CN113842112A - Sleep arousal detection method and device, computer equipment and readable storage medium - Google Patents
Sleep arousal detection method and device, computer equipment and readable storage medium Download PDFInfo
- Publication number
- CN113842112A CN113842112A CN202010599319.6A CN202010599319A CN113842112A CN 113842112 A CN113842112 A CN 113842112A CN 202010599319 A CN202010599319 A CN 202010599319A CN 113842112 A CN113842112 A CN 113842112A
- Authority
- CN
- China
- Prior art keywords
- arousal
- sleep
- sample
- level representation
- detection model
- 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
- 230000037007 arousal Effects 0.000 title claims abstract description 299
- 238000001514 detection method Methods 0.000 title claims abstract description 164
- 230000007958 sleep Effects 0.000 title claims abstract description 134
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000000034 method Methods 0.000 claims abstract description 31
- 238000004590 computer program Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 8
- 230000002045 lasting effect Effects 0.000 claims description 4
- 230000002349 favourable effect Effects 0.000 abstract description 7
- 230000008569 process Effects 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 210000004373 mandible Anatomy 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000000241 respiratory effect Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000003860 sleep quality Effects 0.000 description 2
- 230000001052 transient effect Effects 0.000 description 2
- 230000002618 waking effect Effects 0.000 description 2
- 208000000884 Airway Obstruction Diseases 0.000 description 1
- 206010062519 Poor quality sleep Diseases 0.000 description 1
- 208000032140 Sleepiness Diseases 0.000 description 1
- 206010041235 Snoring Diseases 0.000 description 1
- 206010041349 Somnolence Diseases 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000036403 neuro physiology Effects 0.000 description 1
- 230000008452 non REM sleep Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 230000035790 physiological processes and functions Effects 0.000 description 1
- 238000005498 polishing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000004461 rapid eye movement Effects 0.000 description 1
- 230000036385 rapid eye movement (rem) sleep Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 201000002859 sleep apnea Diseases 0.000 description 1
- 208000019116 sleep disease Diseases 0.000 description 1
- 230000008667 sleep stage Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000003786 synthesis 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
-
- 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
- A61B5/48—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
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The embodiment of the invention provides a sleep arousal detection method, a device, computer equipment and a readable storage medium, wherein the method comprises the following steps: acquiring sleep data to be processed, wherein the sleep data comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal; inputting the sleep data into a arousal detection model, and outputting arousal classification results of the sleep data, wherein the arousal detection model is obtained by taking known sleep data and arousal classification results of the known sleep data as sample training classifiers. The scheme can avoid the influence of manual work or individual difference on detecting the arousal, is favorable for improving the accuracy of detecting the arousal and is favorable for improving the rapidity of detecting the arousal.
Description
Technical Field
The invention relates to the technical field of biological signal processing, in particular to a sleep arousal detection method and device, computer equipment and a readable storage medium.
Background
Arousal refers to a transient phenomenon that may lead to wakefulness or simply to an interruption of transient sleep. Researchers are particularly concerned with arousals, a recurring condition of brief interruptions in sleep. This is because they reflect a wide range of sleep-related physiological phenomena, with sleep apnea being the most common phenomenon, but not the only manifestation of sleep disorders affected by arousals. Arousals during sleep may be influenced by physiological factors and may be caused spontaneously, mainly by tooth polishing, partial airway obstruction and even snoring. In addition, arousals can result in sudden changes from rapid eye movement sleep to non-rapid eye movement sleep, which in some cases can result in prolonged waking. Therefore, in the extensive understanding of human sleep neurophysiology, accurately identifying arousals during sleep plays a key role in monitoring and improving sleep quality.
In the field of sleep research, arousal is one of the important criteria for determining the quality of sleep in a subject. Generally, the arousal label can provide reference data for judging the problems of the sleep quality of the subject, the reason of the somnolence in daytime and the like. Meanwhile, reference data can be provided for filtering the arousal characteristic waveform so as to eliminate the interference of the arousal characteristic waveform on the interpretation of other sleep labels, such as the interpretation of sleep stages and respiratory events.
A trained sleep technician can confirm the occurrence of arousals by interpreting polysomnography waveforms. The arousal interpretation is complex, is greatly influenced by subjective factors of interpretation technicians, has obvious difference among different individuals, influences the accuracy of the arousal interpretation, and is the key point of research on how to accurately select the arousal interpretation characteristics. The traditional manual design has limited number of features and has limitations.
Disclosure of Invention
The embodiment of the invention provides a sleep arousal detection method, which aims to solve the technical problem of low accuracy in arousal detection in the prior art. The method comprises the following steps:
acquiring sleep data to be processed, wherein the sleep data comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal;
inputting the sleep data into a arousal detection model, and outputting arousal classification results of the sleep data, wherein the arousal detection model is obtained by taking known sleep data and arousal classification results of the known sleep data as sample training classifiers.
The embodiment of the invention also provides a sleep arousal detection device, which is used for solving the technical problem of low accuracy in arousal detection in the prior art. The device includes:
the data acquisition module is used for acquiring sleep data to be processed, wherein the sleep data comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal;
and the arousal detection module is used for inputting the sleep data into an arousal detection model and outputting arousal classification division results of the sleep data, wherein the arousal detection model is obtained by taking known sleep data and arousal classification division results of the known sleep data as sample training classifiers.
The embodiment of the invention also provides computer equipment which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor executes the computer program to realize the arbitrary sleep arousal detection method so as to solve the technical problem of low accuracy in arousal detection in the prior art.
An embodiment of the present invention further provides a computer-readable storage medium, where a computer program for executing any one of the above sleep arousal detection methods is stored in the computer-readable storage medium, so as to solve the technical problem in the prior art that the accuracy is low in arousal detection.
In the embodiment of the invention, the automatic sleep arousal detection based on the information (for example, the information of a plurality of channels including electroencephalogram signals, electro-oculogram signals, mandible electromyogram signals and the like) of a plurality of channels is realized by adopting a trained arousal detection model, and the automatic sleep arousal detection process adopts the electroencephalogram signals, the electro-oculogram signals and the mandible electromyogram signals, so that the automatic sleep arousal detection accuracy can be improved compared with the mode of carrying out the automatic sleep arousal detection only according to the single-channel electroencephalogram signals; meanwhile, the arousal classification result of a large amount of known sleep data and known sleep data is adopted by the arousal detection model to be obtained as a sample training classifier, the problem of limited characteristic quantity can be avoided, and then the arousal detection model is used for carrying out arousal detection on the sleep data to be processed.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is a flow chart of a method of sleep arousal detection provided by an embodiment of the present invention;
FIG. 2 is a flowchart of a method embodying the above-described sleep arousal detection method according to an embodiment of the present invention;
FIG. 3 is a block diagram of a computer device according to an embodiment of the present invention;
fig. 4 is a block diagram of a sleep arousal detection apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.
In an embodiment of the present invention, there is provided a sleep arousal detection method, as shown in fig. 1, the method including:
step 102: acquiring sleep data to be processed, wherein the sleep data comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal;
step 104: inputting the sleep data into a arousal detection model, and outputting arousal classification results of the sleep data, wherein the arousal detection model is obtained by taking known sleep data and arousal classification results of the known sleep data as sample training classifiers.
As can be seen from the flow shown in fig. 1, in the embodiment of the present invention, an automatic sleep arousal detection based on information of multiple channels (for example, information of multiple channels including electroencephalogram signals, electro-oculogram signals, and mandibular electromyogram signals) is implemented by using a trained arousal detection model, and the automatic sleep arousal detection process is advantageous to improve the accuracy of automatic sleep arousal detection compared with a mode of performing automatic sleep arousal detection only according to a single-channel electroencephalogram signal because of using electroencephalogram signals, electro-oculogram signals, and mandibular electromyogram signals; meanwhile, the arousal classification result of a large amount of known sleep data and known sleep data is adopted as a sample training classifier to obtain the arousal detection model, the problem of limited characteristic quantity can be avoided, then the arousal detection model is used for carrying out arousal detection on the sleep data to be processed, the arousal classification result can be output by inputting the sleep data to be processed into the arousal detection model, compared with the mode of manually detecting arousal in the prior art, the arousal detection model can avoid the influence of manual work or individual difference on the arousal detection, is favorable for improving the arousal detection accuracy, and is favorable for improving the arousal detection rapidity.
In a specific implementation, the sleep data (i.e. the sleep data to be processed and the known sleep data) may include any one or any combination of an electroencephalogram signal, an electrooculogram signal, and a mandibular electromyogram signal, and the electroencephalogram signal, the electrooculogram signal, and the mandibular electromyogram signal included in the sleep data may be extracted from a PSG (polysomnogram) signal.
In specific implementation, the known sleep data and the arousal classification result of the known sleep data are used as samples to train a classifier, wherein, the known sleep data also comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal, the known sleep data can also be extracted from the known PSG signal, the arousal classification result of the known sleep data can be determined by a doctor or other algorithm analysis methods, the principle and the mode of arousal classification are not limited in the application, the method adopts the known sleep data and the arousal classification result of the known sleep data as samples, obtains the arousal detection model after deep learning training classifier, then detects arousal by using the arousal detection model, the method is characterized in that the to-be-processed sleep data is input into a arousal detection model, and then arousal classification results can be output.
In particular implementations, the classifiers can include, but are not limited to, naive bayes, decision trees, logistic regression, support vector machines, neural networks (e.g., convolutional neural networks and recurrent neural networks in any form), and the like.
In a specific implementation, the arousal classification result may include, but is not limited to: the sleep period may include any one or any combination of a wake period, a sleep period without arousals, and a sleep period with arousals, as the present application is not limited in this respect.
In particular, after extensive research, the present inventors have discovered that internal brain arousals disrupt the typical pattern of electrical activity in healthy sleep. For example, the disruption of the pattern of electrical activity by arousals can be visually monitored from an electroencephalogram signal. In addition to electroencephalograms, arousals affect the physiological state of the entire body, and therefore indirectly affect physiological indices such as electromyograms, electrooculogram, electrocardiogram, oxygen saturation, respiratory airflow, and respiratory movement. Therefore, the inventor of the application provides that polysomnography for collecting all the vital signs under a multidimensional time sequence is a main means for detecting arousal, and the principle of detecting the arousal classification result through a trained arousal detection model can be based on electroencephalogram signals for detection, also can be based on electroencephalogram signals and mandibular electromyogram signals for detection, and also can be based on electroencephalogram signals and electrooculogram signals for detection, even can be based on electrooculogram signals for detection.
Specifically, the arousal detection model is based on the principle of carrying out arousal detection on electroencephalogram signals and mandibular electromyogram signals or the mode of carrying out arousal classification, and the arousal detection model is not specifically limited in the application, for example, the existing detection principle can be adopted, for example, the electroencephalogram frequency is suddenly changed, waves including alpha waves, theta waves and (or) waves with the frequency being more than 16Hz (but not shuttle waves) are generated, the lasting time lasts for at least 3s, and the arousal can be interpreted as arousal after the stable sleep of at least 10s before the frequency change, if the arousal detection model is in the rapid eye movement period, the arousal detection model also requires that the electroencephalogram (electroencephalogram signals) is changed and the mandibular electromyogram (at least 1 s) is increased at the same time, and the arousal can be interpreted as arousal at this time.
In addition, in order to improve the accuracy of the arousal detection, the arousal detection model can adopt other methods based on the principle of carrying out arousal detection on the basis of electroencephalogram signals and eye signals or the mode of carrying out arousal classification.
During specific implementation, partial channel signals in the sleep data may cause poor signal quality under the conditions of electrode falling, poor contact and the like, for example, the electroencephalogram signal quality is poor and cannot be used for detection, the inventor of the application further provides a arousal detection model which can be used for arousal judgment based on the electrooculogram signal alone, and if the judgment principle is that the electrooculogram signal has alpha waves and/or theta waves lasting for a preset time length, arousal can be judged, the judgment mode can partially make up the influence caused by the fact that some signal quality differences cannot be detected, and the judgment mode is better in stability compared with the existing single-channel electroencephalogram arousal judgment method.
In particular, in order to improve the accuracy of the arousal detection model, in this embodiment, the classifier is trained to obtain the arousal detection model by:
training classifiers by adopting all samples to obtain a primary arousal detection model; specifically, in the process of training the classifier, all samples can be divided into a training set, a verification set and a test set, and a preliminary arousal detection model is obtained by training a neural network.
For each sample, inputting the sleep data of the sample into the preliminary arousal detection model, and taking the output of the intermediate layer of the preliminary arousal detection model as a high-level representation sample of the sample to further obtain the high-level representation sample of each sample; specifically, the output of the intermediate layer of the preliminary arousal detection model comprises the output of one or more hidden layers of the preliminary arousal detection model, namely, the output of any hidden layer or the output of any combination of multiple hidden layers of the preliminary arousal detection model can be used as a high-level representation sample.
Performing density clustering on high-level representation samples of the samples belonging to the class of arousal categories according to each class of arousal categories to obtain a plurality of high-level representation subsets belonging to the class of arousal categories; obtaining a high-level representation subset belonging to each class of arousal category; specifically, as the detection of arousal depends mainly on the characteristic waveforms in each signal channel, in order to overcome the problem that the characteristic waveforms are not obvious enough when integrating the information of a plurality of channels, the processing can be carried out by carrying out density clustering on the high-level representation samples, so that a plurality of high-level representation subsets belonging to each type of arousal category respectively contain sufficient waveform information with different characteristics.
And training a classifier by adopting a high-level representation subset belonging to each type of arousal category to obtain the final arousal detection model. Specifically, in the process of training the classifier by using the high-level representation subset belonging to each type of arousal category, the high-level representation subset may be divided into a training set, a verification set and a test set, and then the classifier is trained and fine-tuned to obtain the final arousal detection model.
In specific implementation, in this embodiment, density clustering may be performed on the high-level representation samples of the samples belonging to the class of arousal categories to obtain a plurality of high-level representation subsets belonging to the class of arousal categories by:
reducing the dimension of the high-level representation sample with the dimension larger than a preset dimension threshold value aiming at the high-level representation sample of each sample belonging to the class of arousal categories until the dimension is equal to the preset dimension threshold value; specifically, the dimension may be the number of data points representing the sample at a high level, the preset dimension threshold should be set to ensure that sufficient sample information is retained in the dimension, and the dimension reduction method may include, but is not limited to, methods such as principal component analysis, local retention projection, laplacian feature mapping, local linear embedding, and linear discriminant analysis.
Calculating the distance between every two high-level representation samples aiming at all high-level representation samples belonging to the class of arousal classes;
calculating the number of high-level representation samples with the distance from the high-level representation samples to be smaller than a preset distance threshold value aiming at each high-level representation sample, taking the number as the local density of the high-level representation samples, and taking the high-level representation sample with the maximum local density as the clustering center of the class of the arousal category;
and sequentially dividing all high-level representation samples belonging to the class of arousal categories into a plurality of high-level representation subsets according to the distance sequence from the cluster center.
Specifically, in the process of sequentially dividing all high-level representation samples belonging to the class of arousal categories into a plurality of high-level representation subsets, the high-level representation subsets are formed according to the sequence of the distances from the clustering center to the high-level representation subsets for each class of arousal categories, and the characteristic waveform is more obvious when the high-level representation subsets including the high-level representation samples are more distant from the clustering center.
For example, a set of high-level representative samples belonging to each arousal category is recorded as
Di＝{(x,y)|y＝i}
Where x represents a high level representation sample, y represents a corresponding arousal category, and i ═ 1, 2.
Representing sample sets D at a high level for each arousal categoryiIn (1), each sample is recorded as
dij＝(xij,yj)
Wherein x isijA higher layer denoted by (i 1, 2.. and n) represents a sample, i represents a corresponding arousal category, j represents a corresponding sample number, j 1, 2.. and num _ Di,num_DiRepresents DiTotal number of elements of the set. y isjRepresents the corresponding arousal category, yj＝i。
Calculating the distance between each high-level representation sample, and calculating DiAny two samples d in the setijAnd dik(dikK in (2) also represents the sample number, except that the value of k is different from j, j ≠ k,1 ≦ j, k ≦ num _ Di) Distance between them is recorded as disijk：
disijk＝|xij-xik|2(1≤j,k≤num_Di)
Setting a distance threshold disTDistance threshold disTThe distance between all the high-level representation samples can be selected to be arranged from small to large and then positioned at a proper position, and the specific implementation can be adjusted according to the actual situation.
The distance of the statistics to each high level representation sample is smaller than a distance threshold disTAs the local density p of the high level representation samplesijI.e. by
Wherein I (-) is an indicator function, disTFor all disijkThe value of T% quantile in the order of small to large, T is 0,100]Is a preset parameter.
Taking the high-level representation sample with the maximum local density in each arousal category as the clustering center of the arousal category, and recording the clustering center of the ith arousal category as dic＝(xic,yicI) wherein
Calculating the distance dis between each high-level representation sample and the cluster center of the class of arousal detection classesickAnd sorting is performed. According to the distance, all high-level representation samples corresponding to the class of arousal category can be sequentially divided into N high-level representation subsets.
Any two subsets Dip,Diq(p, q ═ 1, 2.., N) should satisfy
Dip∩Diq＝Φ(p≠q)
And is
Thus, N subsets of high-level representations, denoted as "high-level representation", of the overall arousal detection samples may be obtained for further training of the classifier and adjustment
Is to disickDividing into N sets in descending order by Dip(p ═ 1, 2.., N) denotes DiRepresents a subset by the p-th higher layer of (2), denoted by Diq(q ═ 1, 2.., N) denotes DiIs represented by the qth higher layer of (1), by
The representation contains the pth subset of high level representations for each category of arousal.
In particular, the overall accuracy of the model is affected by the problem of uneven sample numbers found in the individual arousal categories. To further improve the accuracy of the arousal detection model, in this embodiment, data enhancement should be performed on the subset of the high-level representations of each category of arousal before training the classifier with the subset of the high-level representations belonging to each category of arousal. Specifically, the data enhancement method may include, but is not limited to: generating a countermeasure network, adjusting data training weight, increasing and decreasing sampling, and the like.
In specific implementation, the sleep data can be influenced by the problems of electrode falling, poor contact, power supply noise, body movement and interference of other bioelectrical signals and the like in the acquisition process, and meanwhile, the arousal classification is greatly influenced by the skills and habits of technicians, and the variation rate is high. The sample quality is different due to the above reasons, and the model accuracy is affected, and for this embodiment, classifier training is performed by using the high-level representation subsets of the above arousal categories. According to the distance between each high-level representation sample and the clustering center of the type of arousal, sequentially dividing all high-level representation samples corresponding to the type of arousal into a plurality of high-level representation subsets, and sequentially training and adjusting a classifier to further obtain an automatic arousal detection model. Firstly, training a classifier by using a high-level representation subset nearest to a clustering center, then arranging the classifiers from small to large according to the distance from the clustering center, and sequentially adjusting the classifiers by using the high-level representation subset to reduce the influence of sample quality fluctuation, thereby finally obtaining the automatic arousal detection model.
In order to further improve the efficiency of the arousal detection model in specific implementation, in this embodiment, it is proposed to reduce the number of final parameters of the arousal detection model while not losing accuracy or losing less accuracy. For example, a global average pooling layer or a global maximum pooling layer may be employed to replace the fully-connected layer of the classifier, i.e., replace the fully-connected layer of an existing classifier with the global average pooling layer or the global maximum pooling layer.
In the specific implementation, if the judgment of the classification boundary of the arousal classes is fuzzy, the arousal detection needs to be judged by long-time sample data, the inventor of the application finds that the arousal detection is greatly influenced by time domain adjacent signals, proposes a high-level representation sample in a high-level representation subset of each class of arousal classes, splices each frame of data and the time domain adjacent data and trains a classifier to obtain a final arousal detection model, thus splices the adjacent time domain data of each frame of sample as an auxiliary judgment basis and the current frame of data to train the arousal detection model to improve the accuracy of the automatic arousal detection model, and can accurately distinguish the arousal period from the waking period. The time domain adjacent data refers to a signal data segment with a certain length adjacent to a currently judged frame data segment in a time domain, the adjacent can be front and back adjacent to the currently judged frame data segment, specifically, each frame of data can be spliced with the front adjacent and/or back adjacent time domain adjacent data to train a classifier, and when the arousal detection model is applied, each frame of data in the sleep data to be processed can be spliced with the time domain adjacent data and then input into the arousal detection model.
In the specific implementation, the noise influence caused by the factors such as electrode falling, poor contact, power supply noise, body movement and other bioelectricity signal interference is further eliminated. Before the sample training classifier is adopted, noise and artifact removal processing can be carried out on the sample, the data sampling frequency is changed to be the same fixed frequency, and then normalization processing is carried out on the sleep data.
Specifically, the following describes in detail a process of implementing the above-described sleep arousal detection method, as shown in fig. 2, the process including the steps of:
step S1, extracting electroencephalogram signals, electro-oculogram signals, mandibular electromyogram signals and technician-marked arousal labels (namely arousal category division results) from the PSG monitoring file;
step S2, preprocessing the acquired electroencephalogram signal, electro-oculogram signal, mandibular electromyogram signal and arousal label as samples to obtain preprocessed data samples as arousal detection sample data sets;
the step S2 may specifically include S21, S22, S23 and S24:
step S21, recording the electroencephalogram, electro-oculogram, and mandibular electromyogram obtained from the night monitoring record as EEG _ data, EOG _ data, and EMG _ data, respectively, and converting the corresponding Arousal label marked by the technician into a classification label form as Arousal _ label, where the classification may include but is not limited to: a wake period, a sleep period containing no arousals, a sleep period containing arousals;
step S22, selecting a fixed frequency F, and performing frequency conversion processing on the electroencephalogram signal, the ocular electrical signal and the mandibular electromyogram signal obtained in the step S21 to change the sampling frequency into the fixed frequency F and record the fixed frequency F as EEG _ F, EOG _ F, EMG _ F;
step S23, normalization processing is carried out on the electroencephalogram signal, the electro-oculogram signal and the mandibular electromyogram signal obtained in the step S22, and normalized data EEG _ norm, EOG _ norm and EMG _ norm are obtained;
and step S24, splicing the EEG signal EEG _ norm, the eye electrical signal EOG _ norm and the mandible electrical signal EMG _ norm processed in the steps, namely splicing adjacent time domain data and current frame data of each frame of sample. And taking the spliced signal and the Arousal label Arousal _ label as an Arousal detection sample data set.
And step S3, dividing the arousal detection sample data set, and performing neural network training in a training set, test set and verification set mode to obtain a preliminary automatic arousal detection model.
The step S3 may specifically include steps S31 and S32:
and S31, dividing the arousal detection sample data set obtained in the step S2 into a training set, a test set and a verification set.
And step S32, designing a proper neural network, and training the neural network in a training set, a test set and a verification set mode to obtain a preliminary automatic arousal detection model. The neural network form is not limited in this embodiment, and may be any form including a convolutional neural network and a cyclic neural network.
Step S4, processing the sleep data samples using the preliminary automatic arousal detection model, and outputting the intermediate layer as the high-level representation samples, wherein the output of one or more hidden layers of the preliminary automatic arousal detection model should be included.
And step S5, performing density clustering calculation on the high-level representation sample sets of each type of arousal category, and sequentially dividing all the high-level representation samples corresponding to each arousal category into a plurality of high-level representation subsets.
The step S5 may specifically include S51, S52 and S52:
step S51, recording the high-level representation sample set belonging to each arousal detection category as
Di＝{(x,y)|y＝i}
Where x represents a high level representation sample, y represents a corresponding arousal detection category, and i is 1, 2.
Representing sample sets D at a high level for each arousal categoryiIn (1), each sample is recorded as
dij＝(xij,yj)
Wherein xijA higher layer denoted by (i 1, 2.. and n) denotes a sample, i denotes a corresponding arousal detection category, j denotes a corresponding sample number, j 1, 2.. and num _ Di,num_DiRepresents DiTotal number of elements of the set. y isjIndicates the corresponding class of arousal detection, yj＝i。
Calculating the distance between each high-level representation sample, and calculating DiAny two samples d in the setijAnd dik(j≠k,1≤j,k≤num_Di) The distance between them is recorded as
disijk＝|xij-xik|2(1≤j,k≤num_Di)。
Step S52, setting distance threshold disTDistance threshold disTThe distance between all the high-level representation samples can be selected to be arranged from small to large and then positioned at a proper position, and the specific implementation can be adjusted according to the actual situation.
The distance of the statistics to each high level representation sample is smaller than a distance threshold disTAs the local density p of the high level representation sampleijI.e. by
Wherein I (-) is an indicator function.
Taking the high-level representation sample with the maximum local density in each arousal category as the clustering center of the arousal category, and recording the clustering center of the ith arousal category as dic＝(xic,yic＝i)
Wherein
Step S53, calculating the distance dis between each high-level representation sample and the cluster center of the class of arousal categoryickAnd sorting is performed. According to the distance, all high-level representation samples corresponding to the class of arousal category can be sequentially divided into N high-level representation subsets.
Any two subsets Dip,Diq(p, q ═ 1, 2.., N) should satisfy
Dip∩Diq＝Φ(p≠q)
And is
Thus, N subsets of high-level representations, denoted as "high-level representation", of the overall arousal detection samples may be obtained for further training of the classifier and adjustment
Step S6, counting the number of samples in each high-level representation subset in step S5, and carrying out sample balance processing aiming at the problem of unbalanced number of samples in each arousal category;
and step S7, performing staged model training according to a learning strategy by using the processed high-level sample representation subset to obtain a final automatic arousal detection model. According to the distance between each high-level representation sample and the clustering center of the type of arousal, training and adjusting the classifier by a plurality of high-level representation subsets which are sequentially divided by all high-level representation samples corresponding to the type of arousal, and further obtaining the automatic arousal detection model.
In the present embodiment, a computer device is provided, as shown in fig. 3, comprising a memory 302, a processor 304, and a computer program stored on the memory and executable on the processor, the processor implementing any of the sleep arousal detection methods described above when executing the computer program.
In particular, the computer device may be a computer terminal, a server or a similar computing device.
In the present embodiment, there is provided a computer-readable storage medium storing a computer program for executing any of the sleep arousal detection methods described above.
In particular, computer-readable storage media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable storage medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
Based on the same inventive concept, embodiments of the present invention further provide a sleep arousal detection apparatus, as described in the following embodiments. Because the principle of the device for detecting sleep arousal is similar to that of the method for detecting sleep arousal, the implementation of the device for detecting sleep arousal can refer to the implementation of the method for detecting sleep arousal, and repeated parts are not repeated. As used hereinafter, the term "unit" or "module" may be a combination of software and/or hardware that implements a predetermined function. Although the means described in the embodiments below are preferably implemented in software, an implementation in hardware, or a combination of software and hardware is also possible and contemplated.
Fig. 4 is a block diagram showing a configuration of a sleep arousal detection apparatus according to an embodiment of the present invention, as shown in fig. 4, including:
a data acquisition module 402, configured to acquire sleep data to be processed, where the sleep data includes an electroencephalogram signal, an electrooculogram signal, and a mandibular electromyogram signal;
and a arousal detection module 404, configured to input the sleep data into an arousal detection model and output an arousal classification result of the sleep data, wherein the arousal detection model is obtained by training a classifier with known sleep data and arousal classification result of the known sleep data as samples.
In one embodiment, the arousal detection module includes:
the first training unit is used for training the classifier by adopting all samples to obtain a primary arousal detection model;
a high-level representation sample acquisition unit, configured to, for each sample, input sleep data of the sample into the preliminary arousal detection model, and obtain a high-level representation sample of each sample by using an output of the intermediate layer of the preliminary arousal detection model as the high-level representation sample of the sample;
the density clustering unit is used for performing density clustering on the high-level representation samples of the samples belonging to the class of arousal categories according to each class of arousal categories to obtain a plurality of high-level representation subsets belonging to the class of arousal categories; obtaining a high-level representation subset belonging to each class of arousal category;
and the second training unit is used for training the classifier by adopting the high-level representation subset belonging to each type of arousal category to obtain the final arousal detection model.
In an embodiment, the second training unit is configured to, for the high-level representation samples in the high-level representation subset of each class of arousal categories, concatenate each frame of data with time-domain neighboring data to train a classifier, and obtain the final arousal detection model.
In one embodiment, the arousal detection module is configured to determine a waveform as arousal by the arousal detection model when the waveform is detected that satisfies the following conditions:
the electro-ocular signal and the electroencephalogram signal synchronously generate alpha waves and/or theta waves, or the electro-ocular signal generates alpha waves and/or theta waves lasting for preset duration.
In one embodiment, the fully connected layer in the classifier is replaced by a global average pooling layer or a global maximum pooling layer.
The embodiment of the invention realizes the following technical effects: the method adopts a trained arousal detection model to realize automatic sleep arousal detection based on the synthesis of information of a plurality of channels (for example, information of a plurality of channels such as electroencephalogram signals, electro-oculogram signals and mandibular electromyogram signals), and compared with a mode of carrying out automatic sleep arousal detection only according to single-channel electroencephalogram signals, the method is favorable for improving the accuracy of automatic sleep arousal detection due to the adoption of the electroencephalogram signals, the electro-oculogram signals and the mandibular electromyogram signals in the automatic sleep arousal detection process; meanwhile, the arousal classification result of a large amount of known sleep data and known sleep data is adopted as a sample training classifier to obtain the arousal detection model, the problem of limited characteristic quantity can be avoided, then the arousal detection model is used for carrying out arousal detection on the sleep data to be processed, the arousal classification result can be output by inputting the sleep data to be processed into the arousal detection model, compared with the mode of manually detecting arousal in the prior art, the arousal detection model can avoid the influence of manual work or individual difference on the arousal detection, is favorable for improving the arousal detection accuracy, and is favorable for improving the arousal detection rapidity.
It will be apparent to those skilled in the art that the modules or steps of the embodiments of the invention described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and alternatively, they may be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described may be performed in an order different than that described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple ones of them may be fabricated into a single integrated circuit module. Thus, embodiments of the invention are not limited to any specific combination of hardware and software.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made to the embodiment of the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method of sleep arousal detection, comprising:
acquiring sleep data to be processed, wherein the sleep data comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal;
inputting the sleep data into a arousal detection model, and outputting arousal classification results of the sleep data, wherein the arousal detection model is obtained by taking known sleep data and arousal classification results of the known sleep data as sample training classifiers.
2. The method of sleep arousal detection according to claim 1, wherein training a classifier to obtain the arousal detection model comprises:
training classifiers by adopting all samples to obtain a primary arousal detection model;
for each sample, inputting the sleep data of the sample into the preliminary arousal detection model, and taking the output of the intermediate layer of the preliminary arousal detection model as a high-level representation sample of the sample to obtain the high-level representation sample of each sample;
performing density clustering on high-level representation samples of the samples belonging to the class of arousal categories according to each class of arousal categories to obtain a plurality of high-level representation subsets belonging to the class of arousal categories; obtaining a high-level representation subset belonging to each class of arousal category;
and training a classifier by adopting a high-level representation subset belonging to each type of arousal category to obtain the final arousal detection model.
3. The method of sleep arousal detection according to claim 2, wherein the training of the classifier using the subset of high level representations belonging to each category of arousal to obtain the final arousal detection model comprises:
and for the high-level representation samples in the high-level representation subset of each type of arousal category, splicing each frame of data with time domain adjacent data and then training a classifier to obtain the final arousal detection model.
4. The method for sleep arousal detection according to any one of claims 1 to 3 further comprising:
the arousal detection model determines a waveform as arousal when the waveform is detected that meets the following conditions:
the electro-ocular signal and the electroencephalogram signal synchronously generate alpha waves and/or theta waves, or the electro-ocular signal generates alpha waves and/or theta waves lasting for preset duration.
5. The method for sleep arousal detection according to any one of claims 1 to 3 further comprising:
the fully connected layer of the classifier is replaced with a global average pooling layer or a global maximum pooling layer.
6. A sleep arousal detection apparatus comprising:
the data acquisition module is used for acquiring sleep data to be processed, wherein the sleep data comprises an electroencephalogram signal, an electro-oculogram signal and a mandibular electromyogram signal;
and the arousal detection module is used for inputting the sleep data into an arousal detection model and outputting arousal classification division results of the sleep data, wherein the arousal detection model is obtained by taking known sleep data and arousal classification division results of the known sleep data as sample training classifiers.
7. The sleep arousal detection apparatus of claim 6 wherein the arousal detection module comprises:
the first training unit is used for training the classifier by adopting all samples to obtain a primary arousal detection model;
a high-level representation sample acquisition unit, configured to, for each sample, input sleep data of the sample into the preliminary arousal detection model, and obtain a high-level representation sample of each sample by using an output of the intermediate layer of the preliminary arousal detection model as the high-level representation sample of the sample;
the density clustering unit is used for performing density clustering on the high-level representation samples of the samples belonging to the class of arousal categories according to each class of arousal categories to obtain a plurality of high-level representation subsets belonging to the class of arousal categories; obtaining a high-level representation subset belonging to each class of arousal category;
and the second training unit is used for training the classifier by adopting the high-level representation subset belonging to each type of arousal category to obtain the final arousal detection model.
8. The sleep arousal detection apparatus according to claim 7, wherein the second training unit is configured to train a classifier after splicing each frame of data with time domain neighboring data for a high level representation sample in the high level representation subset of each type of arousal category to obtain the final arousal detection model.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the sleep arousal detection method of any one of claims 1 to 5 when executing the computer program.
10. A computer-readable storage medium characterized in that the computer-readable storage medium stores a computer program for executing the sleep arousal detection method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010599319.6A CN113842112A (en) | 2020-06-28 | 2020-06-28 | Sleep arousal detection method and device, computer equipment and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010599319.6A CN113842112A (en) | 2020-06-28 | 2020-06-28 | Sleep arousal detection method and device, computer equipment and readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113842112A true CN113842112A (en) | 2021-12-28 |
Family
ID=78972444
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010599319.6A Pending CN113842112A (en) | 2020-06-28 | 2020-06-28 | Sleep arousal detection method and device, computer equipment and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113842112A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115251939A (en) * | 2022-06-21 | 2022-11-01 | 西南交通大学 | Method, device and equipment for detecting arousal and readable storage medium |
-
2020
- 2020-06-28 CN CN202010599319.6A patent/CN113842112A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115251939A (en) * | 2022-06-21 | 2022-11-01 | 西南交通大学 | Method, device and equipment for detecting arousal and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110141226B (en) | Automatic sleep staging method and device, computer equipment and computer storage medium | |
| KR101395197B1 (en) | Automated detection of sleep and waking states | |
| US5816247A (en) | Monitoring an EEG | |
| Wilson et al. | Spike detection: a review and comparison of algorithms | |
| CN110801221B (en) | Sleep apnea fragment detection equipment based on unsupervised feature learning | |
| Kumari et al. | Seizure detection in EEG using time frequency analysis and SVM | |
| Pearce et al. | Temporal changes of neocortical high-frequency oscillations in epilepsy | |
| US20210267530A1 (en) | Multiclass classification method for the estimation of eeg signal quality | |
| KR20200049930A (en) | The biological signal analysis system and biological signal analysis method for operating by the system | |
| Jiang et al. | Sleep stage classification using covariance features of multi-channel physiological signals on Riemannian manifolds | |
| Agarwal et al. | Digital tools in polysomnography | |
| CN105975900A (en) | Learning scene classification recording method and recording device | |
| CN113842112A (en) | Sleep arousal detection method and device, computer equipment and readable storage medium | |
| Demirel et al. | Single-channel EEG based arousal level estimation using multitaper spectrum estimation at low-power wearable devices | |
| CN114746005A (en) | System and method for specifying REM and wake states | |
| CN116616716A (en) | Integrated learning sleep stage method based on cluster dimension reduction | |
| CN114931362A (en) | Cross-patient epileptic wave detection method and system based on differential matrix and storage medium | |
| RU2751137C1 (en) | Method for determining sleep phase in long-term eeg recording | |
| Schönweiler et al. | Classification of passive auditory event-related potentials using discriminant analysis and self-organizing feature maps | |
| CN113208621A (en) | Dreaming interaction method and system based on EEG signal | |
| Nagpal et al. | Sleep EEG classification using fuzzy logic | |
| Campbell | Mental chronometry. I. Behavioural and physiological techniques | |
| Swapnil et al. | An Ensemble Approach to Classify Mental Stress using EEG Based Time-Frequency and Non-Linear Features | |
| TWI536964B (en) | Eog-based sleep staging method, computer program product with stored programs, computer readable medium with stored programs, and electronic apparatuses | |
| CN118830855A (en) | Multidimensional sleep disorder detection management system |
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 |