CN109833031B - Automatic sleep staging method based on LSTM and utilizing multiple physiological signals - Google Patents
Automatic sleep staging method based on LSTM and utilizing multiple physiological signals Download PDFInfo
- Publication number
- CN109833031B CN109833031B CN201910185683.5A CN201910185683A CN109833031B CN 109833031 B CN109833031 B CN 109833031B CN 201910185683 A CN201910185683 A CN 201910185683A CN 109833031 B CN109833031 B CN 109833031B
- Authority
- CN
- China
- Prior art keywords
- signal
- sequence
- data
- lstm
- sleep
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000007958 sleep Effects 0.000 title claims abstract description 57
- 238000000034 method Methods 0.000 title claims abstract description 44
- 230000001133 acceleration Effects 0.000 claims abstract description 34
- 238000012545 processing Methods 0.000 claims abstract description 34
- 230000029058 respiratory gaseous exchange Effects 0.000 claims abstract description 17
- 230000008667 sleep stage Effects 0.000 claims abstract description 16
- 230000000241 respiratory effect Effects 0.000 claims description 33
- 238000001228 spectrum Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 5
- 238000013145 classification model Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 2
- 230000007547 defect Effects 0.000 abstract description 4
- 238000011160 research Methods 0.000 description 6
- 206010062519 Poor quality sleep Diseases 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000037053 non-rapid eye movement Effects 0.000 description 3
- 230000004461 rapid eye movement Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 208000019116 sleep disease Diseases 0.000 description 2
- 208000020685 sleep-wake disease Diseases 0.000 description 2
- 230000002618 waking effect Effects 0.000 description 2
- 210000001015 abdomen Anatomy 0.000 description 1
- 230000037007 arousal Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 230000002612 cardiopulmonary effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 210000001061 forehead Anatomy 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 210000004072 lung Anatomy 0.000 description 1
- 230000003860 sleep quality Effects 0.000 description 1
- 230000037322 slow-wave sleep Effects 0.000 description 1
Images
Abstract
An automatic sleep staging method based on LSTM and using multiple physiological signals comprises the steps of firstly, acquiring electrocardiosignals, respiration signals and acceleration signals of a tested person; step two, signal processing, step three, extracting features for classification, step four, constructing a model, inputting the manually extracted features into a first-layer long-time memory model, inputting the output probability of the manually extracted features into a second-layer long-time memory model together with the manually extracted features as new features, constructing classifiers for different classification tasks, and step five, using the trained model for classification of sleep stages; the invention adopts various physiological signals including electrocardiosignals, thoracoabdominal respiration signals and head acceleration signals, but does not have electroencephalogram signals, thereby overcoming the defects caused by applying the electroencephalogram to sleep stages; meanwhile, a long-time memory model is adopted, the method is suitable for large-sample and large-data, the time correlation of sleep events is considered, and the accuracy and reliability of sleep staging are improved.
Description
Technical Field
The invention belongs to the technical field of biomedical signal processing, relates to signal processing of electrocardio, respiration, acceleration and the like, and particularly relates to an automatic sleep staging method based on a long-short time memory model (LSTM) and utilizing multiple physiological signals.
Background
Sleep is one of the most important life activities of a human body, and the study on the sleep rule and the sleep structure is helpful for improving the sleep quality of people. Since the 60 s of the 20 th century, sleep medicine has developed into a standard system over several decades, and the most common of them is the sleep staging standard established by the American Academy of Sleep Medicine (AASM) in 2007: based on Polysomnography (PSG), night sleep activity is divided into five different stages, namely, a waking stage (W), a first stage of sleep (N1), a second stage of sleep (N2), a third stage of sleep (N3), and a Rapid Eye Movement stage (REM), according to AASM sleep and related event interpretation manual. The PSG needs to record multi-lead physiological signals such as electroencephalogram, electrooculogram, myoelectricity, Electrocardiogram (ECG), and the like, and then a professional physician with abundant experience interprets the signals every 30s frames to obtain a clinical classification result. In Non-clinical applications, there are also different methods of classification, two classes (W, Sleep) to distinguish arousal from Sleep, three classes (W, NREM, REM) to distinguish wakefulness, Non-rapid eye Movement (NREM) and rapid eye Movement, four classes (W, N1/N2, N3, REM) to distinguish wakefulness, Light Sleep (LS), deep Sleep (Slow Wave Sleep, SWS) and rapid eye Movement. The four-classification approach may be more practical than other classification criteria, taking into account the interrelations and overages between individual sleep stages.
However, the sleep staging method based on the PSG technology has many problems in practical application, which is generally performed in a strict sleep laboratory, but too many signal measurements affect the comfort of sleep, so that the measurement results deviate from the real situation, and the expensive cost also limits the popularization of the method. Therefore, it is of great significance to explore an automatic sleep staging method. At present, the research of the automatic sleep staging method mainly includes feature extraction and pattern recognition of signals, and the method can be divided into three aspects according to used physiological signals: first, sleep staging studies based on electroencephalogram. The research on the aspect is mature, and at present, the accuracy rate of more than 90% can be realized by only utilizing the single-lead brain electrical signal to carry out five classifications on the sleep stage of the healthy people; meanwhile, the five classification of sleep stages of healthy people by combining electroencephalogram, myoelectricity and electrooculogram can realize the accuracy of 92 percent and the accuracy of 86 percent for patients with sleep disorder. Second, sleep staging studies based on cardiopulmonary coupling-related signals. The ECG and respiratory signals are mainly utilized in the aspect, and the highest 71.9% accuracy rate is realized by classifying three patients with sleep disordered breathing by using the cardiac and respiratory signals in Huangwen et al. Thirdly, the study of sleep staging method based on acceleration signal in sleep process mainly classifies wakefulness and sleep, and the consistency with the monitoring result of PSG system can reach 91% at most.
However, the above studies have certain practical problems. Although the electroencephalogram signal based staging method is high in accuracy, electroencephalogram is a weak physiological signal and is easily interfered by various kinds, so that the requirements on electrodes and an acquisition process are strict, and the cost is high. Currently, most of sleep staging methods based on acceleration signals can only be used for distinguishing waking from sleeping, classification results are poor, and actual reference significance is not large. Therefore, the sleep staging method research based on the coupling aspect of the heart and the lung has the most practical significance. The ECG is a physiological signal which is easy to obtain, has larger amplitude and less interference factors, and has huge practical application value. Currently, there are some sleep staging studies that only use ECG signals to extract the Heart Rate Variability (HRV) signal from the ECG signal, and perform sleep staging according to the difference of the HRV signal in different sleep stages. The team of the institute of electrical and electronic engineering of the university of hacelay compares the results of three classification of healthy people by four different classification methods, wherein the highest method achieves an accuracy of 87.11%, but the effect of sleep staging by using HRV is not ideal at present, so that the research on the sleep staging method has great exploration significance.
The human sleep lasts for a certain period of Time, the physiological signal change in the period of Time is a Time sequence, and a Long Short Time Memory (LSTM) based Time sequence has good capability of processing the pattern recognition problem. The LSTM model is additionally provided with a forgetting gate on the basis of the traditional circulating neural network, so that the neural network selectively forgets the learned parameters before, and the problem of long-term dependence is avoided. Yulitita et al used LSTM for sleep stage study, and utilized three signals of electroencephalogram, electrooculogram and myoelectricity to classify patients with sleep disorder in five grades to achieve 86% accuracy, and Radha et al utilized LSTM model to conduct sleep stage study on healthy people to obtain good results.
In summary, the current research methods for automatic sleep staging have respective limitations, and no concise, reliable and efficient method exists. The LSTM-based neural network approach has good classification capability for time series data, but has not been well used in the field of sleep staging research.
Disclosure of Invention
In order to overcome the defects of the existing automatic sleep staging technology, the invention aims to provide an automatic sleep staging method based on LSTM and utilizing multiple physiological signals, which is a universal, easy-to-implement and economic sleep staging method.
In order to achieve the purpose, the invention adopts the technical scheme that:
an automatic sleep staging method using multiple physiological signals based on LSTM, comprising the steps of:
the method comprises the following steps: signal acquisition
The ECG measuring instrument, the respiration signal measuring instrument and the three-axis acceleration sensor are used for measuring to obtain a first-lead ECG signal, a first-lead respiration signal and a third-lead acceleration signal of the measured person.
Step two: signal processing;
for the acquired one-lead ECG signal, oneConducting signal processing on the respiration signal and the three-derivative acceleration signal, wherein the three derivatives are three mutually vertical directions of space x, y and z; processing the ECG signal to obtain the final HRV signal and respiratory wave amplitude signal Rfm(ii) a Processing the respiratory signal to obtain a final RRV signal; processing the acceleration signal to obtain a comprehensive acceleration signal acc (n);
step three: extracting characteristics;
for four signals obtained after the original signal processing: HRV, respiratory wave amplitude signal RfmRRV and the comprehensive acceleration signal acc (n), firstly, segmentation processing is carried out, the sleep data of a whole night is divided into a plurality of data sections with the same length by taking 5 minutes as a window length and 30 seconds as a step length, then, feature extraction is carried out on each section of data, normalization processing is carried out on each feature, and the normalization method is that the feature of a signal acquisition object of the whole night is subtracted by the mean value of the feature sequence, and then, the mean value is divided by the variance of the feature sequence.
Step four: and (5) constructing a model.
The constructed model adopts two layers of LSTMs, and each layer is provided with four different classifiers which are respectively used for realizing the staging tasks of two-classification, three-classification, four-classification and five-classification.
The input sequence of the first layer of LSTM network is the characteristic sequence extracted in the third step, the input characteristic data passes through the first layer of LSTM network, four parallel classifiers are arranged in the input characteristic data, the network structure of each classifier is similar, and the input sequence comprises 5 layers: an input layer, a batch normalization layer, a Bi-LSTM layer, a forgetting layer and a full connection layer.
After the first LSTM layer, each classifier outputs data in a different dimension, representing the probability of each class's stage, which are combined to form a new set of features. And connecting the new set of characteristics in parallel with the characteristics extracted in the second step to serve as the input of a second-layer Bi-LSTM network, wherein four classifiers are also parallelly arranged in the second-layer LSTM network, the structure of each classifier is the same as that of the previous layer, and each classifier is trained to obtain a classification model finally used for prediction.
Step five: predicting sleep staging
And (4) using the four classifiers obtained in the fourth step for predicting the sleep period, performing signal processing and feature extraction on data needing to be predicted, and then sending the extracted features into the classifiers according to the sequence during model training, so as to output the result of sleep staging.
The invention has the advantages that: in order to overcome the defects of the existing automatic sleep staging technology, the invention provides a general, easy-to-realize and economic sleep staging method. Firstly, the physiological signals applied by the invention only comprise ECG, respiration signals and acceleration signals, the three physiological signals are easy to obtain and simple to operate, and the defects of high cost, complexity and the like of electroencephalogram signals are overcome. Secondly, the invention uses the LSTM model, can well utilize the physiological signals in the sleeping process as the correlation before and after the time sequence, and improves the accuracy of the sleep stage. Certainly, the invention has wide application scenes, can be conveniently applied to the fields of monitoring wards, sleeping departments, family sleep and the like, and can be conveniently transplanted into portable equipment to promote the development of mobile medical treatment.
Drawings
Fig. 1 is an overall block diagram of the method.
FIG. 2 is an overall block diagram of the construction of the LSTM model.
Fig. 3 is a network structure of each classifier.
DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION
In order to more clearly illustrate the operation of the present invention, the following detailed description is given with reference to the accompanying drawings and examples.
Referring to fig. 1, an automatic sleep staging method using multiple physiological signals based on LSTM includes the steps of:
the method comprises the following steps: signal acquisition:
the ECG signal is collected by using an ECG measuring instrument, the ECG signal of the II lead is selected, the sampling rate is 100Hz, the chest and abdomen respiration signal is measured by using a respiration signal measuring instrument to be taken as a respiration signal, the sampling rate is 100Hz, the forehead information of the patient is measured by using a three-axis acceleration sensor to be taken as an acceleration signal, and the sampling rate is 100 Hz.
Step two: and (6) signal processing.
The method comprises the following steps of processing signals of a first-lead ECG signal, a first-lead respiration signal and a third-lead acceleration signal which are acquired, wherein the three leads are in three mutually perpendicular directions of space x, y and z, and the signal processing step can be carried out in three aspects:
processing of ECG signals.
First, the raw ECG signal is filtered;
secondly, extracting HRV signals from the ECG signals, identifying the peak point positions of each R wave according to a maximum slope method, obtaining the time interval between two adjacent wave peaks through the difference between the peak values of the adjacent R waves, wherein the obtained time interval sequence is uneven, performing equal interval interpolation on the sequence by utilizing a cubic spline interpolation method to obtain a time sequence under a target sampling frequency, and finally taking the reciprocal of the sequence to obtain the final HRV signal; finally, according to actual requirements, the HRV signals are subjected to down-sampling output;
then, a respiratory wave amplitude signal R is extracted on the basis of HRVfmBy using frequency modulation method, HRV passes through a three-order Butterworth high-pass filter with cut-off frequency of 0.15Hz and a three-order Butterworth low-pass filter with cut-off frequency of 0.5Hz to obtain respiratory wave amplitude signal, denoted as Rfm。
Processing the respiratory signal.
Filtering an original respiratory signal, respectively filtering out 0.01-1 Hz effective components of the respiratory signal by using a low-pass filter with a cut-off frequency of 1Hz and a high-pass filter with a cut-off frequency of 0.01Hz, then extracting RRV signals from the effective respiratory signals, identifying the peak point position of each respiratory wave according to a maximum slope method, obtaining the time interval between two adjacent peaks by using the difference value between the adjacent peak positions, then performing equal interval interpolation on the sequence by using a cubic spline interpolation method to obtain a time sequence under a target sampling frequency, finally taking the reciprocal of the sequence to obtain a final RRV signal, and finally performing RRV down-sampling output according to actual requirements.
And processing the acceleration signal.
To three axesThe acceleration signal is obtained by calculating the arithmetic square root of the sum of squares thereof according to the formula (1) to obtain a signal acctemp(n); for the signal acctemp(n) performing 10-point smoothing filtering to obtain a final comprehensive acceleration signal, which is recorded as acc (n).
Where M is 3, the acceleration signals of the three channels are indicated.
Step three: extracting characteristics;
for four signals obtained after the original signal processing: HRV, respiratory wave amplitude signal RfmRRV and the comprehensive acceleration signal acc (n), firstly, segmentation processing is carried out, the sleep data of a whole night is divided into a plurality of data sections with the same length by taking 5 minutes as a window length and 30 seconds as a step length, then, feature extraction is carried out on each section of data, normalization processing is carried out on each feature, and the normalization method is that the feature of a signal acquisition object of the whole night is subtracted by the mean value of the feature sequence, and then, the mean value is divided by the variance of the feature sequence.
Respiratory wave amplitude signal RfmExtractable features include: dividing the median value of the whole data by the sequence range; dividing the median of the sequence of the interval of the quarter-quartile and the three-quarter-quartile by the sequence range; dividing the mean value of the whole section of data by the square difference; a median of a sequence of respiratory wave peak-to-peak time intervals; the median of a sequence of respiratory wave peak-trough time intervals; the median of a sequence of respiratory wave trough-peak time intervals; dividing the median of the sequence of respiratory wave peak-peak time intervals by the sequence range; dividing the median of the respiratory wave trough-peak time interval sequence by the sequence range; the median of the sequence of breath wave trough-peak time intervals is divided by the sequence range.
For HRV and RRV, extractable features include: mean value of whole data; variance of the whole data; dividing the mean value of the whole section of data by the square difference; the sum of power spectrums of the whole data low frequency band (0.01Hz-0.04 Hz); the sum of power spectrums of frequency bands (0.04Hz-0.15Hz) in the whole data; the sum of power spectrums of the whole data high frequency band (0.15Hz-0.4 Hz); the sum of the total power spectrum of the whole data; the ratio of the sum of the power spectrums of the whole high frequency band (0.15Hz-0.4Hz) to the sum of the power spectrums of the middle frequency band (0.04Hz-0.15 Hz).
For the integrated acceleration signal acc (n), the extractable features include: mean value of whole data; variance of the whole data; dividing the mean value of the whole section of data by the square difference; dividing the median value of the whole data by the sequence range; the median of the sequence of the quarter-and three-quarters interval is divided by the sequence range.
On the basis of 30 features listed above, data segments are still selected with 5 minutes as a window length and 30 seconds as a step length, the extracted feature sequences are arranged in a descending order, values at seven tenths and three tenths of the arranged feature data segments are respectively used as one feature of the segment, and finally, 90 feature parameters in total are obtained.
Step four: and (5) constructing a model.
The model constructed by the invention adopts double-layer LSTM, each layer is provided with four different classifiers for realizing the staging tasks of two-classification, three-classification, four-classification and five-classification, and refer to FIG. 2.
The input sequence of the first layer LSTM network is the 90 feature sequences extracted in step two. The input feature data passes through a first LSTM layer, four parallel classifiers are arranged in the first LSTM layer, the network structure of each classifier is similar, and the first LSTM layer comprises 5 network layers: an input layer, a batch normalization layer, a Bi-LSTM layer, a forgetting layer and a full connection layer.
After the first LSTM layer, each classifier outputs data in a different dimension, representing the probability of each class's stage, which are combined to form 14 new features. Connecting the 14 new features in parallel with the features extracted in the second step as input of a second-layer LSTM network, wherein four classifiers are also parallelly arranged in the second-layer LSTM network, the structure of each classifier is the same as that of the previous layer, and each classifier is trained by a large number of samples to obtain a classification model finally used for prediction;
step five: predicting sleep staging
And (3) using the four classifiers obtained in the fourth step for predicting sleep stages, using corresponding instruments to measure ECG signals, respiration signals and acceleration signals of other tested persons, carrying out signal processing and feature extraction on data, and then sending the extracted features into the classifiers according to the sequence of model training, thus outputting the results of the sleep stages.
Claims (4)
1. An automatic sleep staging method using multiple physiological signals based on LSTM, comprising the steps of:
the method comprises the following steps: signal acquisition
Measuring by using an electrocardio measuring instrument, a respiration signal measuring instrument and a three-axis acceleration sensor to obtain a first-lead ECG signal, a first-lead respiration signal and a third-lead acceleration signal of the measured person;
step two: signal processing;
processing the acquired one-lead ECG signal, one-lead respiration signal and three-axis acceleration signal, wherein the three leads are three mutually perpendicular directions of space x, y and z; processing the ECG signal to obtain the final HRV signal and respiratory wave amplitude signal Rfm(ii) a Processing the respiratory signal to obtain a final RRV signal; processing the triaxial acceleration signal to obtain a comprehensive acceleration signal acc (n);
step three: extracting characteristics;
for four signals obtained after the original signal processing: HRV, respiratory wave amplitude signal RfmRRV, comprehensive acceleration signal acc (n), firstly, carrying out segmentation processing, dividing the sleep data of a whole night into a plurality of data sections with the same length by taking 5 minutes as a window length and taking 30 seconds as a step length, then carrying out feature extraction on each section of data, and carrying out normalization processing on each feature, wherein the normalization method is that the feature of a signal acquisition object of the whole night is subtracted by the mean value of the feature sequence, and then the difference is divided by the variance of the feature sequence;
step four: constructing a model;
the constructed model adopts double-layer LSTM, and each layer is provided with four different classifiers for realizing the staging tasks of two-classification, three-classification, four-classification and five-classification;
the input sequence of the first layer of LSTM network is the characteristic sequence extracted in the third step, the input characteristic data passes through the first layer of LSTM network, four parallel classifiers are arranged in the input characteristic data, the network structure of each classifier is similar, and the input characteristic data comprises 5 network layers: an input layer, a batch normalization layer, a Bi-LSTM layer, a forgetting layer and a full connection layer;
after passing through the first LSTM layer, each classifier outputs data of different dimensions, representing the probability of each class of stage, and combining the probabilities to form a new set of characteristics; connecting the new set of characteristics in parallel with the characteristics extracted in the second step to serve as the input of a second-layer LSTM network, wherein four classifiers are also parallelly arranged in the second-layer LSTM network, the structure of each classifier is the same as that of the previous layer, and each classifier is trained to obtain a classification model finally used for prediction;
step five: predicting sleep staging
And (4) using the four classifiers obtained in the fourth step for predicting the sleep stages, performing signal processing and feature extraction on data needing to be predicted, and then sending the extracted features into the classifiers according to the sequence during model training, so as to output the results of the sleep stages.
2. The LSTM-based method for automated sleep staging using multiple physiological signals according to claim 1, wherein the processing of the respiration signal of step one is specifically:
filtering an original respiratory signal, respectively filtering out 0.01-1 Hz effective components of the respiratory signal by using a low-pass filter with a cut-off frequency of 1Hz and a high-pass filter with a cut-off frequency of 0.01Hz, then extracting RRV signals from the effective respiratory signal, identifying the peak point position of each respiratory wave according to a maximum slope method, obtaining the time interval between two adjacent peaks by using the difference value between the adjacent peak positions, carrying out equal interval interpolation on the sequence by using a cubic spline interpolation method, changing the sequence into a time sequence under a target sampling frequency, finally taking the reciprocal of the sequence to obtain the RRV signals, and finally carrying out down-sampling output on the RRV according to actual requirements.
3. The LSTM-based method for automatic sleep staging using multiple physiological signals according to claim 1, wherein the processing of the acceleration signal in step one includes:
for the triaxial acceleration signal, the arithmetic square root of the square sum of the triaxial acceleration signal is obtained according to the formula (1) to obtain a signal acctemp(n); for the signal acctemp(n) smoothing the 10 points to obtain a comprehensive acceleration signal recorded as acc (n)
Where M is 3, the acceleration signals of the three channels are indicated.
4. The LSTM-based automated sleep staging method using multiple physiological signals according to claim 1,
step three, the respiratory wave amplitude signal RfmExtractable features include: dividing the median value of the whole data by the sequence range; dividing the median of the sequence of the interval of the quarter-quartile and the three-quarter-quartile by the sequence range; dividing the mean value of the whole section of data by the square difference; a median of a sequence of respiratory wave peak-to-peak time intervals; the median of a sequence of respiratory wave peak-trough time intervals; the median of a sequence of respiratory wave trough-peak time intervals; dividing the median of the sequence of respiratory wave peak-peak time intervals by the sequence range; dividing the median of the respiratory wave trough-peak time interval sequence by the sequence range; dividing the median of the respiratory wave trough-peak time interval sequence by the sequence range;
in the third step, the HRV and RRV, the extractable characteristics include: mean value of whole data; variance of the whole data; dividing the mean value of the whole section of data by the square difference; the sum of the power spectrums of the whole data in the low frequency range of 0.01Hz-0.04 Hz; the sum of power spectrums of the frequency bands of 0.04Hz to 0.15Hz in the whole data; the sum of power spectrums of 0.15Hz-0.4Hz in the whole high-frequency range of the data; the sum of the total power spectrum of the whole data; the ratio of the sum of the power spectrums of the whole data in the high frequency range of 0.15Hz to 0.4Hz to the sum of the power spectrums of the whole data in the medium frequency range of 0.04Hz to 0.15 Hz;
the comprehensive acceleration signal acc (n) in the third step, the extractable characteristics include: mean value of whole data; variance of the whole data; dividing the mean value of the whole section of data by the square difference; dividing the median value of the whole data by the sequence range; dividing the median of the sequence of the interval of the quarter-quartile and the three-quarter-quartile by the sequence range;
on the basis of 30 features listed above, data segments are still selected with 5 minutes as a window length and 30 seconds as a step length, the extracted feature sequences are arranged in a descending order, values at seven tenths and three tenths of the arranged feature data segments are respectively used as one feature of the segment, and finally, 90 feature parameters in total are obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185683.5A CN109833031B (en) | 2019-03-12 | 2019-03-12 | Automatic sleep staging method based on LSTM and utilizing multiple physiological signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185683.5A CN109833031B (en) | 2019-03-12 | 2019-03-12 | Automatic sleep staging method based on LSTM and utilizing multiple physiological signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109833031A CN109833031A (en) | 2019-06-04 |
| CN109833031B true CN109833031B (en) | 2020-08-14 |
Family
ID=66885629
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910185683.5A Active CN109833031B (en) | 2019-03-12 | 2019-03-12 | Automatic sleep staging method based on LSTM and utilizing multiple physiological signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109833031B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110432870A (en) * | 2019-08-13 | 2019-11-12 | 重庆邮电大学 | A kind of sleep signal based on 1D CNN-LSTM method by stages automatically |
| CN111134666B (en) * | 2020-01-09 | 2021-05-11 | 中国科学院软件研究所 | Emotion recognition method of multi-channel electroencephalogram data and electronic device |
| CN111436904B (en) * | 2020-03-17 | 2022-11-15 | 深圳赛安特技术服务有限公司 | Intelligent sleep staging method and device and computer readable storage medium |
| CN111493828B (en) * | 2020-04-23 | 2021-11-16 | 四川大学 | Sequence-to-sequence sleep disorder detection method based on full convolution network |
| CN113842111A (en) * | 2020-06-28 | 2021-12-28 | 珠海格力电器股份有限公司 | Sleep staging method and device, computing equipment and storage medium |
| CN112232161B (en) * | 2020-09-30 | 2023-04-07 | 华中科技大学 | Complex motion continuous estimation method based on electromyography mapping model switching |
| CN113925459B (en) * | 2021-09-27 | 2023-05-30 | 西安交通大学 | Sleep stage method based on electroencephalogram feature fusion |
| CN115775630A (en) * | 2023-02-10 | 2023-03-10 | 北京海思瑞格科技有限公司 | Postoperative lung complication probability prediction method based on sleep stage data before operation |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956558A (en) * | 2016-04-26 | 2016-09-21 | 陶大鹏 | Human movement identification method based on three-axis acceleration sensor |
| CN107595243A (en) * | 2017-07-28 | 2018-01-19 | 深圳和而泰智能控制股份有限公司 | A kind of illness appraisal procedure and terminal device |
| US9949714B2 (en) * | 2015-07-29 | 2018-04-24 | Htc Corporation | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection |
| CN108542386A (en) * | 2018-04-23 | 2018-09-18 | 长沙学院 | A kind of sleep state detection method and system based on single channel EEG signal |
| CN109044396A (en) * | 2018-06-25 | 2018-12-21 | 广东工业大学 | A kind of intelligent recognition of heart sound method based on two-way length Memory Neural Networks in short-term |
| KR20190004430A (en) * | 2017-07-04 | 2019-01-14 | 주식회사 알고리고 | Apparatus and method for posture analysis of time series using artificial inteligence |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10783900B2 (en) * | 2014-10-03 | 2020-09-22 | Google Llc | Convolutional, long short-term memory, fully connected deep neural networks |
| US11373672B2 (en) * | 2016-06-14 | 2022-06-28 | The Trustees Of Columbia University In The City Of New York | Systems and methods for speech separation and neural decoding of attentional selection in multi-speaker environments |
| US11222253B2 (en) * | 2016-11-03 | 2022-01-11 | Salesforce.Com, Inc. | Deep neural network model for processing data through multiple linguistic task hierarchies |
-
2019
- 2019-03-12 CN CN201910185683.5A patent/CN109833031B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9949714B2 (en) * | 2015-07-29 | 2018-04-24 | Htc Corporation | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection |
| CN105956558A (en) * | 2016-04-26 | 2016-09-21 | 陶大鹏 | Human movement identification method based on three-axis acceleration sensor |
| KR20190004430A (en) * | 2017-07-04 | 2019-01-14 | 주식회사 알고리고 | Apparatus and method for posture analysis of time series using artificial inteligence |
| CN107595243A (en) * | 2017-07-28 | 2018-01-19 | 深圳和而泰智能控制股份有限公司 | A kind of illness appraisal procedure and terminal device |
| CN108542386A (en) * | 2018-04-23 | 2018-09-18 | 长沙学院 | A kind of sleep state detection method and system based on single channel EEG signal |
| CN109044396A (en) * | 2018-06-25 | 2018-12-21 | 广东工业大学 | A kind of intelligent recognition of heart sound method based on two-way length Memory Neural Networks in short-term |
Non-Patent Citations (6)
| Title |
|---|
| 《基于CNN-LSTM网络的睡眠分期研究》;张秀丽，夏斌;《微型机与应用》;20170913;第36卷;全文 * |
| 《基于周期分割的睡眠自动分期研究》;李同庆，邹俊忠，张见，王蓓，魏作臣;《计算机工程与应用》;20180629;全文 * |
| 《基于心电与呼吸信号的睡眠分期算法研究》;黄文汉，张伟，胡立刚，周陈旺，向丹阳;《智能计算机与应用》;20180228;第8卷;全文 * |
| 《心电信号及各组分的频率分布和有效带宽研究》;何伟，陈良迟，徐晓红，谢正祥;《生物医学工程学杂志》;19961225;全文 * |
| Instantaneous Heart Rate as a Robust Feature for Sleep Apnea;Rahul K. Pathinarupothi,Vinaykumar R., Ekanath Rangan;《2017 IEEE EMBS International Conference on Biomedical & Health Information》;20170413;全文 * |
| Sleep Stage Classification Based on Multi-level Feature Learning and Recurrent Neural Networks via Wearable Device;Zhang X , Kou W , Chang I C , et al.;《Computers in Biology and Medicine》;20181201;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109833031A (en) | 2019-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109833031B (en) | Automatic sleep staging method based on LSTM and utilizing multiple physiological signals | |
| Kortelainen et al. | Sleep staging based on signals acquired through bed sensor | |
| Zhang et al. | Sleep stage classification using bidirectional lstm in wearable multi-sensor systems | |
| Kuo et al. | Automatic stage scoring of single-channel sleep EEG based on multiscale permutation entropy | |
| CN109770892A (en) | A kind of sleep stage method based on electrocardiosignal | |
| Almutairi et al. | Detection of obstructive sleep apnoea by ecg signals using deep learning architectures | |
| Prabha et al. | Automated system for obstructive sleep apnea detection using heart rate variability and respiratory rate variability | |
| CN114781465B (en) | rPPG-based non-contact fatigue detection system and method | |
| Cai et al. | Application of electroencephalography-based machine learning in emotion recognition: A review | |
| CN111407262A (en) | Sleep staging method based on B L STM and device for sleep staging based on B L STM | |
| CN106343992B (en) | Heart rate variance analyzing method, device and purposes | |
| Dehkordi et al. | Sleep stage classification in children using photoplethysmogram pulse rate variability | |
| Zhang et al. | Rass: A portable real-time automatic sleep scoring system | |
| TWI288875B (en) | Multiple long term auto-processing system and method thereof | |
| Bahador et al. | Multimodal spatio-temporal-spectral fusion for deep learning applications in physiological time series processing: A case study in monitoring the depth of anesthesia | |
| Jiang et al. | A multi-scale parallel convolutional neural network for automatic sleep apnea detection using single-channel EEG signals | |
| CN114648040A (en) | Method for extracting and fusing multiple physiological signals of vital signs | |
| AU2020102907A4 (en) | Novel automated machine learning algorithms based system for sleep staging features analysis | |
| Saran et al. | Validation of dozee, a ballistocardiography-based device, for contactless and continuous heart rate and respiratory rate measurement | |
| Samavati et al. | Efficient deep learning-based estimation of the vital signs on smartphones | |
| CN115316957A (en) | Perioperative anesthesia depth monitoring system based on multi-parameter indexes | |
| Dehkordi et al. | Sleep/wake classification using cardiorespiratory features extracted from photoplethysmogram | |
| CN114903445A (en) | Intelligent monitoring and early warning system for cardiovascular and cerebrovascular diseases | |
| Bianchi et al. | Automatic detection of sleep macrostructure based on a sensorized T-shirt | |
| Shanmugham et al. | Sleep apnea detection using ResNet |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231227 Address after: Room 705, 7th Floor, Building A3, Nanshan Zhiyuan, No. 1001 Xueyuan Avenue, Xili Street, Nanshan District, Shenzhen City, Guangdong Province, 518055 Patentee after: Shenzhen Ruixinyu Technology Co.,Ltd. Address before: Beilin District Xianning West Road 710049, Shaanxi city of Xi'an province No. 28 Patentee before: XI'AN JIAOTONG University |