CN116509335A - Gradual evolution automatic sleep stage-dividing method - Google Patents
Gradual evolution automatic sleep stage-dividing method Download PDFInfo
- Publication number
- CN116509335A CN116509335A CN202310690872.4A CN202310690872A CN116509335A CN 116509335 A CN116509335 A CN 116509335A CN 202310690872 A CN202310690872 A CN 202310690872A CN 116509335 A CN116509335 A CN 116509335A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- sleep stage
- sleep
- training
- 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
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000007958 sleep Effects 0.000 title claims abstract description 30
- 230000008667 sleep stage Effects 0.000 claims abstract description 50
- 238000011156 evaluation Methods 0.000 claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 36
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 238000012795 verification Methods 0.000 claims abstract description 19
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 12
- 238000012216 screening Methods 0.000 claims description 11
- 239000012634 fragment Substances 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 description 5
- 230000036541 health Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 208000019116 sleep disease Diseases 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000005189 cardiac health Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 230000003860 sleep quality Effects 0.000 description 2
- 230000036962 time dependent Effects 0.000 description 2
- 206010062519 Poor quality sleep Diseases 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 230000005056 memory consolidation Effects 0.000 description 1
- 230000004630 mental health Effects 0.000 description 1
- 230000006996 mental state Effects 0.000 description 1
- 230000008452 non REM sleep Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000036385 rapid eye movement (rem) sleep Effects 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 208000020685 sleep-wake disease Diseases 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4812—Detecting sleep stages or cycles
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
A gradual evolving automatic sleep staging method comprising the steps of: step one, constructing a plurality of sleep stage models based on a plurality of feature extraction methods; training a model by using a tag data set, and calculating an evaluation index of the model on a verification set; step three, the sleep stage model automatically marks the unlabeled dataset, and a credibility sample is screened based on a credibility evaluation method; step four, forming a training set by the label data and the trusted samples screened in the step three, and repeating the step two and the step three; step five, testing the sleep stage model by a test set, and calculating a gradual evolution result of the model; the invention can continuously improve the performance of the automatic sleep stage model under the conditions of a small number of labeled data sets and a large number of unlabeled data sets, and effectively relieve the difficulty of medical data shortage and complex labeled data.
Description
Technical Field
The invention relates to the technical field of sleep monitoring, in particular to a gradual evolution automatic sleep staging method.
Background
Sleep is an essential process for life, is an important link for organism restoration, integration and memory consolidation, and is an indispensable component for health. Poor sleep quality for a long period of time can affect the mental state and health quality of people, and serious diseases of other bodies can be induced. Investigation by the world health organization shows that 27% of people have sleep problems, and sleep disorders have become a prominent problem threatening the health of the world public. Sleep staging is of great significance for sleep quality analysis and sleep disorder diagnosis.
Clinically, sleep sessions were interpreted as "gold standard" by Polysomnography (PSG) and expert. PSG refers to the simultaneous acquisition of multichannel physiological signals such as Electrocardiograph (ECG), electroencephalogram (EEG), electrooculogram (EOG), myoelectricity (EMG), respiration, etc. during sleep of a patient. The clinician divides night sleep into awake, non-rapid eye movement sleep and rapid eye movement sleep, which may be further divided into phase i, phase ii, phase iii, in units of 30 s. However, PSG is costly, and patients need to wear a large number of sensors in a professional sleep laboratory during monitoring, and monitoring data needs to be manually analyzed, which severely restricts the popularization of sleep monitoring.
In recent years, with the rise of the concept of great health and the wide application of artificial intelligence technology, more and more researchers are focusing on the research and application of artificial intelligence technology in the field of sleep monitoring. Moreover, massive sleep data provides a research basis for an automatic sleep stage algorithm.
Semi-supervised learning (Semi-Supervised Learning, SSL) is a key problem in research in the field of machine learning, and is a learning method combining supervised learning and unsupervised learning. Semi-supervised learning uses a training model of a small amount of tagged data and a large amount of untagged data, and abstract information of the untagged data can improve the overfitting phenomenon caused by the small amount of tagged data. Self-Training (Self-Training) is a common semi-supervised learning method, and the principle is that the prediction result of a model is marked as a real label, so that a Training set is expanded and used for supervised learning. Training data is gradually increased by successive iterations to improve model performance.
The automatic sleep stage algorithm training classifier usually adopts a supervised training strategy, and the classification precision is positively correlated with the scale of the training set. However, due to the specificity of medical data and the complexity of manual labeling, available labeling data is very scarce. Furthermore, classification accuracy and generalization ability of the stage model trained using a limited number of marker data are hardly ensured. For example: the Chinese patent invention 201710002025.9 discloses an automatic sleep stage method of single-lead electroencephalogram, but the scheme has the defect that model training only depends on initial small amount of data, the obtained model cannot adapt to mass data in practical application, and the accuracy and generalization capability are required to be improved.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a gradual evolution automatic sleep stage method, which aims to gradually improve the classification accuracy and generalization capability of a sleep stage model by using limited marked data and a large amount of unmarked data.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a gradual evolving automatic sleep staging method comprising the steps of:
step one: constructing a plurality of sleep stage models based on a plurality of feature extraction methods;
step two: training a sleep stage model by using a labeled data set, and calculating an evaluation index of the sleep stage model on a verification set;
step three: automatically marking a label-free data set by the sleep stage model, and screening a credible sample based on a credibility evaluation method;
step four: the label data and the trusted samples screened in the step III form a training set, and the step II and the step III are repeated;
step five: the test set tests the sleep stage model, and the gradual evolution result of the model is calculated.
The first step is specifically as follows:
selecting multimode physiological signals or single physiological signals as the original data of the sleep stage task according to the actual application scene;
when multiple physiological signals are used, according to the characteristics of different physiological signals, the local features are automatically extracted by adopting a convolutional neural network, the features are extracted by adopting a one-dimensional convolutional network for one-dimensional data, and the features are extracted by adopting a two-dimensional convolutional network for two-dimensional data, so that multiple groups of features are obtained;
when a single physiological signal is used, as the single physiological signal has different characteristics of a time domain and a frequency domain, the characteristic can be automatically extracted by adopting a convolutional neural network for the original signal, and the characteristic can be automatically extracted by adopting the convolutional neural network for the data converted from the original signal to the frequency domain, so that the two groups of characteristics of the time domain and the frequency domain are ensured to have distinction;
and the obtained multiple groups of characteristics respectively adopt a classifier to realize classification tasks, so that the construction of multiple sleep stage models is completed.
The second step is specifically as follows:
the data set is divided into a labeled data set, an unlabeled data set, a verification set and a test set according to a proportion, wherein the unlabeled data set is more than 50% of the total number of the data sets, and the verification set and the test set are always unchanged in the model evolution process and are not overlapped with other data sets;
dividing the labeled data set and the verification set into data fragments according to a fixed length, namely sleep fragments in a sleep stage task, and inputting the sleep fragments into a training model of the sleep stage model constructed in the step one;
in the training process, a label data set is used for fitting a model, a verification set is used for verifying the performance of the model after each iteration, and calculated evaluation indexes comprise Accuracy (Accuracy), an F1 fractional average value (MF 1) and Cohen's Kappa coefficient; continuously adjusting super parameters, wherein the super parameters comprise learning rate, batch size and maximum training times; and finally, taking the model with the best calculated evaluation index as an optimal model.
The third step is specifically as follows:
automatically marking the label-free data set by using the optimal model obtained in the second step, and screening a credible sample based on a credibility evaluation method;
the credibility evaluation method is divided into a data credibility evaluation method and a model credibility evaluation method according to different use objects; the data reliability evaluation method is used for judging whether the original data is reliable or not, the data reliability evaluation of EEG signals is used for judging whether the data acquired under the conditions of signal acquisition failure caused by electrode falling, strong interference inundation of real signals and the like exists or not, the model reliability evaluation is used for judging whether the prediction result of a model is reliable or not, and the reliability evaluation of sleep stage models is used for evaluating the output probability distribution of a single model and the classification consistency among a plurality of models.
The fourth step is specifically as follows:
forming a training set by the trusted sample obtained through the screening in the step three and the labeled data used in the step two, continuously training the sleep stage model, and repeating the step two and the step three for a plurality of times until the model evolution termination condition is met;
the model evolution termination condition is determined according to the data quantity of the trusted samples and the evaluation index of the verification set, and usually, the evolution is terminated or the iteration number is customized under the condition that the screened trusted samples are very few.
The fifth step is specifically as follows:
and gradually improving the model performance by screening the trusted samples for multiple times and updating the training set, and testing the model final classification accuracy by using the testing set.
Compared with the prior art, the invention has the advantages that:
1. the method uses a semi-supervised training method, namely only limited or even a small amount of manual marking data is needed for the first training in the step two, and the label-free data is completely used for the subsequent model evolution, so that the difficulties of medical data shortage and complicated marking data are effectively relieved, and the data cost for developing a deep learning model is reduced.
2. In the model evolution process, namely in the step three, the unlabeled data are used in a large amount to continuously enrich the training set, so that the data diversity is increased, and the model classification accuracy and generalization capability are improved.
In summary, the invention provides a gradually evolving automatic sleep stage method, which is suitable for the sleep stage model evolution of various physiological signals and classifiers. The feature extraction method, the deep learning network and the credibility evaluation method can be determined according to actual application scenes and specific physiological signals.
Drawings
FIG. 1 is a flow chart of a gradual evolution automatic sleep staging method of the present invention.
Fig. 2 is a diagram illustrating a sleep stage model according to an embodiment of the present invention, wherein (a) in fig. 2 is a RAW-BiLSTM model, and (B) in fig. 2 is a CWT-TCN model.
FIG. 3 is a model evolution framework of an embodiment of the present invention.
FIG. 4 shows the model evolution results according to the embodiment of the present invention, wherein A in FIG. 4 is the RAW-BiLSTM model evolution results, and (B) in FIG. 4 is the CWT-TCN model evolution results.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings.
Referring to fig. 1, a gradual evolving automatic sleep staging method includes the steps of:
step one: based on a plurality of feature extraction methods, a plurality of sleep stage models are constructed.
The first step is specifically as follows:
selecting multimode physiological signals or single physiological signals as the original data of the sleep stage task according to the actual application scene;
when multiple physiological signals are used, according to the characteristics of different physiological signals, the local characteristics are automatically extracted by adopting a convolutional neural network, the characteristics of One-dimensional data are extracted by adopting a One-dimensional convolutional network (One-dimensional Deep Convolutional Neural Network, 1-DCNN), the characteristics of Two-dimensional data are extracted by adopting a Two-dimensional convolutional network (Two-dimensional Deep ConvolutionalNeuralNetwork, 2-DCNN), and a plurality of groups of characteristics are obtained;
when a single physiological signal is used, the single physiological signal has different characteristics of a time domain and a frequency domain, so that the characteristic can be automatically extracted by adopting a convolutional neural network for an original signal, and the characteristic can be automatically extracted by adopting the convolutional neural network for data converted from the original signal to the frequency domain, thereby ensuring that the two groups of characteristics of the time domain and the frequency domain have distinction.
For example: EEG has both time domain and frequency domain features. Thus, on the one hand, the local features Σf can be extracted directly from the original EEG using the 1-DCNN 1 The method comprises the steps of carrying out a first treatment on the surface of the Alternatively, the original EEG may be time-frequency analyzed in advance using a Continuous wavelet transform (Continuous WaveletTransform, CWT), and then the local features Σf may be extracted from the wavelet map using a 2-DCNN 2 。
The physiological signal is a time sequence signal, and the signal characteristics have a certain time dependency relationship. Therefore, after the above convolution network automatically extracts local features, a time convolution network needs to be used to further acquire global features. For example: features sigma f 1 A two-way long and short Term Memory network (Bi-directional Long Short-Term Memory, biLSTM) may be used, biLSTM being adapted to capture the time dependence between EEG segments. Features sigma f 2 A time series convolutional neural network (Temporal Convolutional Network, TCN) may be used, the TCN being adapted to capture long-term dependencies of the time series.
The multiple groups of characteristics obtained by the method respectively adopt the classifier to realize classification tasks, and a softmax layer is generally adopted as the classifier in deep learning, so that the construction of multiple sleep stage models is completed.
Specific: two sleep stage models are constructed according to EEG signal characteristics. Referring to fig. 2 (a), the RAW-BiLSTM model is composed of a 1-DCNN module for extracting local features of the original EEG signal and a BiLSTM module for extracting time-dependent relationships between EEG segments. Referring to fig. 2 (B), the CWT-TCN model is composed of a 2-DCNN module and a TCN module, where the original EEG signal is first CWT transformed to obtain a wavelet map, and then the 2-DCNN module is used to extract local features, and the TCN module extracts time-dependent relationships between EEG segments.
Step two: and training the sleep stage model by using the labeled data set, and calculating the evaluation index of the sleep stage model on the verification set.
The data set is proportionally divided into a labeled data set, an unlabeled data set, a verification set and a test set, wherein the unlabeled data set is more than 50% of the total number of the data sets, and the verification set and the test set are always unchanged in the model evolution process and are not overlapped with other data sets.
The length of time of the data segments typically used for sleep stages is fixed. Therefore, the labeled dataset and the validation set are divided into data segments according to a fixed length, which are called sleep segments in the sleep stage task, and then the sleep segments are input into the sleep stage model constructed in the step one to train the model.
In the training process, a label data set is used for fitting a model, a verification set is used for verifying the performance of the model after each iteration, and calculated evaluation indexes comprise Accuracy (Accuracy), an F1 fractional average value (MF 1) and Cohen's Kappa coefficient; continuously adjusting super parameters, wherein the super parameters comprise learning rate, batch size and maximum training times; and finally, taking the model with the best calculated evaluation index as an optimal model.
Specific: the dataset is the C3/A2 channel EEG signals of the sleep heart health study (Sleep Heart Health Study, SHHS) phase 1 database. 5793 samples were taken according to 1:7:1: the 1 scale is divided into a labeled data set, an unlabeled data set, a validation set and a test set. The EEG signal divides sleep fragments for a length of 30 s. The label data set trains the sleep stage model, the verification set verifies the performance of the model, and evaluation indexes such as accuracy, MF1 and Kappa coefficients are calculated.
Step three: and step two, the obtained two sleep stage models automatically mark the unlabeled data set. Then, the tag data is subjected to reliability analysis using a reliability evaluation method.
The reliability evaluation method is classified into a data reliability evaluation method and a model reliability evaluation method according to the object of use. The data credibility evaluation method is used for judging whether the original data is credible or not, and the data credibility evaluation of the EEG signals is used for judging whether the data acquired under the conditions of signal acquisition failure caused by electrode falling, strong interference inundation of real signals and the like exist or not, and the unreliable data can seriously influence the training effect of the model. The model credibility evaluation refers to judging whether a prediction result of a model is credible or not, and the sleep stage model credibility evaluation refers to evaluation of single model output probability distribution and evaluation of classification consistency among a plurality of models.
The data credibility evaluation is to obtain a low-interference or non-interference signal by carrying out threshold screening on the time-frequency characteristics of the EEG signal. The single model confidence rating is to output a probability distribution using a predictive entropy metric model. The lower the prediction entropy, the more reliable the model classification result. The multiple model confidence scores are multiple output consistency metrics using Kappa measures. The higher the Kappa measure, the higher the consistency of the multiple model classification results. And screening out the credible samples by the three credibility evaluation methods.
Step four: referring to fig. 3, the training set is updated, namely: the trusted sample and the labeled data form a new training set to continue training the model, and the second step and the third step are repeated until the model evolution termination condition is met. The method is characterized in that the number of the screened trusted samples is less than 10 or the performance of the model on a verification set is not improved to be a model evolution termination condition in three iterations.
Referring to fig. 4, in the gradual evolution process, the accuracy, MF1, kappa coefficients gradually increase, indicating that the performance of the sleep stage model on the validation set gradually increases.
Step five: and testing the final sleep stage model by using a test set, and calculating the gradual evolution result of the model.
And gradually improving the model performance by screening the trusted samples for multiple times and updating the training set, and finally testing the model by using the testing set to finally classify the accuracy. The five classification accuracy of the RAW-BiLSTM model was 75.9%, the MF1 was 66.17, and the kappa coefficient was 0.67. The five classification accuracy of the CWT-TCN model is 80.28%, MF1 is 69.01, and kappa coefficient is 0.72.
Claims (6)
1. A gradual evolving automatic sleep staging method, comprising the steps of:
step one: constructing a plurality of sleep stage models based on a plurality of feature extraction methods;
step two: training a sleep stage model by using a labeled data set, and calculating an evaluation index of the sleep stage model on a verification set;
step three: automatically marking a label-free data set by the sleep stage model, and screening a credible sample based on a credibility evaluation method;
step four: the label data and the trusted samples screened in the step III form a training set, and the step II and the step III are repeated;
step five: the test set tests the sleep stage model, and the gradual evolution result of the model is calculated.
2. The method for gradually evolving automatic sleep stages according to claim 1, wherein the first step is specifically:
selecting multimode physiological signals or single physiological signals as the original data of the sleep stage task according to the actual application scene;
when multiple physiological signals are used, according to the characteristics of different physiological signals, the local features are automatically extracted by adopting a convolutional neural network, the features are extracted by adopting a one-dimensional convolutional network for one-dimensional data, and the features are extracted by adopting a two-dimensional convolutional network for two-dimensional data, so that multiple groups of features are obtained;
when a single physiological signal is used, as the single physiological signal has different characteristics of a time domain and a frequency domain, the characteristic can be automatically extracted by adopting a convolutional neural network for the original signal, and the characteristic can be automatically extracted by adopting the convolutional neural network for the data converted from the original signal to the frequency domain, so that the two groups of characteristics of the time domain and the frequency domain are ensured to have distinction;
and the obtained multiple groups of characteristics respectively adopt a classifier to realize classification tasks, so that the construction of multiple sleep stage models is completed.
3. The method for gradually evolving automatic sleep stages according to claim 1, wherein the first step is specifically:
the data set is divided into a labeled data set, an unlabeled data set, a verification set and a test set according to a proportion, wherein the unlabeled data set is more than 50% of the total number of the data sets, and the verification set and the test set are always unchanged in the model evolution process and are not overlapped with other data sets;
dividing the labeled data set and the verification set into data fragments according to a fixed length, namely sleep fragments in a sleep stage task, and inputting the sleep fragments into a training model of the sleep stage model constructed in the step one;
in the training process, a label data set is used for fitting a model, a verification set is used for verifying the performance of the model after each iteration, and calculated evaluation indexes comprise Accuracy (Accuracy), an F1 fractional average value (MF 1) and a Cohen's kappa coefficient; continuously adjusting super parameters, wherein the super parameters comprise learning rate, batch size and maximum training times; and finally, taking the model with the best calculated evaluation index as an optimal model.
4. The gradual evolving automatic sleep staging method according to claim 1, wherein the step three is specifically:
automatically marking the label-free data set by using the optimal model obtained in the second step, and screening a credible sample based on a credibility evaluation method;
the credibility evaluation method is divided into a data credibility evaluation method and a model credibility evaluation method according to different use objects; the data reliability evaluation method is used for judging whether the original data is reliable or not, the data reliability evaluation of EEG signals is used for judging whether the data acquired under the conditions of signal acquisition failure caused by electrode falling, strong interference inundation of real signals and the like exists or not, the model reliability evaluation is used for judging whether the prediction result of a model is reliable or not, and the reliability evaluation of sleep stage models is used for evaluating the output probability distribution of a single model and the classification consistency among a plurality of models.
5. The gradual evolving automatic sleep staging method according to claim 1, wherein the fourth step is specifically:
forming a training set by the trusted sample obtained through the screening in the step three and the labeled data used in the step two, continuously training the sleep stage model, and repeating the step two and the step three for a plurality of times until the model evolution termination condition is met;
the model evolution termination condition is determined according to the data quantity of the trusted samples and the evaluation index of the verification set, and usually, the evolution is terminated or the iteration number is customized under the condition that the screened trusted samples are very few.
6. The gradual evolution automatic sleep staging method according to claim 1, characterized in that the fifth step is specifically:
and gradually improving the model performance by screening the trusted samples for multiple times and updating the training set, and testing the model final classification accuracy by using the testing set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310690872.4A CN116509335A (en) | 2023-06-12 | 2023-06-12 | Gradual evolution automatic sleep stage-dividing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310690872.4A CN116509335A (en) | 2023-06-12 | 2023-06-12 | Gradual evolution automatic sleep stage-dividing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116509335A true CN116509335A (en) | 2023-08-01 |
Family
ID=87399602
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310690872.4A Pending CN116509335A (en) | 2023-06-12 | 2023-06-12 | Gradual evolution automatic sleep stage-dividing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116509335A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118097669A (en) * | 2024-04-23 | 2024-05-28 | 成都大学 | Remote sensing image automatic labeling method based on multi-model coupling evaluation |
-
2023
- 2023-06-12 CN CN202310690872.4A patent/CN116509335A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118097669A (en) * | 2024-04-23 | 2024-05-28 | 成都大学 | Remote sensing image automatic labeling method based on multi-model coupling evaluation |
| CN118097669B (en) * | 2024-04-23 | 2024-06-21 | 成都大学 | Remote sensing image automatic labeling method based on multi-model coupling evaluation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107495962B (en) | Sleep automatic staging method for single-lead electroencephalogram | |
| Zhao et al. | SleepContextNet: A temporal context network for automatic sleep staging based single-channel EEG | |
| CN110246577B (en) | Method for assisting gestational diabetes genetic risk prediction based on artificial intelligence | |
| CN108492877B (en) | Cardiovascular disease auxiliary prediction method based on DS evidence theory | |
| CN113768519B (en) | Method for analyzing consciousness level of patient based on deep learning and resting state electroencephalogram data | |
| CN111248859A (en) | Automatic sleep apnea detection method based on convolutional neural network | |
| CN109009102A (en) | A kind of aided diagnosis method and system based on electroencephalogram deep learning | |
| CN111685774B (en) | OSAHS Diagnosis Method Based on Probability Integrated Regression Model | |
| CN115530847A (en) | Electroencephalogram signal automatic sleep staging method based on multi-scale attention | |
| Fang et al. | A dual-stream deep neural network integrated with adaptive boosting for sleep staging | |
| CN114841216B (en) | Electroencephalogram signal classification method based on model uncertainty learning | |
| CN113593708A (en) | Sepsis prognosis prediction method based on integrated learning algorithm | |
| CN116509335A (en) | Gradual evolution automatic sleep stage-dividing method | |
| CN112932501A (en) | Method for automatically identifying insomnia based on one-dimensional convolutional neural network | |
| CN115500843A (en) | Sleep stage staging method based on zero sample learning and contrast learning | |
| CN116269212A (en) | Multi-mode sleep stage prediction method based on deep learning | |
| CN113576472B (en) | Blood oxygen signal segmentation method based on full convolution neural network | |
| CN113796830B (en) | Automatic evaluation method for sleep signal stage credibility | |
| CN114366038B (en) | Sleep signal automatic staging method based on improved deep learning algorithm model | |
| CN116473514B (en) | Parkinson disease detection method based on plantar pressure self-adaptive directed space-time graph neural network | |
| CN117064333A (en) | Primary screening device for obstructive sleep apnea hypopnea syndrome | |
| CN113808735B (en) | Mental disease assessment method based on brain image | |
| CN115399735A (en) | Multi-head attention mechanism sleep staging method based on time-frequency double-current enhancement | |
| CN114129138A (en) | Automatic sleep staging method based on time sequence multi-scale mixed attention model | |
| Lv et al. | Ssleepnet: a structured sleep network for sleep staging based on sleep apnea severity |
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 |