CN112426131A - Sleep stage automatic interpretation method based on step-by-step clustering model - Google Patents
Sleep stage automatic interpretation method based on step-by-step clustering model Download PDFInfo
- Publication number
- CN112426131A CN112426131A CN202011315916.8A CN202011315916A CN112426131A CN 112426131 A CN112426131 A CN 112426131A CN 202011315916 A CN202011315916 A CN 202011315916A CN 112426131 A CN112426131 A CN 112426131A
- Authority
- CN
- China
- Prior art keywords
- clustering
- sleep
- sample
- sleep stage
- distance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000008667 sleep stage Effects 0.000 title claims abstract description 92
- 238000000034 method Methods 0.000 title claims abstract description 67
- 230000007958 sleep Effects 0.000 claims abstract description 87
- 238000012937 correction Methods 0.000 claims abstract description 33
- 230000008569 process Effects 0.000 claims abstract description 32
- 238000005457 optimization Methods 0.000 claims abstract description 5
- 230000002618 waking effect Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 12
- 238000003064 k means clustering Methods 0.000 claims description 11
- 230000037053 non-rapid eye movement Effects 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 10
- 238000006243 chemical reaction Methods 0.000 abstract description 8
- 238000013461 design Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 14
- 230000007704 transition Effects 0.000 description 10
- 238000000819 phase cycle Methods 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 208000019116 sleep disease Diseases 0.000 description 7
- 208000020685 sleep-wake disease Diseases 0.000 description 6
- 238000012360 testing method Methods 0.000 description 5
- 206010062519 Poor quality sleep Diseases 0.000 description 4
- 230000004461 rapid eye movement Effects 0.000 description 4
- 210000004556 brain Anatomy 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000004622 sleep time Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 210000005252 bulbus oculi Anatomy 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003860 sleep quality Effects 0.000 description 1
- 230000037322 slow-wave sleep Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4812—Detecting sleep stages or cycles
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Fuzzy Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention relates to a sleep stage automatic interpretation method based on a step-by-step clustering model. The method designs and constructs a step-by-step clustering model by utilizing the association between different electroencephalogram signal characteristics and each sleep stage. Meanwhile, in the clustering process, optimization adjustment is carried out on key steps of the algorithm, an initial clustering center is selected by combining a method considering both density and distance, and a Gaussian kernel function is adopted to calculate a weight factor to adjust and update the clustering center. In addition, after clustering, a distance correction coefficient is designed to improve the rationality of a clustering result by combining with an actual conversion rule of a sleep state. The sleep stage automatic interpretation method based on the step-by-step clustering model can reflect the objective clustering effect of individual complex characteristic distribution in the sleep stage, also improves the classification performance of a clustering algorithm on the sleep stage problem, enables the classification performance to be closer to the experience mode of manual interpretation, and can provide an effective and feasible auxiliary interpretation tool for clinical application.
Description
Technical Field
The invention relates to the technical field of sleep evaluation, in particular to a sleep stage automatic interpretation method based on a step-by-step clustering model.
Background
The sleep process is a dynamic process consisting of several sleep stages representing different sleep states. In the sleep process, except for the waking period W, the sleep is divided into a rapid eye movement period REM and a non-rapid eye movement period according to the movement state of human eyeballs. In the non-rapid eye movement period, the sleep state is further divided into four sleep stages from shallow to deep, namely a shallow sleep stage S1, a shallow sleep stage S2, a deep sleep stage S3 and a deep sleep stage S4. Wherein, S3 and S4 are often combined into one slow wave sleep phase SS.
The interpretation of the sleep stages, namely analyzing the characteristics of bioelectricity signals such as electroencephalogram and the like synchronously recorded in the whole night sleep process, and corresponding to different sleep stages according to the change of the characteristics. The results of the sleep stage interpretation reflect the change of the sleep state in the whole night sleep process, and are important basis for clinically diagnosing and treating sleep related diseases. Conventionally, the evaluation of different sleep stages still depends on manual interpretation of expert doctors, professional manpower and a large amount of time are required for manual sleep stage division, interpretation results are subjective, and deviation can occur, so that subsequent sleep disease diagnosis and treatment are affected.
The automatic sleep staging method can reduce the manual burden of specialist doctors, can improve the interpretation efficiency of sleep staging, is an important auxiliary tool for evaluating sleep quality and diagnosing and treating sleep-related diseases, and has very important value for practical clinical application. However, since the actual sleep data is complicated and varied and there are large differences between individuals, these all bring difficulties to the research and application of the automatic interpretation method.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides an automatic sleep stage interpretation method based on a step-by-step clustering model, the step-by-step clustering model is designed and constructed aiming at the characteristics of sleep electroencephalograms and the association between the characteristics and different sleep stages, the clustering effect of individualized complex characteristics distributed in the sleep stages is objectively embodied, the classification performance of a clustering algorithm on the sleep stage discrimination problem is improved, the clustering algorithm is closer to the experience mode of manual interpretation, and an effective and feasible auxiliary interpretation tool can be provided for clinical application.
The purpose of the invention can be realized by the following technical scheme:
a sleep stage automatic interpretation method based on a step-by-step clustering model comprises the following steps:
step 1: acquiring electroencephalogram signals recorded in the sleep process of the whole night, and dividing the electroencephalogram signals into a plurality of data sections according to sampling frequency;
step 2: for each data segment, determining a plurality of frequency domain characteristics, including frequency domain characteristics of 1 high frequency band, frequency domain characteristics of 3 middle and low frequency bands and frequency domain characteristics of 2 middle and high frequency bands;
and step 3: extracting frequency domain characteristics of a high frequency band by taking all the electroencephalogram signals as input signals, constructing a characteristic sample A as a clustered input sample, and distinguishing data classification of an awake period W and a non-awake period by adopting an optimized and adjusted K-means clustering algorithm;
and 4, step 4: taking data in a non-waking period as an input signal, extracting frequency domain characteristics of a medium-low frequency band, constructing a characteristic sample B as a clustered input sample, and adopting an optimized and adjusted K-means clustering algorithm to distinguish SS (signal to noise ratio) in a deep sleep period, S2 in a shallow sleep period and other data classifications;
and 5: extracting frequency domain characteristics of medium and high frequency bands by taking other data as input signals, constructing a characteristic sample C as a clustering input sample, and distinguishing data classification of a waking period W, a shallow sleep period S1 and a non-rapid eye movement period REM by adopting an optimized and adjusted K mean value clustering algorithm;
step 6: and (4) performing distance correction processing on all clustering results obtained by a clustering algorithm, clustering the corrected data again, and outputting to obtain a sleep stage automatic interpretation result.
Further, the step 1 specifically includes: collecting the electroencephalogram signals recorded in the whole night sleep process, and dividing 3000 data points every 30 seconds into a plurality of data segments according to the sampling frequency.
Further, the step 2 comprises the following sub-steps:
step 201: extracting 6 frequency domain characteristics corresponding to different frequency bands from the electroencephalogram signals of every 30 seconds;
step 202: fourier transform is carried out on an original time sequence of the electroencephalogram signal, energy sum of 6 frequency bands is obtained through calculation, and then the percentage of the energy sum to total energy corresponding to a total frequency segment is obtained, and the energy sum serves as frequency domain characteristics and comprises frequency domain characteristics of 1 high frequency band, frequency domain characteristics of 3 middle and low frequency bands and frequency domain characteristics of 2 middle and high frequency bands.
Further, the different frequency bands in step 201 include frequency bands of 0 Hz-4 Hz, 4 Hz-8 Hz, 8 Hz-12 Hz, 12 Hz-15 Hz, 15 Hz-30 Hz and 30 Hz-49.5 Hz; the total frequency band in step 202 is 0.5Hz to 50 Hz.
Further, the optimization adjustment of the K-means clustering algorithm in the steps 3, 4 and 5 includes selection of an initial clustering center and an update rule of the clustering center.
Further, the selection process of the initial cluster center includes the following steps:
step 01: calculating to obtain a corresponding comprehensive index according to the local density and the relative distance of each sample point;
step 02: and arranging all the calculated comprehensive indexes in a descending order, and selecting sample points corresponding to the first K comprehensive index values as initial clustering centers according to the required clustering number K values, so as to finish the selection of the initial clustering centers.
Furthermore, the updating rule of the clustering center comprises a method for setting a weight to reduce the influence of outliers in each cluster, a Gaussian kernel function is selected to set the weight, the weight of a sample point farther away from the clustering center is smaller, the weight of a sample point closer to the clustering center is larger, and the clustering center is updated according to the weighted distance.
Further, the corresponding description formula of the comprehensive index in the step 01 is as follows:
in the formula, η is a comprehensive index, ρ is the local density of a sample point, and ξ is the relative distance of the sample point.
Further, the gaussian kernel function has a corresponding description formula as follows:
in the formula, xcThe center of the kernel function refers to the clustering center point of each cluster, and sigma is the width parameter of the kernel function and corresponds to the variance of the distance from the sample point to the clustering center.
Further, the process of performing distance correction processing on all the clustering results obtained by the clustering algorithm in the step 6 specifically includes: when the sleep stage to which the sample of one data segment i belongs is determined to be SiCorrecting the distance value corresponding to the sample of the next data segment i +1 according to the sleep stage S to which the sample of the data segment i belongsiCombining with each sleep stage, multiplying the distance correction coefficients corresponding to different combinations with the distance corresponding to the sample of the data segment i +1 to obtain a group of distance values corresponding to the sample of the corrected data segment i +1, and selecting the sleep stage class corresponding to the minimum value in the corrected distance as the final clustering result of the sample of the data segment i + 1;
the distance correction coefficient is as follows:
compared with the prior art, the invention has the following advantages:
(1) the invention relates to a sleep stage automatic interpretation method based on a step-by-step clustering model in sleep interpretation application, which aims at the correlation between sleep electroencephalogram characteristics and sleep stages, adopts a combination form of different characteristics, can flexibly construct characteristic samples, and realizes the automatic interpretation of the sleep stages step by combining with an optimized and adjusted K mean value clustering algorithm. The step-by-step clustering model has a better classification effect on solving the problem of sleep stages with complex shapes, and can improve the overall interpretation performance of the sleep stages.
(2) The invention provides a distance correction process after clustering, which is fused in a step-by-step clustering model, aims at the problem of unreasonable classification which may occur during sleep stage judgment, and adjusts the clustering result which is obtained by adopting a clustering algorithm and does not accord with the actual state conversion rule by using a correction coefficient according to the regularity of sleep state conversion all night. The distance correction processing can make the periodic change of the sleep stage more clear in the sleep process, and is helpful for professional doctors to further diagnose the sleep-related diseases.
(3) The invention relates to a sleep stage automatic interpretation method based on a step-by-step clustering model in sleep interpretation application, which mainly adopts a clustering mode to classify according to the actual distribution condition of characteristic samples. The model is simple and quick to realize, on one hand, prior knowledge and training data are not needed, and the process of model training is omitted; on the other hand, the problem of difference among different subjects does not need to be considered, and the clustering effect of individualized complex feature distribution in the sleep stage can be reflected. The model is suitable for processing complex and changeable sleep data in clinical practical application, and is an effective and feasible tool for auxiliary diagnosis and treatment.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention;
fig. 2 is a sleep time phase sequence diagram of a whole-night sleep process of a healthy adult under test in the embodiment of the present invention, in which fig. 2(a) is a schematic diagram of an artificial interpretation result of a sleep stage of a healthy adult under a test condition of the whole-night sleep process, and fig. 2(b) is a schematic diagram of an automatic interpretation result of a sleep stage of a healthy adult under a test condition of the whole-night sleep process based on a step-and-step clustering model;
fig. 3 is a sleep phase sequence chart of a whole-night sleep process of a certain sleep disorder patient in an embodiment of the present invention, where fig. 3(a) is a schematic diagram of an artificial interpretation result of a sleep stage of the certain sleep disorder patient under a test condition of the whole-night sleep process, and fig. 3(b) is a schematic diagram of an automatic interpretation result of the sleep stage based on a step-by-step clustering model under a test condition of the whole-night sleep process of the certain sleep disorder patient.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.
DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION
The invention provides a sleep stage automatic interpretation method based on a step-by-step clustering model, which comprises the following steps:
step 1: dividing the electroencephalogram signals recorded in the whole-night sleep process into data segments by taking 3000 data points every 30 seconds according to the sampling frequency;
step 2: for each data segment, 6 frequency domain characteristics are calculated, and the energy of different frequency segments respectively accounts for the percentage of the total frequency segment;
and step 3: extracting frequency domain characteristics of 1 high frequency band by taking all the brain electrical signals as input signals, and constructing a characteristic sample A as a clustered input sample;
and 4, step 4: distinguishing the waking period W and the non-waking period by adopting an optimized and adjusted K mean value clustering algorithm;
and 5: taking the non-awake period obtained after clustering in the step 4 as an input signal, extracting frequency domain characteristics of 3 middle and low frequency bands, and constructing a characteristic sample B as a clustered input sample;
step 6: distinguishing SS in a deep sleep period, S2 in a shallow sleep period and others by adopting an optimized and adjusted K-means clustering algorithm;
and 7: extracting frequency domain characteristics of 2 medium and high frequency bands by taking other samples obtained after clustering in the step 6 as input signals, and constructing a characteristic sample C as a clustered input sample;
and 8: distinguishing a waking period W, a shallow sleep period S1 and a non-rapid eye movement period REM by adopting an optimized and adjusted K-means clustering algorithm;
and step 9: distance correction processing is carried out on the clustering result, and unreasonable sleep state transition is reduced;
step 10: and re-clustering the corrected distances, and outputting to obtain a sleep stage automatic interpretation result.
The method designs and constructs a step-by-step clustering model by utilizing the association between different electroencephalogram signal characteristics and each sleep stage, simultaneously optimizes and adjusts the key steps of the algorithm in the clustering process, selects an initial clustering center by combining a density and distance method, and adjusts and updates the clustering center by adopting a Gaussian kernel function to calculate a weight factor. In addition, after clustering, a distance correction coefficient is designed to improve the rationality of a clustering result by combining with an actual conversion rule of a sleep state.
The sleep stage automatic interpretation method based on the step-by-step clustering model can reflect the objective clustering effect of individual complex characteristic distribution in the sleep stage, also improves the classification performance of a clustering algorithm on the sleep stage problem, enables the classification performance to be closer to the experience mode of manual interpretation, and can provide a feasible auxiliary interpretation tool for clinical application.
The specific technical details of the embodiment of the invention are as follows:
feature extraction
Under different sleep states, the energy of each frequency section of the electroencephalogram signal also presents different characteristics. The characteristic change of the brain electrical signals is the main basis for sleep stage interpretation. The invention extracts energy characteristics R of different frequency bands from the sleep electroencephalogram signalδ、Rθ、Rα、Rσ、RβAnd RγThe frequency bands respectively correspond to 0-4 Hz, 4-8 Hz, 8-12 Hz, 12-15 Hz, 15-30 Hz and 30-49.5 Hz. By adopting different combinations of the electroencephalogram characteristics, the whole-night sleep process is divided into a waking period W, a light sleep period S1, a light sleep period S2, a deep sleep period SS and a rapid eye movement period REM.
Step-by-step clustering model
Different sleep states have different characteristics, which are in turn characterized by parameters. Combining these parameters into a feature vector and clustering, which often reduces the overall effectiveness. In order to avoid the mutual influence of all parameters during classification, the invention relates to a step clustering model. FIG. 1 shows the general structure diagram of the step-by-step clustering model, the input of the model is sleep electroencephalogram data, and the output is the automatic interpretation result of sleep stages. The whole process comprises three step-by-step clustering modules and distance correction processing. For each step-by-step clustering module, different feature combinations can be flexibly applied, and a K-means clustering algorithm after optimization and adjustment is combined to distinguish sleep stages with strong correlation with the current feature combination, so that different sleep stages are distinguished step by step, and the mutual confusion phenomenon among the sleep stages is reduced. And adjusting the distance between the samples obtained after step-by-step clustering and the clustering center by adopting a distance correction coefficient to reduce unreasonable clustering results and improve the overall classification performance of the clustering algorithm in the sleep stage problem.
The clustering treatment is carried out in three steps as shown in the figure. The same point of each step of clustering is that the clustering method adopts an adjusted K mean value clustering algorithm; the difference is in the input samples and output classes for clustering.
Firstly, the high frequency component R in the EEG signal is correspondedγAs a characteristic, after the step 1 clustering is carried out on all sample data, distinguishing a part of waking periods W; secondly, the R of the medium and low frequency components is adjustedδ、RθAnd RσForming a characteristic vector as a sample characteristic, and carrying out the 2 nd step clustering on the rest sample data, wherein the shallow sleep period S2 and the deep sleep period SS are mainly distinguished; then, the R of the medium-high frequency component is addedαAnd RβThe combination of (3) is used as a sample characteristic, and then the 3 rd step clustering is carried out on the residual sample data to obtain a shallow sleep period S1, a rapid eye movement period REM and a waking period W.
In fig. 1, sleep stages S1, S2, SS, and REM in the sleep state are clustered at step 2 and step 3, respectively, considering that there are two different phenomena in the awake period WLike this, the awake period W was subjected to clustering twice in steps 1 and 3. In the step 1 of clustering, high-frequency components R are mainly combinedγTo distinguish the waking period W which does not contain alpha wave and has more active high-frequency components of the brain electrical signal. In light sleep stage S2, the intermediate frequency component RθIs more remarkable, and in the SS stage of deep sleep, the low-frequency component RδSignificantly, when the phase S2 enters the deep sleep SS phase, the θ wave gradually decreases and the δ wave dominates. Further, the intermediate frequency component RσCorresponding to the frequency components of 12-15 Hz, the frequency band is related to the frequency band of the characteristic waves such as the sleep spindle waves appearing in the S2 period. Therefore, in the step 2 clustering, the middle-low frequency component R is adoptedδ、RθAnd RσThe constituent feature vectors are used to determine S2 and SS. In the 3 rd step of clustering processing, the W, S1 EEG signals and the REM EEG signals have the characteristic of mixed frequency, the energy proportion is slightly different, and the medium-frequency component R is usedαTo obtain alpha wave-containing waking period W, and further combined with high frequency component RβTo distinguish W, S1 from REM.
Adjusted clustering algorithm
In a conventional K-means clustering algorithm, an initial cluster center is randomly selected from all samples, and a mean value is calculated for the clustered samples to update the cluster centers. On one hand, the randomly selected clustering center can make the algorithm unstable during iteration and reduce the algorithm classification effect; on the other hand, outliers inevitably exist due to differences of data samples, and positions of the outliers are often deviated from the current clustering center, so that the updated clustering center also generates a certain degree of deviation, and the clustering effect is influenced.
In the sleep stage automatic interpretation method based on the step-by-step clustering model, the optimization and adjustment of key steps in a K-means clustering algorithm comprise the selection of an initial clustering center and the updating of the clustering center, so that the randomness problem of the selection of the initial clustering center is solved, and the classification performance of the clustering algorithm is improved.
The selection of the initial clustering center mainly focuses on two measurement modes of distance or density. Calculating the local density as the minimum intra-class distance and the relative distance as the maximum inter-class distance, combining the two measurement indexes, and calculating the parameters:
in the formula, η is a comprehensive index, ρ is the local density of a sample point, and ξ is the relative distance of the sample point.
And (3) representing the comprehensive index of the density and the distance by eta, and serving as a basis for selecting an initial clustering center. Thus, a corresponding comprehensive index η (i) is calculated from the local density ρ (i) and the relative distance ξ (i) of each sample point. And then, arranging the eta according to a descending order from big to small, and selecting the sample points with the larger eta values as initial clustering centers according to the required clustering quantity K values, thereby completing the selection of the initial clustering centers.
In order to reduce the offset influence of the outlier generated in the updating of the cluster center, a method for setting a weight value is adopted to reduce the influence of the outlier in each cluster. A Gaussian kernel function is selected to set the weight, and the expression of the Gaussian kernel function is shown as the following formula:
in the formula, xcThe center of the kernel function refers to the clustering center point of each cluster, and sigma is the width parameter of the kernel function and corresponds to the variance of the distance from the sample point to the clustering center.
For each sample point, a weight factor ω ═ K (| | x-x) corresponding to the sample point is calculated according to the formulac| |). According to the characteristics of the gaussian kernel function, if a sample point is farther from the clustering center, the value of ω is smaller, otherwise, the value of ω is larger, and when x is xcWhen the sample point is the cluster center point, the value of the weighting factor ω is 1. Firstly, solving a weight factor omega for all samples in a cluster; then, normalization is performed, i.e. all sample weights in the cluster are summed and then each sample weight is divided byThe weight sum is used for obtaining omega ', the normalized sample weight sum is 1, and omega' is used as a weight factor corresponding to each sample; and finally, updating the clustering center by using the value of x omega' as the updated sample point. If the clustering center is changed, the basic steps of the clustering algorithm are continuously executed in an iterative mode until the clustering center is not changed any more, clustering is completed, and a clustering result is output.
Distance correction
In the whole sleep process, each sleep stage lasts for a period of time, and then the transition is made to another sleep stage. The transitions between sleep stages follow the natural laws of the sleep cycle, that is, each sleep stage transitions to a sleep stage in the state adjacent to it, such as from the light sleep stage S2 to the deep sleep stage SS; whereas transitions between non-adjacent sleep sessions, such as a transition from the awake period W directly to the deep sleep period SS, are generally not, or only with minimal probability, occurring during overnight sleep.
In the sleep stage automatic interpretation method based on the step-by-step clustering model, distance correction processing is designed for improving the rationality of the step-by-step clustering model according to the actual rules of different sleep stages in a sleep cycle. As shown in table 1, the distance correction coefficient from the awake period W to the non-adjacent deep sleep period SS and rapid eye movement period REM is 1, indicating that there is a very low possibility of direct transition from the awake period W to SS or REM. Correspondingly, the distance correction coefficient is set to 1, which means that the condition is not corrected, and the originally calculated distance value is reserved. The distance correction coefficients from the awake period W to the adjacent shallow sleep periods S1 and S2 are 0.84 and 0.98, respectively, indicating that there is a certain probability of switching from the awake period W to S1 or S2, and the probability of W to S1 is greater than the probability of W to S2. Thus, the distance correction coefficients in table 1 also correspond to the transition probabilities between the respective sleep stages, and the distance correction processing is performed for both cases, and the distance correction coefficient corresponding to the combination of W to S1 is smaller than the distance correction coefficient corresponding to the combination of W to S2. In table 1, the distance correction coefficient indicating that the awake period continues is relatively small, 0.18, indicating that the probability of the awake period continuing is relatively high.
TABLE 1 distance correction factor
When the sleep stage combination to which the samples of the front and rear data sections belong has higher conversion probability, correcting the clustering result, and multiplying the original distance by a smaller distance correction coefficient; on the contrary, the distance is multiplied by a larger distance correction coefficient; if the conversion probability is very small, no correction is made. After the distance correction processing step is added, under the condition that the sleep stage corresponding to the sample of the current data segment is determined, the sample of the next data segment is judged in the sleep stage with higher conversion probability as far as possible. When the sleep stages are in the continuous stage, the distance correction coefficient corresponding to the same sleep stage is far smaller than that of other sleep stages, and the classification accuracy of the continuous stage is improved by using the distance correction coefficient; when the sleep stages are in the transition stage, the transition stage of the adjacent sleep stages is corrected and identified to a certain extent through the distance coefficient.
Examples of the applications
Selecting data from different Sleep electroencephalogram databases, wherein one data is a Sleep-EDF public data set which is Sleep electroencephalogram data of healthy adults; the other is actual clinical data, which is sleep electroencephalogram data of patients suffering from sleep disorder and receiving continuous positive airway pressure treatment. The acquisition of the sleep electroencephalogram data obtains the informed consent of the subject.
The sleep stage automatic interpretation method based on the step-by-step clustering model is applied to sleep electroencephalogram data of healthy adults and sleep disorder patients. Fig. 2 shows a diagram of sleep phase sequences for a healthy adult tested for the whole night sleep. In fig. 2, 2(a) is a sleep phase sequence diagram of the result of the sleep stage manual interpretation, and 2(b) is a sleep phase sequence diagram obtained based on the step-by-step clustering model. The abscissa in the figure represents the duration of the sleep process in hours (h) and the ordinate corresponds to different sleep stages, W, S1, S2, SS and REM from top to bottom. On the whole, the result obtained by the sleep stage automatic interpretation method based on the step-by-step clustering model, namely the sleep state in the sleep time phase sequence diagram is periodically changed, the whole change trend is more consistent with the manual interpretation result, and the change condition of the sleep state can be observed more clearly. Similarly, fig. 3 shows a sleep phase sequence diagram of the whole night sleep process of a sleep disorder patient, in fig. 3, 3(a) is a sleep phase sequence diagram of the sleep stage manual interpretation result, and 3(b) is a sleep phase sequence diagram obtained based on the step-by-step clustering model. By comparison with the results of manual interpretation, a similar classification effect as in fig. 2 can be observed. The step-by-step clustering model not only can enable the change of the sleep state to be more clear, but also corrects the classification result which does not accord with the actual sleep state conversion rule in the automatic interpretation process to a certain extent, and obtains better effects on reducing the confusion phenomenon of the sleep cycle and improving the classification effect of the sleep stages on the whole.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A sleep stage automatic interpretation method based on a step-by-step clustering model is characterized by comprising the following steps:
step 1: acquiring electroencephalogram signals recorded in the sleep process of the whole night, and dividing the electroencephalogram signals into a plurality of data sections according to sampling frequency;
step 2: for each data segment, determining a plurality of frequency domain characteristics, including frequency domain characteristics of 1 high frequency band, frequency domain characteristics of 3 middle and low frequency bands and frequency domain characteristics of 2 middle and high frequency bands;
and step 3: extracting frequency domain characteristics of a high frequency band by taking all the electroencephalogram signals as input signals, constructing a characteristic sample A as a clustered input sample, and distinguishing data classification of an awake period W and a non-awake period by adopting an optimized and adjusted K-means clustering algorithm;
and 4, step 4: taking data in a non-waking period as an input signal, extracting frequency domain characteristics of a medium-low frequency band, constructing a characteristic sample B as a clustered input sample, and adopting an optimized and adjusted K-means clustering algorithm to distinguish SS (signal to noise ratio) in a deep sleep period, S2 in a shallow sleep period and other data classifications;
and 5: extracting frequency domain characteristics of medium and high frequency bands by taking other data as input signals, constructing a characteristic sample C as a clustering input sample, and distinguishing data classification of a waking period W, a shallow sleep period S1 and a non-rapid eye movement period REM by adopting an optimized and adjusted K mean value clustering algorithm;
step 6: and (4) performing distance correction processing on all clustering results obtained by a clustering algorithm, clustering the corrected data again, and outputting to obtain a sleep stage automatic interpretation result.
2. The sleep staging automatic interpretation method based on the step-by-step clustering model as claimed in claim 1, wherein the step 1 specifically comprises: collecting the electroencephalogram signals recorded in the whole night sleep process, and dividing 3000 data points every 30 seconds into a plurality of data segments according to the sampling frequency.
3. The method for automatically judging sleep stages based on the step-by-step clustering model as claimed in claim 2, wherein the step 2 comprises the following sub-steps:
step 201: extracting 6 frequency domain characteristics corresponding to different frequency bands from the electroencephalogram signals of every 30 seconds;
step 202: fourier transform is carried out on an original time sequence of the electroencephalogram signal, energy sum of 6 frequency bands is obtained through calculation, and then the percentage of the energy sum to total energy corresponding to a total frequency segment is obtained, and the energy sum serves as frequency domain characteristics and comprises frequency domain characteristics of 1 high frequency band, frequency domain characteristics of 3 middle and low frequency bands and frequency domain characteristics of 2 middle and high frequency bands.
4. The method according to claim 3, wherein the different frequency bands in step 201 include frequency bands of 0 Hz-4 Hz, 4 Hz-8 Hz, 8 Hz-12 Hz, 12 Hz-15 Hz, 15 Hz-30 Hz, and 30 Hz-49.5 Hz; the total frequency band in step 202 is 0.5Hz to 50 Hz.
5. The method for automatically judging sleep stages based on the step-by-step clustering model as claimed in claim 1, wherein the optimization adjustment of the K-means clustering algorithm in the steps 3, 4 and 5 comprises selection of initial clustering centers and update rules of the clustering centers.
6. The method for automatically judging sleep stages based on the step-by-step clustering model as claimed in claim 5, wherein the selection process of the initial clustering center comprises the following steps:
step 01: calculating to obtain a corresponding comprehensive index according to the local density and the relative distance of each sample point;
step 02: and arranging all the calculated comprehensive indexes in a descending order, and selecting sample points corresponding to the first K comprehensive index values as initial clustering centers according to the required clustering number K values, so as to finish the selection of the initial clustering centers.
7. The method as claimed in claim 5, wherein the updating rule of the clustering center includes using a weight setting method to reduce the influence of the outliers in each cluster, using a Gaussian kernel function to set the weight, wherein the weight of the sample points farther away from the clustering center is smaller, and the weight of the sample points closer to the clustering center is larger, and updating the clustering center with the weighted distance.
8. The method for automatically judging sleep stages based on the step-by-step clustering model as claimed in claim 6, wherein the synthetic index in the step 01 corresponds to a description formula:
in the formula, η is a comprehensive index, ρ is the local density of a sample point, and ξ is the relative distance of the sample point.
9. The method according to claim 7, wherein the Gaussian kernel function has a corresponding description formula:
in the formula, xcThe center of the kernel function refers to the clustering center point of each cluster, and sigma is the width parameter of the kernel function and corresponds to the variance of the distance from the sample point to the clustering center.
10. The sleep stage automatic interpretation method based on the step-by-step clustering model as claimed in claim 1, wherein the process of performing distance correction processing on all clustering results obtained by the clustering algorithm in the step 6 specifically comprises: when the sleep stage to which the sample of one data segment i belongs is determined to be SiCorrecting the distance value corresponding to the sample of the next data segment i +1 according to the sleep stage S to which the sample of the data segment i belongsiCombining with each sleep stage, multiplying the distance correction coefficients corresponding to different combinations with the distance corresponding to the sample of the data segment i +1 to obtain a group of distance values corresponding to the sample of the corrected data segment i +1, and selecting the sleep stage class corresponding to the minimum value in the corrected distance as the final clustering result of the sample of the data segment i + 1;
the distance correction coefficient is as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011315916.8A CN112426131A (en) | 2020-11-22 | 2020-11-22 | Sleep stage automatic interpretation method based on step-by-step clustering model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011315916.8A CN112426131A (en) | 2020-11-22 | 2020-11-22 | Sleep stage automatic interpretation method based on step-by-step clustering model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112426131A true CN112426131A (en) | 2021-03-02 |
Family
ID=74693452
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011315916.8A Pending CN112426131A (en) | 2020-11-22 | 2020-11-22 | Sleep stage automatic interpretation method based on step-by-step clustering model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112426131A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113080864A (en) * | 2021-04-07 | 2021-07-09 | 电子科技大学 | Common sleep disease detection method through automatic sleep staging results |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011083393A (en) * | 2009-10-14 | 2011-04-28 | Osaka Bioscience Institute | Apparatus and method for automatically identifying sleep stage, and computer program for the same |
| CN106175697A (en) * | 2016-09-18 | 2016-12-07 | 广州视源电子科技股份有限公司 | Sleep state detection method and device |
| CN110141226A (en) * | 2019-05-29 | 2019-08-20 | 清华大学深圳研究生院 | Sleep mode automatically method, apparatus, computer equipment and computer storage medium by stages |
| CN110728231A (en) * | 2019-10-10 | 2020-01-24 | 华东理工大学 | Sleep staging method based on improved particle swarm algorithm and twin support vector machine |
| CN110742585A (en) * | 2019-10-10 | 2020-02-04 | 北京邮电大学 | Sleep staging method based on BCG (BCG-broadcast) signals |
| CN110897639A (en) * | 2020-01-02 | 2020-03-24 | 清华大学深圳国际研究生院 | Electroencephalogram sleep staging method based on deep convolutional neural network |
| CN111067503A (en) * | 2019-12-31 | 2020-04-28 | 深圳安视睿信息技术股份有限公司 | Sleep staging method based on heart rate variability |
| US20200181707A1 (en) * | 2012-09-13 | 2020-06-11 | Yrt Limited | Method and software to determine probability of sleep/wake states and quality of sleep and wakefulness from an electroencephalogram |
-
2020
- 2020-11-22 CN CN202011315916.8A patent/CN112426131A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011083393A (en) * | 2009-10-14 | 2011-04-28 | Osaka Bioscience Institute | Apparatus and method for automatically identifying sleep stage, and computer program for the same |
| US20200181707A1 (en) * | 2012-09-13 | 2020-06-11 | Yrt Limited | Method and software to determine probability of sleep/wake states and quality of sleep and wakefulness from an electroencephalogram |
| CN106175697A (en) * | 2016-09-18 | 2016-12-07 | 广州视源电子科技股份有限公司 | Sleep state detection method and device |
| CN110141226A (en) * | 2019-05-29 | 2019-08-20 | 清华大学深圳研究生院 | Sleep mode automatically method, apparatus, computer equipment and computer storage medium by stages |
| CN110728231A (en) * | 2019-10-10 | 2020-01-24 | 华东理工大学 | Sleep staging method based on improved particle swarm algorithm and twin support vector machine |
| CN110742585A (en) * | 2019-10-10 | 2020-02-04 | 北京邮电大学 | Sleep staging method based on BCG (BCG-broadcast) signals |
| CN111067503A (en) * | 2019-12-31 | 2020-04-28 | 深圳安视睿信息技术股份有限公司 | Sleep staging method based on heart rate variability |
| CN110897639A (en) * | 2020-01-02 | 2020-03-24 | 清华大学深圳国际研究生院 | Electroencephalogram sleep staging method based on deep convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| 于莹 等: "基于改进K均值聚类及其距离修正的睡眠分期方法", 计算机应用, vol. 40, no. 1, pages 269 - 273 * |
| 肖姝源 等: "基于改进的Ｋ均值聚类算法的睡眠自动分期研究", 生物医学工程学杂志, vol. 33, no. 5, pages 847 - 854 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113080864A (en) * | 2021-04-07 | 2021-07-09 | 电子科技大学 | Common sleep disease detection method through automatic sleep staging results |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Korkalainen et al. | Accurate deep learning-based sleep staging in a clinical population with suspected obstructive sleep apnea | |
| CN107495962B (en) | Sleep automatic staging method for single-lead electroencephalogram | |
| CN111493822B (en) | Sleep electroencephalogram based rapid eye movement period sleep behavior disorder classification method | |
| CN110897639A (en) | Electroencephalogram sleep staging method based on deep convolutional neural network | |
| CN105451801B (en) | The sleep method and system for the method for sound induction | |
| CN112656427A (en) | Electroencephalogram emotion recognition method based on dimension model | |
| CN113729707A (en) | FECNN-LSTM-based emotion recognition method based on multi-mode fusion of eye movement and PPG | |
| KR20120012364A (en) | Bio-signals based automatic sleep-stage identification system | |
| CN112088408A (en) | Method for sleep stage detection, computing device and wearable device | |
| CN110432870A (en) | A kind of sleep signal based on 1D CNN-LSTM method by stages automatically | |
| JP2002534144A (en) | Method and apparatus for performing a visual field test and computer program for processing the result | |
| CN107874750A (en) | Pulse frequency variability and the psychological pressure monitoring method and device of sleep quality fusion | |
| WO2022134242A1 (en) | Method for establishing automatic sleep staging and application thereof | |
| CN112263253B (en) | Depression recognition system, medium and equipment based on deep learning and electrocardiosignal | |
| CN117609813A (en) | Intelligent management method for intensive patient monitoring data | |
| CN115530847A (en) | Electroencephalogram signal automatic sleep staging method based on multi-scale attention | |
| CN112932501A (en) | Method for automatically identifying insomnia based on one-dimensional convolutional neural network | |
| CN114569096B (en) | Non-contact continuous blood pressure measurement method and system based on video stream | |
| CN115500843A (en) | Sleep stage staging method based on zero sample learning and contrast learning | |
| CN112426131A (en) | Sleep stage automatic interpretation method based on step-by-step clustering model | |
| CN112353401B (en) | Staged regulation and control method based on physiological state evaluation | |
| CN113729648B (en) | Wearable pulse-taking bracelet system based on multiple pulse sensors | |
| CN114129147A (en) | System and method for predicting effects of Parkinson patients after DBS operation based on brain function network | |
| CN115399735A (en) | Multi-head attention mechanism sleep staging method based on time-frequency double-current enhancement | |
| Lv et al. | A multilevel temporal context network for sleep stage classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |