which the sleep state determining device 60 further includes other modules. The reduced sampling module 63 is used to reduce the sampling process of the physiological signal. The first filter module 65A is configured to perform a band pass filtering process on the physiological signal, and the band pass frequency band is approximately 0.5 Hz to 30 Hz. The second filter module 65B is configured to perform another band pass filtering process on the physiological signal, and the band pass band is approximately 4 Hz to 8 Hz. The autoregressive module 67 is configured to perform an autoregressive model operation procedure on the physiological signals to generate a plurality of autoregressive coefficients. The correction module 68 is configured to selectively correct the sleep state according to a before and after sleep state. The operation of these modules can be referred to the previous descriptions related to the judgment method, and will not be described again.

Figure 11 shows the average consistency between the automatic judgment method and the manual interpretation when different autoregressive model series are used for each sleep stage. The horizontal axes of the six graphs are the autoregressive model series, and the vertical axis is the average consistency (%). In this experiment, the physiological signals were analyzed by using the autoregressive model series 5 to 20 respectively, and the thirteen entropy values obtained by multi-scale entropy analysis were used as the basis for judging the sleep state. If the consistency of various sleep stages is averaged, it can be found that the average value of the autoregressive model is 8 and the judgment is the best.

Figure 12 (A) is an example of an evaluation of the sleep state of a brainwave signal manually analyzed by the expert according to the Rechtschaffen & Kales rule. The horizontal axis is time (in hours) and the vertical axis is five sleep stages. Fig. 12(B) and Fig. 12(C) are examples of the results obtained by evaluating the same brain wave signal using the judgment method according to the present invention; the difference between the two figures is that the experiment of Fig. 12(B) is not added with the correction procedure. Comparing these three figures, it can be seen that the experimental results after adding the correction program are quite similar to the results of the manual interpretation. In short, it is true that the judgment method and the judging device according to the present invention can achieve a good judgment effect.

Figure 13 (A) is an example of an evaluation of the sleep state of an eye-moving signal by an expert according to the Rechtschaffen & Kales rule. The horizontal axis is time and the vertical axis is five sleep stages. Fig. 13 (B) and Fig. 13 (C) are examples of the results obtained by evaluating the same eye movement signal by the judgment method according to the present invention; the difference between the two figures is that the experiment of Fig. 13 (B) is not added with the correction procedure. As can be seen from these three figures, the judging method and the judging device according to the present invention can also be applied to the analysis of the eye movement signal to achieve a good judgment effect.

In summary, the judging method and the judging device according to the present invention use the multi-scale entropy to analyze the brain wave signal or the eye movement signal of the subject, and only need to measure the single channel physiological signal to automatically determine the sleep state, and has It has been proved by experiments that a fairly high degree of correctness can be achieved. In addition, by combining the multi-scale entropy analysis and the autoregressive model operation, the judging method and the judging device according to the present invention can effectively distinguish the first stage and the fast moving eye stage. Since the judging method and the judging device according to the present invention can be designed to be fully automated, a large amount of time and effort for manually interpreting the signal can be omitted. In addition, the judging method and the judging device according to the present invention can generate the judgment result according to the signal of the single channel, and can reduce the interference of the detecting device on the subject.

The features and spirit of the present invention will be more apparent from the detailed description of the preferred embodiments. On the contrary, the intention is to cover various modifications and equivalents within the scope of the invention as claimed.

S12~S28. . . Process step

60. . . Sleep state judging device

62. . . Collection module

63. . . Reduced sampling module

64. . . Analysis module

65A. . . First filter module

65B. . . Second filter module 65B

66. . . Judging module

67. . . Autoregressive module

68. . . Correction module

FIG. 1 , FIG. 4 and FIG. 5 are flowcharts of a sleep state determination method according to a specific embodiment of the present invention.

Fig. 2(A) to Fig. 2(C) are diagrams showing the results of analysis using the method according to the present invention and expert judgment.

Figure 3 is a statistical legend of the analysis results from 20 subjects.

Figure 6, Figure 8, and Figure 9 are the confusion matrices for comparing the results of the experiments according to the method of the present invention with the results of expert judgment.

Figure 7 shows the average consistency between the automatic judgment method and the manual interpretation for each sleep stage when multi-scale entropy analysis with different scales is used.

Figure 10 (A) and Figure 10 (B) are block diagrams of a sleep state judging device in accordance with an embodiment of the present invention.

Figure 11 shows the average consistency between the automatic judgment method and the manual interpretation when different autoregressive model series are used for each sleep stage.

Figures 12(A) to 12(C) and Figs. 13(A) to 13(C) are examples of analysis results obtained by experiments according to the method of the present invention and expertly judged.

S12~S16. . . Process step

Claims (20)

A sleep state judging method includes: (a) obtaining a physiological signal of a subject during a measurement period, wherein the physiological signal is a brain wave signal or a eye movement signal; (b) performing a physiological signal for the physiological signal Multi-scale entropy analysis to generate a plurality of entropy values; and (c) determining, according to the plurality of entropy values, a sleep state of the subject during the measurement period. The sleep state judging method according to claim 1, wherein the method further comprises: stepping down the sampling process and a band pass filtering program on the physiological signal, and the step of the step (b) The pass band is approximately 0.5 Hz to 30 Hz. The sleep state judging method according to claim 1, wherein the step (c) comprises determining the sleep state based on an average value of the plurality of entropy values. The sleep state judging method according to claim 1, wherein the plurality of entropy values respectively correspond to one of the scales used in the multi-scale entropy analysis, and the scale ranges from 1 to 13. The sleep state judging method according to claim 1, wherein the physiological signal is a C3-A2 brain wave signal. The sleep state judging method according to claim 1, wherein the method further comprises: (d) performing an autoregressive model operation procedure on the physiological signal to generate a plurality of steps; and (c) The autoregressive coefficients; wherein the step (c) comprises determining the sleep state according to the plurality of entropy values and the plurality of autoregressive coefficients. The sleep state judging method according to claim 6, wherein the autoregressive model operation program has eight levels and generates eight autoregressive coefficients. The sleep state judging method according to claim 6 , wherein the method further comprises: performing a band pass filtering process on the physiological signal between the step (a) and the step (d), and the band pass band is substantially 4 Hertz to 8 Hz; wherein step (d) performs the autoregressive model operation procedure for the filtered physiological signal. The sleep state judging method according to claim 6, wherein the step (c) comprises analyzing the isentropic value and the autoregressive coefficients by a linear discriminant analysis program to determine the sleep state. The sleep state judging method according to claim 1, wherein the method further comprises, after the step (c), selectively correcting the sleep state according to a before and after sleep state. A sleep state judging device includes: a collecting module, configured to obtain a physiological signal of a subject during a measurement period, wherein the physiological signal is a brain wave signal or an eye movement signal; an analysis module, And performing a multi-scale entropy analysis on the physiological signal to generate a plurality of entropy values; and a determining module, configured to determine, according to the plurality of entropy values, a sleep state of the subject during the measurement period. The sleep state judging device of claim 11, the device further comprising: a reduced sampling module for performing a reduced sampling procedure on the physiological signal; and a first filtering module for The physiological signal performs a band pass filtering process, and the band pass band is approximately 0.5 Hz to 30 Hz; wherein the analysis module performs the multi-scale entropy analysis on the physiological signal passing through the reduced sampling program and the filtering program. The sleep state judging device according to claim 11, wherein the judging module judges the sleep state based on an average value of the plurality of entropy values. The sleep state judging device according to claim 11, wherein the plurality of entropy values respectively correspond to one of the scales used in the multi-scale entropy analysis, and the scale ranges from 1 to 13. The sleep state judging device according to claim 11, wherein the physiological signal is a C3-A2 brain wave signal. The sleep state judging device of claim 11, the device further comprising: an autoregressive module for performing an autoregressive model operation procedure on the physiological signal to generate a plurality of autoregressive coefficients; wherein The determining module determines the sleep state based on the plurality of entropy values and the plurality of autoregressive coefficients. The sleep state judging device according to claim 16, wherein the autoregressive model operation program has eight levels and generates eight autoregressive coefficients. The sleep state judging device according to claim 16, wherein the device further comprises: a second filtering module, configured to perform a band pass filtering process on the physiological signal, and the band pass band is approximately 4 Hz to 8 Hertz; wherein the autoregressive module performs the autoregressive model operation procedure for the filtered physiological signal. The sleep state judging device according to claim 16, wherein the judging module analyzes the isentropic value and the autoregressive coefficients by a linear discriminant analysis program to determine the sleep state. The sleep state judging device according to claim 11, wherein the device further comprises: a correction module for selectively correcting the sleep state according to a before and after sleep state.