CN111743513B - S1 and REM sleep state detection method and system based on mixed channel EEG signal - Google Patents

S1 and REM sleep state detection method and system based on mixed channel EEG signal Download PDF

Info

Publication number
CN111743513B
CN111743513B CN202010606885.5A CN202010606885A CN111743513B CN 111743513 B CN111743513 B CN 111743513B CN 202010606885 A CN202010606885 A CN 202010606885A CN 111743513 B CN111743513 B CN 111743513B
Authority
CN
China
Prior art keywords
rem
sleep
data
state
states
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.)
Active
Application number
CN202010606885.5A
Other languages
Chinese (zh)
Other versions
CN111743513A (en
Inventor
袁志勇
安攀峰
杜博
赵俭辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN202010606885.5A priority Critical patent/CN111743513B/en
Publication of CN111743513A publication Critical patent/CN111743513A/en
Application granted granted Critical
Publication of CN111743513B publication Critical patent/CN111743513B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • A61B5/4809—Sleep detection, i.e. determining whether a subject is asleep or not
    • 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
    • A61B5/4815—Sleep quality
    • 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
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device

Abstract

The invention discloses a method and a system for detecting S1 and REM sleep states based on mixed channel EEG signals, which comprises the steps of firstly, acquiring EEG signal data of Fpz-Cz and Pz-Oz channels; then taking the S1 and REM sleep states as a whole state, and detecting the whole state by using Fpz-Cz single-channel signal data; and finally, carrying out two classifications on the detected S1 and REM overall states by using signal data of two channels of Fpz-Cz and Pz-Oz so as to respectively detect two different sleep states of S1 and REM. According to the method, the similarity and difference characteristics of the S1 and the REM sleep state are combined, the multi-model fusion classification detection method is established by utilizing different channel signal data and extracting different classification characteristics, and the detection precision of the S1 and the REM sleep state can be effectively improved.

Description

S1 and REM sleep state detection method and system based on mixed channel EEG signal
Technical Field
The invention belongs to the technical field of sleep state detection, relates to a sleep state detection method and system based on an EEG signal, and particularly relates to a sleep state detection method and system based on an S1 and REM of a mixed channel EEG signal.
Background
Sleep is the natural state of brain activity, and different sleep states reflect the physiological, cognitive, health, etc. characteristics of a person. The various sleep states comprise six stages of Wake, S1, S2, S3, S4 and REM, and the S1 and the S2 are light sleep states, wherein the S1 state is directly related to the cognitive behaviors of people and can be used for detecting fatigue and other applications, such as driver fatigue driving detection; s3 and S4 are deep sleep states, reflecting sleep quality changes; REM is a special sleep state that can be used to diagnose related sleep disorders by detecting the REM sleep state. Because sleep directly affects the mood, cognition, memory, etc. of humans, EEG-based sleep staging research has gained increasing attention [1-4 ]. Related research uses intelligent algorithms to classify different sleep states, including methods such as deep neural networks, support vector machines, random forests, etc., and how to select effective signal channels and classification features is also the key to improve sleep stages [5-7 ].
Although many current sleep staging methods have achieved good detection effects, due to the similarity of different sleep state signals and the imbalance of data samples, there are great differences in the detection performance of various sleep states, such as the detection rate of Wake state is more than 95%, and the detection rates of S1 and REM sleep state are relatively low, especially the detection rate of S1, which is generally 30% -50% [8-10 ]. By analyzing the changing characteristics of the sleep EEG signal, S1 and REM are two similar sleep states and have the characteristic of few data samples, which makes it difficult to perform effective detection. S1 is the initial state of sleep stage, the brain begins to sleep, this state directly affects the cognitive behavior of people, and REM is a special sleep state very related to various mental diseases, so how to increase the detection rate of S1 and REM in the sleep stage is very important for practical application.
Document [11] combines S1 and REM into one state according to the change characteristics of the sleep EEG signal, and performs classification detection of 5 types of sleep states, but although the overall sleep staging accuracy is improved, detection of S1 and REM sleep states is lacking; document [12] performs sleep staging studies using a cascaded LSTM approach, the first LSTM network classification including four sleep stages of S1 and REM global states, and the second network performing a two-class detection of S1 and REM sleep states. Document [13] extracts classification features from various physiological signals such as EEG, EOG, EMG, and the like in a polysomnogram, and detects the REM sleep state using a random forest algorithm.
Because the S1 and the REM are two sleep states with important application and research values, and the current sleep staging method lacks effective detection of the S1 and REM sleep states, it is necessary to select a high-efficiency signal classification characteristic in combination with the change characteristics of the EEG signal, and meanwhile, the detection model is to avoid the problem of inconsistent detection performance of various sleep states in the sleep staging method, and overcome the influence of data sample imbalance and the like.
Reference to the literature
[1]H.Ghimatgar,K.Kazemi,M.S.Helfroush,and A.Aarabi,“An automatic single-channel EEG-based sleep stage scoring method based on hidden markov model,”Journal of neuroscience methods,p.108320,2019.
[2]P.Ghasemzadeh,H.Kalbkhani,S.Sartipi,and M.G.Shayesteh,“Classification of sleep stages based on LSTAR model,”Applied Soft Computing,vol.75,pp.523–536,2019.
[3]F.Dehnavi,S.Moghimi,S.SadrabadiHaghighi,M.Safaie,and M.Ghorbani,“Opposite effect of motivated forgetting on sleep spindles during stage 2and slow wave sleep,”Sleep,2019.
[4]M.Sokolovsky,F.Guerrero,S.Paisarnsrisomsuk,C.Ruiz,and S.A.Alvarez,“Deep learning for automated feature discovery and classification of sleep stages,”IEEE/ACM transactions on computational biology and bioinformatics,2019.
[5]X.Li,L.Cui,S.Tao,J.Chen,X.Zhang,and G.-Q.Zhang,“Hyclasss:A hybrid classifier for automatic sleep stage scoring,”IEEE journal of biomedical and health informatics,vol.22,no.2,pp.375–385,2017.
[6]D.Y.Kang,P.N.DeYoung,A.Malhotra,R.L.Owens,and T.P.Coleman,“A state space and density estimation framework for sleep staging in obstructive sleep apnea,”IEEE Transactions on Biomedical Engineering,vol.65,no.6,pp.1201–1212,2017.
[7]A.Supratak,H.Dong,C.Wu,and Y.Guo,“Deepsleepnet:A model for automatic sleep stage scoring based on raw single-channel EEG,”IEEE Transactions on Neural Systems and Rehabilitation Engineering,vol.25,no.11,pp.1998–2008,2017.
[8]D.Jiang,Y.-n.Lu,M.Yu,and W.Yuanyuan,“Robust sleep stage classification with single-channel EEG signals using multimodal decomposition and hmm-based refinement,”Expert Systems with Applications,vol.121,pp.188–203,2019.
[9]S.Seifpour,H.Niknazar,M.Mikaeili,and A.M.Nasrabadi,“A new automatic sleep staging system based on statistical behavior of local extrema using single channel EEG signal,”Expert Systems with Applications,vol.104,pp.277–293,2018.
[10]A.R.Hassan and M.I.H.Bhuiyan,“Computer-aided sleep staging using complete ensemble empirical mode decomposition with adaptive noise and bootstrap aggregating,”Biomedical Signal Processing and Control,vol.24,pp.1–10,2016.
[11]K.A.Aboalayon,W.S.Almuhammadi,and M.Faezipour,“A comparison of different machine learning algorithms using single channel EEG signal for classifying human sleep stages,”in 2015Long Island Systems,Applications and Technology.IEEE,2015,pp.1–6.
[12]N.Michielli,U.R.Acharya,and F.Molinari,“Cascaded LSTM recurrent neural network for automated sleep stage classification using single-channel EEG signals,”Computers in biology and medicine,vol.106,pp.71–81,2019.
[13]N.Cooray,F.Andreotti,C.Lo,M.Symmonds,M.T.Hu,and M.De Vos,“Detection of REM sleep behaviour disorder by automated polysomnography analysis,”Clinical Neurophysiology,vol.130,no.4,pp.505–514,2019.
Disclosure of Invention
In order to solve the above technical problem, the present invention provides a method for detecting S1 and REM sleep states based on a mixed channel EEG signal.
The method adopts the technical scheme that: a method for detecting S1 and REM sleep states based on mixed channel EEG signals, comprising the steps of:
step 1: acquiring EEG signal data of two channels of Fpz-Cz and Pz-Oz;
step 2: taking the S1 sleep state and the REM sleep state as a whole state, and detecting the whole state by using Fpz-Cz single-channel signal data;
and step 3: the detected overall states of S1 and REM are classified by using signal data of two channels of Fpz-Cz and Pz-Oz to detect two different sleep states of S1 and REM, respectively.
Preferably, the specific implementation of step 2 comprises the following sub-steps:
step 2.1: processing Fpz-Cz single-channel EEG signal data, and taking a continuous N-second time length signal as a data sample, wherein N is a preset value and is generally 30;
step 2.2: extracting EEG signal classification characteristics, specifically including signal sample range, mean value and variance; number of peak points, mean, variance, average interval; the number of mode points, average interval, variance and other classification characteristics reflecting the EEG amplitude and frequency change characteristics;
step 2.3: establishing a C-SVM two-classification model of the S1 and REM overall state and other sleep states (including Wake, S2, S3 and S4), marking the S1 and REM two-class data samples as a +1 class, marking Wake, S2, S3 and S4 as a "-1 class, and establishing a detection model through training the data samples;
step 2.4: establishing an OC-SVM single-class detection model for S1 and REM integral detection, wherein the modeling process is to only use two types of data samples, namely S1 and REM, and mark the two types of data samples as a +1 type, so as to establish a detection model for detecting S1 and REM integral sleep state;
step 2.5: judging whether an N-second data sample belongs to S1 or REM two sleep states by using a C-SVM two-classification model; if yes, executing step 3; if not, executing the step 2.6;
step 2.6: and (3) judging the data sample by using the established OC-SVM detection model, if the data sample belongs to the integral state of S1 and REM, executing the step 3, and otherwise, judging the data sample to be in other sleep states.
Preferably, in step 2.3, based on the similarity of the sleep states of S1 and REM, S1 and REM are taken as one class, and Wake, S2, S3 and S4 are taken as one class; establishing a C-SVM two-classification model of S1 and REM overall state and other sleep states, wherein the solved C-SVM classification model is as follows:
Figure GDA0002986928240000041
wherein α ═ (α)1,α2,...,αN) For Lagrangian, K (x)i,xj) Is a Gaussian kernel function, xiAnd xjRespectively an ith sleep state data sample and a jth sleep state data sample in the training data; thus, a decision function is constructed:
Figure GDA0002986928240000042
where sign is a sign function, b is a separate hyperplane offset term, αi *The ith Lagrangian, K (x, x), learned for model trainingi) Is a Gaussian kernel function, x is a new test data sample, yiThe label category of the ith training sample; if f (x) is 1, the sample x is determined to be in S1 or REM sleep state, and if f (x) is-1, the sample x is determined to be in other sleep states.
Preferably, the OC-SVM model in step 2.4 is solved by:
Figure GDA0002986928240000051
wherein α ═ (α)1,α2,…,αN) For Lagrangian, K (x)i,xj) Is a Gaussian kernel function, xiAnd xjRespectively an ith sleep state data sample and a jth sleep state data sample in the training data; thus, a decision function is constructed:
Figure GDA0002986928240000052
wherein sign is a sign function, rho is a compensation term of the decision function of the OC-SVM, and alphai *The ith Lagrangian, K (x, x), learned for model trainingi) Is a Gaussian kernel function, x is a new test data sample, yiThe label category of the ith sample; if f (x) is 1, the sample x is determined to be S1 orREM sleep state, otherwise, determining to be other class of sleep state.
Preferably, the specific implementation of step 3 comprises the following sub-steps:
step 3.1: processing EEG signal data of two channels of Fpz-Cz and Pz-Oz, and taking a continuous N-second time length signal as a data sample, wherein N is a preset value and is generally 30;
step 3.2: decomposing the two channel signal data of each data sample into signal sub-bands with different frequencies;
step 3.3: extracting classification features of each decomposed frequency signal sub-band, wherein the classification features specifically comprise a signal sample range, a mean value and a variance; number of peak points, mean, variance, average interval; the number of mode points, average interval, variance and other classification characteristics reflecting the EEG amplitude and frequency change characteristics;
step 3.4: establishing a C-SVM two-classification model of S1 and REM sleep states by using extracted sub-band characteristics of different frequency signals, marking an S1 data sample as a +1 class and an REM data sample as a "-1 class in the modeling process, and establishing a detection model by training the data sample;
step 3.5: and judging the detected data samples of the S1 and REM overall state by using the C-SVM classification model and the characteristic data of the sub-bands of the two channels of the multi-frequency signals, and outputting the data samples into specific S1 and REM sleep states.
Preferably, in step 3.2, the EEG signals of two channels are decomposed by using FFT algorithm, signal sub-bands of four frequencies of 1-8Hz, 1-9Hz, 9-50Hz, and 30-50Hz are obtained, feature extraction is performed on each corresponding 8 decomposed signal sub-bands, and in step 3.4, C-SVM binary classification models of two sleep states of S1 and REM are trained, and the algorithm model is as in step 2.3.
The technical scheme adopted by the system of the invention is as follows: a mixed channel EEG signal based S1 and REM sleep state detection system, comprising: the system comprises an EEG signal data acquisition module, an S1 and REM integral state detection module, and an S1 and REM sleep state classification detection module;
the EEG signal data acquisition module is used for acquiring EEG signal data of two channels of Fpz-Cz and Pz-Oz;
the S1 and REM overall state detection module is used for taking the S1 and REM two sleep states as an overall state and detecting the overall state by utilizing Fpz-Cz single-channel signal data;
the S1 and REM sleep state classification detection module is used for carrying out two classifications on the detected S1 and REM overall state by utilizing signal data of two channels of Fpz-Cz and Pz-Oz so as to respectively detect two different sleep states of S1 and REM.
Compared with the prior art, the invention has the following innovation and advantages:
1. the invention researches a sleep state detection method of S1 and REM based on mixed channel EEG signals, selects different signal data with high efficiency for different classification tasks in the detection process, such as classifying the integral state of S1 and REM and other sleep states only by using single channel EEG signals, classifying the two sleep states of S1 and REM with strong similarity by using double channel EEG signals, and on the basis of ensuring the detection performance, uses less EEG signal data as much as possible.
2. The invention adopts a multi-model fusion method to detect the overall state of S1 and REM, wherein the detection rate of the overall sleep state of S1 and REM can be improved and the influence caused by data imbalance can be overcome by establishing an OC-SVM single-class detection model of the overall state and detecting the negative-class data sample detected by the C-SVM again.
3. In order to effectively classify the two sleep states of S1 and REM, the EEG signals of the two channels are decomposed to respectively obtain a plurality of sub-band signal data with different frequencies, and the corresponding frequency sub-bands respectively reflect the difference of the two sleep states in terms of amplitude and frequency characteristics, so that the classification accuracy of S1 and REM can be improved.
Drawings
Fig. 1 is a schematic diagram of the application of S1 and REM sleep state detection in sleep quality analysis, drowsiness detection and sleep disease diagnosis according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of detecting S1 and REM sleep states by the multi-model fusion method according to the embodiment of the present invention
Fig. 3 is a schematic sub-band diagram of EEG signal decomposition for S1 and REM detection according to an embodiment of the present invention.
Fig. 4 is a flowchart illustrating a method for detecting S1 and REM sleep states based on a mixed channel EEG signal according to an embodiment of the present invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail with reference to the accompanying drawings and examples, it is to be understood that the embodiments described herein are merely illustrative and explanatory of the present invention and are not restrictive thereof.
The S1 and REM sleep state detection in this embodiment are performed by performing various classification detection tasks for different sleep states to effectively detect two sleep states, S1 and REM. Since the sleep states include six different states, namely Wake, S1, S2, S3, S4 and REM, the method provided by the present invention can be adjusted by combining specific target tasks, and as shown in fig. 1, in the three sleep state detection embodiments, by acquiring continuous EEG signals and processing the signals as a data sample according to 30S, applications such as sleep quality analysis, drowsiness detection, sleep disease diagnosis, etc. can be performed according to different sleep state staging tasks. The sleep quality analysis needs to classify six different sleep states, the drowsiness detection mainly detects a small amount of S1 states from Wake states, and the sleep disease diagnosis focuses on the detection of REM sleep states. The invention aims at specific application targets, and from the viewpoint of solving practical problems, the detection of the S1 and REM sleep states is divided into two stages, and in order to overcome the influence of the similarity of the two states on the detection performance, the two sleep states are taken as a whole and classified and detected with other sleep states. Because the two sleep states of S1 and REM have larger signal difference with other states, the signal channel data and sample characteristics can be utilized as less as possible. The method provided by the invention establishes a C-SVM two-class model of multi-level S1 and REM integral states and other states in the sleep quality analysis process, gradually detects different sleep states of S3, S4, Wake and S2 according to the difference of EEG signal changes, and in the process, uses an OC-SVM single-class detection model of S1 and REM integral sleep states to carry out correction detection so as to overcome the influence of less data samples and reduce the false alarm rate. In the application of drowsiness detection and sleep disease diagnosis, the two states are detected as a whole by a multi-model fusion method, and then the two states are classified, so that the non-target detection state is classified into other classification states.
The S1 and REM sleep state detection method based on the mixed channel EEG firstly detects the overall states of S1 and REM by using a multi-model fusion method, and then classifies the overall states. As shown in fig. 2, the S1 and REM sleep states in the sleep EEG data samples are marked as positive class "+ 1", and other sleep states are marked as negative class "-1", each data sample is detected by the C-SVM binary model, if the detection result is "+ 1", it is indicated as S1 or REM sleep state, and if the detection result is "-1", the OC-SVM single-class detection model of the S1 and REM overall sleep state is used for re-detection. Because the OC-SVM model only utilizes the positive class data sample, if the detection result is in the type of "+ 1", the data sample is judged to be in the sleeping state of S1 and REM, and if not, other detection states are judged. The OC-SVM single-class detection model is used for correcting the detection result of the C-SVM two-class model and detecting misclassified samples caused by data imbalance again so as to reduce the false alarm rate of S1 and REM sleep states. The C-SVM binary classification model is adopted to classify the integral state and other sleep states so as to avoid the influence of the similarity of EEG signals in the S1 and REM sleep states on each detection task, and the OC-SVM model only utilizes a small number of one class of data samples so as to overcome the influence of misclassification caused by imbalance of the data samples. The OC-SVM model training does not use the detection precision of the S1 and REM whole sleep states as a learning target, but uses the detection precision of other sleep states as a target, and aims to improve the detection precision of the S1 and REM sleep states as much as possible on the basis of ensuring the detection performance of other sleep states and avoid the reduction of the detection performance of other states caused by the improvement of the detection rate of a certain state.
In the embodiment of the invention, the parameter setting and optimization of the OC-SVM model do not directly aim at the detection precision of the positive-class sample, and the OC-SVM model assists in detecting the S1 and REM integral sleep state, so that the detection performance of other sleep states is not reduced on the basis of improving the detection rate of the integral state, and the OC-SVM model is optimized by combining the detection performances of other sleep states in the process of setting the model parameters and training the OC-SVM single-class detection model, thereby improving the detection performance of each sleep state.
After the overall states of S1 and REM sleep are detected, the C-SVM model is used for carrying out classification detection on the two sleep states, different from the C-SVM classification model for detecting the overall states, EEG signal data of two channels of Fpz-Cz and Pz-Oz are applied to model learning in the embodiment, and EEG signal data of a single channel of Fpz-Cz are only used for classification S1 and REM overall states and other sleep states. In addition, since the similarity between the S1 and the REM signals directly causes the problem of low detection rate of the two sleep states, the present invention does not directly use EEG signal sample data of the two channels, but decomposes the data samples to obtain EEG signal sub-bands reflecting the difference between the S1 and the REM sleep states. As shown in FIG. 3, the present invention utilizes fast Fourier transform to perform signal decomposition, and obtains signal sub-bands with frequencies of 1-8Hz, 1-9Hz, 9-50Hz and 30-50Hz, respectively. In the above signal sub-bands, the EEG signals of S1 and REM sleep states have power and coherent oscillation difference, and the classification accuracy of the two sleep states of S1 and REM is improved from 83.17% to 91.25% by extracting the classification characteristics of each signal sub-band and carrying out model learning.
Fig. 4 is a flowchart illustrating the method for detecting the sleep state of the S1 and REM based on the mixed channel EEG signal according to the present invention. The embodiment of the invention is mainly divided into three stages, firstly EEG signal data of Fpz-Cz and Pz-Oz channels are collected, then the integral states of S1 and REM are detected, in the process, only original signal data of the Fpz-Cz channels are utilized, according to an actual detection task, C-SVM two-class models of S1 and REM integral sleep states and other states are established, an OC-SVM single-class detection model of S1 and REM integral states is established, and the detection result of the C-SVM two-class model is corrected to detect the misclassification S1 or REM sleep state caused by data imbalance again; after detecting the integral states of S1 and REM sleep, classifying S1 and REM sleep states by utilizing EEG signal data of two channels, fully considering key factors of low detection rates of S1 and REM sleep states in the process, obtaining a plurality of signal sub-bands with different frequencies by adopting a signal decomposition method, wherein certain power and coherent oscillation differences exist in the signal sub-bands of the S1 and REM sleep states, EEG signal characteristics reflecting the two sleep states can be extracted through characteristics, and finally classifying and detecting the two sleep states which are difficult to distinguish by utilizing a C-SVM binary classification model.
According to the invention, by extracting the classification characteristics of a single-channel original EEG signal, the S1 and REM sleep states are taken as a whole to carry out two-classification detection with other sleep states, a two-classification detection model is established by utilizing a C-SVM algorithm, a single-class OC-SVM detection model is established for a whole state data sample, and the results of the C-SVM two-classification model are corrected by the S1 and the single-class detection model of the REM sleep states through a multi-model fusion detection method, so that the detection precision of the whole state is improved; further, acquiring dual-channel EEG signal data, decomposing the original signal into different frequency signal sub-bands representing the difference between S1 and REM sleep states, thereby improving the classification characteristics of S1 and REM sleep states of each signal sub-band, and establishing two classification models of the two sleep states so as to detect S1 and REM two brain sleep activity states with important research significance. According to the method, the similarity and difference characteristics of the S1 and the REM sleep state are combined, the multi-model fusion classification detection method is established by utilizing different channel signal data and extracting different classification characteristics, and the detection precision of the S1 and the REM sleep state can be effectively improved.
It should be understood that parts of the specification not set forth in detail are well within the prior art.
It should be understood that the above description of the preferred embodiments is given for clarity and not for any purpose of limitation, and that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (6)

1. A method for detecting S1 and REM sleep states based on mixed channel EEG signals, comprising the steps of:
step 1: acquiring EEG signal data of two channels of Fpz-Cz and Pz-Oz;
step 2: taking the S1 sleep state and the REM sleep state as a whole state, and detecting the whole state by using Fpz-Cz single-channel signal data;
the specific implementation of the step 2 comprises the following substeps:
step 2.1: processing Fpz-Cz single-channel EEG signal data, and taking a continuous N-second time length signal as a data sample, wherein N is a preset value;
step 2.2: extracting EEG signal classification features, specifically including features reflecting EEG amplitude variation and frequency variation characteristics;
step 2.3: establishing a C-SVM two-classification model of the S1 and REM overall state and other sleep states, wherein the modeling process is to mark the S1 and REM data samples as a +1 class and mark other sleep states as a "-1 class, and establishing a detection model through training the data samples;
step 2.4: establishing an OC-SVM single-class detection model for S1 and REM integral detection, wherein the modeling process is to only use two types of data samples, namely S1 and REM, and mark the two types of data samples as a +1 type, so as to establish a detection model for detecting S1 and REM integral sleep state;
step 2.5: judging whether an N-second data sample belongs to S1 or REM two sleep states by using a C-SVM two-classification model; if yes, executing step 3; if not, executing the step 2.6;
step 2.6: judging the data sample by using the established OC-SVM single-class detection model, if the data sample belongs to the integral state of S1 and REM, executing the step 3, otherwise, judging the data sample to be in other sleep states;
and step 3: the detected overall states of S1 and REM are classified by using signal data of two channels of Fpz-Cz and Pz-Oz to detect two different sleep states of S1 and REM, respectively.
2. The mixed channel EEG signal based S1 and REM sleep state detection method of claim 1, wherein: in step 2.3, according to the similarity of S1 and REM sleep states, taking S1 and REM as a class and Wake, S2, S3 and S4 as a class; establishing a C-SVM classification model of S1 and REM overall state and other sleep states, wherein the solved C-SVM two classification models are as follows:
Figure FDA0002998625830000011
wherein α ═ (α)1,α2,...,αN) For Lagrangian, K (x)i,xj) Is a Gaussian kernel function, xiAnd xjI and j sleep state data samples, y, in the training data, respectivelyiAnd yjLabel categories of the ith sample and the j sample in the model solving process are respectively obtained;
thus, a decision function is constructed:
Figure FDA0002998625830000021
where sign is a sign function, b is a separate hyperplane offset term, αi *The ith Lagrangian, K (x, x), learned for model trainingi) Is a Gaussian kernel function, x is a new test data sample, yiThe label category of the ith training sample; if f (x) is 1, the sample x is determined to be in S1 or REM sleep state, and if f (x) is-1, the sample x is determined to be in other sleep states.
3. The mixed channel EEG signal based S1 and REM sleep state detection method according to claim 1, wherein said OC-SVM single-class detection model in step 2.4 is:
Figure FDA0002998625830000022
wherein α ═ (α)1,α2,...,αN) For Lagrangian, K (x)i,xj) Is a Gaussian kernel function, xiAnd xjRespectively an ith sleep state data sample and a jth sleep state data sample in the training data; thus, a decision function is constructed:
Figure FDA0002998625830000023
wherein sign is a sign function, rho is a compensation term of the decision function of the OC-SVM, and alphai *The ith Lagrangian, K (x, x), learned for model trainingi) Is a Gaussian kernel function, x is a new test data sample, yiThe label category of the ith training sample; if f (x) is 1, the sample x is determined to be in S1 or REM sleep state, otherwise, the sample x is determined to be in other sleep state.
4. The mixed channel EEG signal based S1 and REM sleep state detection method as claimed in claim 1, wherein the detailed implementation of step 3 comprises the following sub-steps:
step 3.1: processing EEG signal data of two channels of Fpz-Cz and Pz-Oz, and taking a continuous N-second time length signal as a data sample, wherein N is a preset value;
step 3.2: decomposing the two channel signal data of each data sample into signal sub-bands with different frequencies;
step 3.3: extracting classification features of each decomposed frequency signal sub-band, wherein the classification features specifically comprise characteristics reflecting EEG amplitude change and characteristics reflecting frequency change;
step 3.4: establishing a C-SVM two-classification model of S1 and REM sleep states by using extracted sub-band characteristics of different frequency signals, marking an S1 data sample as a +1 class and an REM data sample as a "-1 class in the modeling process, and establishing a detection model by training the data sample;
step 3.5: and judging the detected data samples of the S1 and REM overall state by utilizing a C-SVM binary model and the characteristic data of the sub-bands of the two channels of the multi-frequency signals, and outputting the data samples into specific S1 and REM sleep states.
5. The mixed channel EEG signal based S1 and REM sleep state detection method according to claim 4, wherein: in step 3.2, the EEG signals of the two channels are decomposed by using an FFT algorithm, signal sub-bands with four frequencies of 1-8Hz, 1-9Hz, 9-50Hz and 30-50Hz are respectively obtained, feature extraction is carried out on each corresponding 8 decomposed signal sub-bands, and C-SVM two classification models of S1 and REM in two sleep states are trained in step 3.4.
6. A mixed channel EEG signal based S1 and REM sleep state detection system, comprising: the system comprises an EEG signal data acquisition module, an S1 and REM integral state detection module, and an S1 and REM sleep state classification detection module;
the EEG signal data acquisition module is used for acquiring EEG signal data of two channels of Fpz-Cz and Pz-Oz;
the S1 and REM overall state detection module is used for taking the S1 and REM two sleep states as an overall state and detecting the overall state by utilizing Fpz-Cz single-channel signal data;
the S1 and REM overall state detection module further includes the following sub-modules:
the first submodule is used for processing Fpz-Cz single-channel EEG signal data, and taking a continuous N-second time length signal as a data sample, wherein N is a preset value;
the sub-module II is used for extracting EEG signal classification features, and specifically comprises features reflecting EEG amplitude variation and frequency variation;
the submodule III is used for establishing a C-SVM two-classification model of the S1 and the REM overall state and other sleep states, the modeling process is that the S1 and the REM two-class data samples are marked as a +1 class, the other sleep states are marked as a "-1 class, and a detection model is established through training of the data samples;
the submodule IV is used for establishing an OC-SVM single-class detection model for S1 and REM integral detection, wherein the modeling process is to only use two types of data samples of S1 and REM, and mark the two types of data samples as a +1 type, so as to establish a detection model for detecting S1 and REM integral sleep states;
a submodule V for judging whether the N second data sample belongs to S1 or REM two sleep states by utilizing a C-SVM two-classification model; if yes, operating S1 and a REM sleep state classification detection module; if not, operating the submodule six;
a submodule six, which is used for judging the data sample by using the established OC-SVM single-class detection model, if the data sample belongs to the integral state of S1 and REM, operating the S1 and REM sleep state classification detection module, and otherwise, judging the data sample to be in other sleep states;
the S1 and REM sleep state classification detection module is used for carrying out two classifications on the detected S1 and REM overall state by utilizing signal data of two channels of Fpz-Cz and Pz-Oz so as to respectively detect two different sleep states of S1 and REM.
CN202010606885.5A 2020-06-29 2020-06-29 S1 and REM sleep state detection method and system based on mixed channel EEG signal Active CN111743513B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010606885.5A CN111743513B (en) 2020-06-29 2020-06-29 S1 and REM sleep state detection method and system based on mixed channel EEG signal

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010606885.5A CN111743513B (en) 2020-06-29 2020-06-29 S1 and REM sleep state detection method and system based on mixed channel EEG signal

Publications (2)

Publication Number Publication Date
CN111743513A CN111743513A (en) 2020-10-09
CN111743513B true CN111743513B (en) 2021-05-18

Family

ID=72678087

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010606885.5A Active CN111743513B (en) 2020-06-29 2020-06-29 S1 and REM sleep state detection method and system based on mixed channel EEG signal

Country Status (1)

Country Link
CN (1) CN111743513B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113080864B (en) * 2021-04-07 2022-02-01 电子科技大学 Common sleep disease detection method through automatic sleep staging results
CN113180667B (en) * 2021-04-29 2022-09-23 南昌虚拟现实研究院股份有限公司 Cognitive load evaluation method and device, storage medium and computer equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104720748A (en) * 2013-12-24 2015-06-24 中国移动通信集团公司 Sleep stage determining method and sleep stage determining system
WO2018027141A1 (en) * 2016-08-05 2018-02-08 The Regents Of The University Of Colorado, A Body Corporate In-ear sensing systems and methods for biological signal monitoring
CN110432870A (en) * 2019-08-13 2019-11-12 重庆邮电大学 A kind of sleep signal based on 1D CNN-LSTM method by stages automatically

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104720748A (en) * 2013-12-24 2015-06-24 中国移动通信集团公司 Sleep stage determining method and sleep stage determining system
WO2018027141A1 (en) * 2016-08-05 2018-02-08 The Regents Of The University Of Colorado, A Body Corporate In-ear sensing systems and methods for biological signal monitoring
CN110432870A (en) * 2019-08-13 2019-11-12 重庆邮电大学 A kind of sleep signal based on 1D CNN-LSTM method by stages automatically

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Cascaded LSTM recurrent neural network for automated sleep stage classification using single-channel EEG signals;Michielli N，Acharya U R，Molinari F;《Computers in Biology and Medicine》;20190331;第106卷;第71-81页 *
On the Classification of Sleep States By Means of Statistical and Spectral Features from Single Channel Electroencephalogram;Hassan A R，Bashar S K，Bhuiyan M I H;《International Conference on Advances in Computing，Communications and Informatics》;20150813;第2238-2243页 *

Also Published As

Publication number Publication date
CN111743513A (en) 2020-10-09

Similar Documents

Publication Publication Date Title
CN106886792B (en) Electroencephalogram emotion recognition method for constructing multi-classifier fusion model based on layering mechanism
Hussein et al. Optimized deep neural network architecture for robust detection of epileptic seizures using EEG signals
Ryu et al. Prediction model of dementia risk based on XGBoost using derived variable extraction and hyper parameter optimization
Yaman et al. Comparison of bagging and boosting ensemble machine learning methods for automated EMG signal classification
Li et al. Classification of epilepsy EEG signals using DWT-based envelope analysis and neural network ensemble
Qin et al. Combining low-dimensional wavelet features and support vector machine for arrhythmia beat classification
Hu Comparison of different features and classifiers for driver fatigue detection based on a single EEG channel
Abdulla et al. Sleep EEG signal analysis based on correlation graph similarity coupled with an ensemble extreme machine learning algorithm
CN110897639A (en) Electroencephalogram sleep staging method based on deep convolutional neural network
Kalaivani et al. Analysis of EEG signal for the detection of brain abnormalities
Sekkal et al. Automatic sleep stage classification: From classical machine learning methods to deep learning
Sharaf et al. An automated approach for epilepsy detection based on tunable Q-wavelet and firefly feature selection algorithm
CN111743513B (en) S1 and REM sleep state detection method and system based on mixed channel EEG signal
Khalighi et al. Adaptive automatic sleep stage classification under covariate shift
Alvarez-Meza et al. Kernel-based relevance analysis with enhanced interpretability for detection of brain activity patterns
Al-Salman et al. Sleep stage classification in EEG signals using the clustering approach based probability distribution features coupled with classification algorithms
Echtioui et al. A novel ensemble learning approach for classification of EEG motor imagery signals
Taran et al. Drowsiness detection using instantaneous frequency based rhythms separation for EEG signals
Zan et al. Local Pattern Transformation-Based convolutional neural network for sleep stage scoring
Ohannesian et al. Epileptic seizures detection from EEG recordings based on a hybrid system of Gaussian mixture model and random forest classifier
Ji et al. 3DSleepNet: A multi-channel bio-signal based sleep stages classification method using deep learning
Mahato et al. Analysis of region of interest (RoI) of brain for detection of depression using EEG signal
Chavan et al. A review on BCI emotions classification for EEG signals using deep learning
Goshvarpour et al. An Innovative Information-Based Strategy for Epileptic EEG Classification
Rukhsar et al. Barnes–Hut approximation based accelerating t-SNE for seizure detection

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
GR01 Patent grant
GR01 Patent grant