CN116671894A

CN116671894A - Method and device for detecting apnea and hypopnea syndrome and computer equipment

Info

Publication number: CN116671894A
Application number: CN202310782792.1A
Authority: CN
Inventors: 潘俊; 谭泉
Original assignee: Wonly Security And Protection Technology Co ltd
Current assignee: Wonly Security And Protection Technology Co ltd
Priority date: 2023-06-28
Filing date: 2023-06-28
Publication date: 2023-09-01

Abstract

The invention relates to the technical field of respiratory state monitoring, and discloses an apnea and hypopnea syndrome detection method, an apnea and hypopnea syndrome detection device and computer equipment, wherein an electroencephalogram signal, an abdomen displacement signal and an oronasal airflow respiratory signal are obtained; sleep staging is carried out based on the electroencephalogram signals, and sleep state information is determined; screening and preprocessing wavelet transformation based on the abdomen displacement signals to extract a feature vector space; and converting to a sleep state space according to the oral-nasal airflow respiratory signal, the sleep state information and a preset neural network model, and carrying out SVM statistics and respiratory state classification.

Description

Method and device for detecting apnea and hypopnea syndrome and computer equipment

Technical Field

The invention relates to the technical field of respiratory state monitoring, in particular to an apnea and hypopnea syndrome detection method, an apnea and hypopnea syndrome detection device and computer equipment.

Background

SAHS hazard is great and not readily noticeable for apnea and hypopnea syndrome.

The conventional sleep polysomnography PSG has the defect of high cost. The present invention aims to monitor abdominal displacement signal data by low cost uwb, evaluate the current respiratory state as: in particular normal breathing, hypopnea, obstructive apnea, mixed apneas.

Disclosure of Invention

In view of the above, the present invention provides a method, apparatus and computer device for detecting apnea and hypopnea syndrome, so as to solve the safety problem caused by the passenger's lack of observation of the surrounding environment.

In a first aspect, the present invention provides a method for detecting apnea and hypopnea syndrome, the method comprising:

acquiring an electroencephalogram signal, an abdomen displacement signal and an oronasal airflow respiration signal;

sleep staging is carried out based on the electroencephalogram signals, and sleep state information is determined;

extracting a feature vector space based on abdomen displacement signal screening and wavelet transformation preprocessing;

and converting into a sleep state space according to the mouth-nose airflow respiratory signals, the sleep state information and the preset neural network model, and carrying out SVM statistics and respiratory state classification.

In a second aspect, the present invention provides an apnea and hypopnea syndrome detection device, the device comprising:

the acquisition signal module is used for acquiring brain electrical signals, abdomen displacement signals and mouth-nose airflow respiration signals;

the determining state module is used for carrying out sleep stage based on the electroencephalogram signals and determining sleep state information;

the feature extraction module is used for extracting a feature vector space based on abdomen displacement signal screening and wavelet transformation preprocessing;

the statistics module is used for converting into a sleep state space according to the oral-nasal airflow breathing signals, the sleep state information and the preset neural network model, and carrying out SVM statistics and breathing state classification.

In a third aspect, the present invention provides a computer device comprising: the device comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so that the method for detecting the apnea and hypopnea syndrome according to the first aspect is executed.

In a fourth aspect, the present invention provides a computer readable storage medium having stored thereon computer instructions for causing a computer to perform the above-described apnea and hypopnea syndrome detection method of the first aspect.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIGS. 4a, 4b and 4c are schematic diagrams of methods for detecting apnea and hypopnea syndrome according to embodiments of the present invention;

FIG. 5 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention;

fig. 10 is a block diagram of an apnea and hypopnea syndrome detection device according to an embodiment of the present invention;

fig. 11 is a schematic diagram of a hardware structure of a computer device according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

According to an embodiment of the present invention, there is provided an embodiment of an apnea and hypopnea syndrome detection method, it is to be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system, such as a set of computer executable instructions, and that although a logical sequence is illustrated in the flowchart, in some cases, the steps illustrated or described may be performed in a different order than that illustrated herein.

The method comprises the steps of carrying out sleep stage on the basis of electroencephalogram signals, carrying out data screening and wavelet transformation pretreatment on the basis of abdomen displacement signals to extract a feature vector space, and carrying out SVM statistics and respiratory state classification by converting (Long-Short Term Memory-Convolutional Neural Network, LSTM-CNN) into a sleep state space by assisting with respiratory filtration of oral-nasal airflow.

SAHS hazard is great and not readily noticeable for apnea and hypopnea syndrome. The conventional sleep polysomnography PSG has the defect of high cost. The present invention aims to monitor abdominal displacement signal data by low cost uwb, evaluate the current respiratory state as: in particular normal breathing, hypopnea, obstructive apnea, mixed apneas.

And the experiment combines the brain electrical signal sleep state information to remove the data of the wake-up period, namely, the data of the sleep period is reserved. And then training an LSTM-CNN model according to a ten-fold cross validation method, and carrying out four-component classification on the respiratory state based on 60-second data fragments, wherein the average accuracy is 83.94%. And the comparison of clinical diagnosis results and KAPPA coefficients is 0.8, and the average accuracy, sensitivity, specificity and precision of detection based on the severity of the AHI index illness respectively reach 90.5%, 95.2%, 96.4% and 96.3%.

The prevalence of sleep apnea and hypopnea syndrome was investigated in an obese adult population, showing a male incidence of between 42% -48% and a female incidence of between 8% -38%, data confirm that the incidence is closely related to patient gender, and the investigation showed that the incidence is also gradually rising with increasing patient age. According to PubMed and Embase databases, it was shown that about 9.36 million people had varying degrees of SAHS and about 4.25 million moderately severe SAHS patients in the population between 30-69 years worldwide by 8 months in 2019 based on diagnostic criteria promulgated by the american sleep society (American Academy of Sleep Medicine, AASM) 2012. The country with the greatest number of SAHS lesions is China, the second world of the disease is the United states, followed by Brazil and India in turn. The number of patients with SAHS in China is about 1.76 hundred million, the incidence rate is up to 23.6%, the number of patients with moderate to severe SAHS is about 6552 ten thousand, the incidence rate is up to 8.8%, and the data prove that the incidence rate has a tendency to rise year by year [8,9]. SAHS is serious in hazard and high in morbidity, so that the disease is found as early as possible, the disease type is diagnosed, and a special treatment plan is formulated according to the self condition and the disease degree of a patient.

In this embodiment, a method for detecting apnea and hypopnea syndrome is provided, and fig. 1 is a flowchart of a method for detecting apnea and hypopnea syndrome according to an embodiment of the present invention.

As shown in fig. 1, the process includes the steps of:

step S101, acquiring an electroencephalogram signal, an abdomen displacement signal and an oronasal airflow respiration signal.

Step S102, sleep stage is carried out based on the brain electrical signals, and sleep state information is determined.

And step S103, extracting a feature vector space based on abdomen displacement signal screening and wavelet transformation preprocessing.

Step S104, converting to a sleep state space according to the oral-nasal airflow respiratory signals, the sleep state information and the preset neural network model, and carrying out SVM statistics and respiratory state classification.

Respiration is the process of exchanging gases between the human body and the outside environment, by way of example, the human body completing the inhalation of oxygen and the expulsion of carbon dioxide through the action of breathing, providing energy for the biological activity. Clinically, the respiratory condition of a subject is evaluated by using an oral-nasal airflow signal, the normal respiratory frequency of a human body is approximately 12-20 times per minute, the waveform is approximately a sine wave, the amplitude is similar, no large fluctuation occurs, and if the amplitude or frequency of the oral-nasal airflow signal of the subject is changed to a large extent, the abnormal respiratory event of the subject can be judged.

Two methods are commonly used clinically to measure the oral nasal airflow of a subject: the first is a thermistor sensor and the second is a flow sensor. The principle of measuring the oral-nasal airflow by the thermistor sensor is as follows: the temperature of the exhaled air from the lung is higher, the temperature of the inhaled air from the external environment is lower, the thermistor changes the resistance value along with the change of the ambient temperature, and the volume of the inhaled and exhaled air of the subject can be calculated by comparing the change of the resistance value of the thermistor. The principle of measuring the oral-nasal airflow by the flow sensor is as follows: the sensor is placed in the respiratory flow path of the subject,

when the subject breathes, the two sides of the sensor generate gas pressure difference, and the breathing gas flow can be converted according to the following formula:

where q is the volume of gas flowing through, Δp is the difference in flow rate across the flow sensor, ρ is the density of the gas, and k is a coefficient, all related to a number of parameters such as the orifice diameter and expansion coefficient of the sensor.

According to the definition: the reduction of the oronasal respiratory airflow by more than 90% from the baseline amplitude in the sleep state, the duration of 10s and more, defined as an apneic event; the reduction in oronasal respiratory airflow from baseline amplitude was more than 30% and the duration was 10s and more defined as a hypopnea event. Therefore, after the oral-nasal airflow signal is measured, the characteristics of amplitude, median frequency, power spectrum entropy, LZ complexity, sample entropy and the like are generally extracted to detect and classify SAHS morbidity.

Figure 2 shows the hypopnea events and apnea events corresponding to the oronasal airflow signals, and box is the expert's noted morbidity time window. Based on the difference in the characteristics of the oral-nasal airflow signal, this signal can be used to classify normal, suspended and hypopnea in an SAHS event, but the SAHS subtype cannot be completed by this signal alone: classification of OSAS, MSAS, CSAS.

In sleep apnea events, the chest and abdomen movement states corresponding to the three pathogenesis types are different according to pathogenesis differences: OSAS is characterized by a cessation of oronasal airflow and a chest-abdomen respiratory effort; CSAS onset is characterized by cessation of both oronasal airflow and thoracoabdominal respiratory action; MSAS is characterized in that the respiratory actions of the front half part of the oral-nasal airflow and the chest and abdomen are stopped in the process of one-time apnea, and the respiratory actions of the rear half part of the oral-nasal airflow and the chest and abdomen are stopped. Four typical respiratory events corresponding to abdominal displacement signals are shown in fig. 3, with boxes being expert-labeled morbidity time windows. The difference between waveforms of different disease types and normal is very obvious, and SAHS detection and four classification can be performed by using the signals based on the difference of the abdominal movement characteristics.

The preparation work required to be done before formal data acquisition: the patient is informed that he/she should stop taking sleep affecting drugs within 24 hours before the test and should not consume alcohol or stimulating beverage within 8 hours. The instruments and medical supplies needed in the data acquisition process are prepared in advance: polysomnography, a corresponding electrode and a sensor, 75% medical alcohol, scrub, conductive paste, medical adhesive tape and other devices. The one-time complete sleep polysomnography detection flow comprises the following four steps:

(1) Before detection, the subject is asked to fill in a questionnaire to acquire basic information and sign an informed consent.

The questionnaire mainly comprises physiological information such as gender, age and the like of patients, and pathological information such as current medical history, past medical history and the like. The patient is informed of the attention points in the detection process by the simple detection process of the patient, and an informed consent is signed.

(2) The alcohol is used for disinfecting the part of the patient, where the electrode needs to be installed, and the scrub cream is used for removing dead skin, so that the electrode is better contacted with the collected part, and the signal collection quality is improved. The electrodes are arranged on the head, chest, legs, shoulders and other body parts of the subject by using the conductive paste, and the electrodes are in good contact with the skin by using the conductive paste. The abdomen displacement signal is acquired by a displacement sensor, and the abdomen motion sensor is tied at the position with the largest abdomen circumference. The electroencephalogram electrode is placed at the position of the right central-left earlobe. The flow sensor is used for the oral-nasal airflow signal, the two probes are respectively placed in nostrils on two sides of a subject, the two probes are wound to the rear of auricles from two sides and then meet at the front side of the neck, the nose wings are bonded to prevent falling off during night turning, and the original sampling rate is 10Hz. After the polysomnography is connected, the sleep monitoring process is formally started. The sleep laboratory turns off the lights at about 10 pm on the day so that the patient falls asleep as early as possible.

(3) In the detection process, a duty worker monitors whether the state of a patient is normal, whether an electrode falls off or other emergency situations through a camera outside a sleep laboratory. The acquisition time is about 7-12 hours.

(4) And after the patient wakes up in the daytime on the next day, the power supply of the sleep polysaccharidemeter is turned off, the electrode on the patient is removed after data acquisition is stopped, and the whole detection process is finished. The data of 126 subjects are totally 6 times to complete collection, the collection date and the number of samples collected each time are shown in the following table 1, and the basic information of the subjects is summarized in the following table 2:

TABLE 1

TABLE 2

The data acquired by the polysomnography is exported as EDF (European Data Format), which is a standard file format and is widely used for exchanging and storing medical time sequences. The format is characterized by the ability to store multi-channel data simultaneously and allow signals of each channel to have different sampling frequencies. The file contains a title and a plurality of data records, wherein the title contains information such as patient identification, measurement starting time, technical parameters of each signal and the like.

Alice sleep is business workstation software developed by Philips corporation, and can be matched with a plurality of sleep breath detection devices with different models to realize analysis of EDF files. The Alice sleep software operation interface is shown in fig. 4a, fig. 4b and fig. 4c, and can show all signals collected by the sleep polysomnography.

The sample data are marked back to back by two auxiliary primary and secondary doctors in the thoracic hospital in Tianjin, and the marking flow is to firstly integrate brain electrical signals, eye movement signals and the like to mark sleep information. The sleep information contains the sleep structure, onset and duration of each stage overnight for each patient following sleep stage rules established by the american sleep medical society, one sleep stage label for each 30s sample. And then, removing data corresponding to the wake-up period in the data after sleep stage, and then, integrating multiple signal waveforms such as an oral-nasal airflow breathing signal, an abdomen displacement signal, an electrocardiosignal, a snore signal and the like in software to label the abnormal sleep breathing event. Different types of episodes are noted using rectangular blocks of different depth, as shown in fig. 4a, 4b and 4c, the noted content including type, onset and duration of sleep disordered breathing events, to the nearest 0.1s.

Note that the upper box in fig. 4a marks the hypopnea event, the lower box marks the snore event (which is not relevant to the study), the box in fig. 4b marks the OSAS event, and the box in fig. 4c marks the CSAS event. Taking fig. 4a as an example, the 1 st to 2 nd signals from top to bottom are respectively left and right eye electric signals, the 3 rd to 4 th signals are brain electric signals, the 5 th signals are frequency tongue myoelectric signals, the 6 th is pressure type mouth-nose breathing airflow signals, the 7 th is heat-collecting type mouth-nose breathing airflow signals, the 8 th to 10 th is chest-abdomen displacement signals, the 11 th is snore signals, the 12 th to 13 th signals are electrocardiosignals, the 14 th is heart beat period (heart rate), the 15 th is percutaneous blood oxygen saturation signals, the 16 th is pulse wave signals, the 17 th to 19 th signals are pulse conduction time, the 20 th is pulse rate, the 21 st is sleep position signals, and the 22 nd is sleep stage labels marked by doctors.

If a dispute is encountered, three more physicians are presented to make a conclusion. The formed csv file format label file contains two parts of sleep information and sleep breathing abnormality information of the subject, a patient number A000035 sleep information label (section) is shown in table 3, and a patient number A000035 sleep breathing abnormality label (section) is shown in table 4.

TABLE 3 Table 3

TABLE 4 Table 4

And counting the number of times of the illness of the patient at night according to the label information, dividing the number of times of the illness of the patient at night by the sleeping time length to calculate the AHI index of the patient, and judging the severity of the illness of the patient according to the AHI index. The statistics of the disease levels for 96 patients herein are shown in table 5.

TABLE 5

The electroencephalogram signal is introduced to identify the sleep state of the subject, complete data screening and further improve SAHS detection and classification accuracy, so that the algorithm is designed to divide the night monitoring data of the subject into sleep stage and wake stage by utilizing the characteristics of the electroencephalogram signal. The sleeping state labels are marked on two paths of signals of mouth-nose airflow respiration and abdomen displacement, the sleeping labels are deleted to be signal parts corresponding to the wake-up period, irrelevant signals are removed to complete signal screening, the quality of a training set is improved, and the detection and classification accuracy is improved. A technical route diagram of the sleep state recognition algorithm based on the electroencephalogram signals is shown in fig. 5.

And extracting the required electroencephalogram signals, abdomen displacement signals and oronasal airflow signals from the EDF file, and respectively preprocessing the electroencephalogram signals, the abdomen displacement signals and the oronasal airflow signals. For the electroencephalogram signals, due to the limitation of an acquisition method, interference of physiological factors such as respiration, eye movement, myoelectricity and the like can be inevitably introduced, and meanwhile, 50Hz power frequency interference can be brought, and filtering treatment is needed.

Wavelet transformation is currently applied to a large range in various fields such as subject signal processing, image compression, voice analysis and the like, and is often used in the analysis field of sleep electroencephalogram because the time window and the frequency window can be adaptively changed to generate variable resolution characteristics.

The wavelet transformation is an improved signal analysis method based on short-time Fourier transformation, meets the localization requirement of time-frequency analysis, and is characterized in that the length of an observation window is fixed, and the shape of the window can be adjusted in a self-adaptive manner. The wavelet transformation is to select proper basic wavelet first, and the following compatibility conditions are needed to be satisfied:

wherein the method comprises the steps ofIs the fourier transform of ψ (ω). Forming a cluster of wavelet system psi by shifting and scaling the basic wavelet psi (t) _a，b (t)，ψ _a，b The expression of (t) is:

b is time shift and a is scale factor. The cluster of wavelets forms as basis a series of subspace models of different sizes, and then projects the signal to be analyzed into the respective subspaces, and the time-scale characteristics of the signal can be obtained through decomposition on different scales.

The wavelet transformation principle is shown in fig. 6, where D represents the decomposed profile signal and mainly includes a high-frequency signal portion, and a represents the detail signal and mainly includes a low-frequency signal portion; the lower layer decomposition object is a low-frequency part obtained by upper layer decomposition, and then decomposed according to the same rule.

The wavelet transform can be divided into the following according to the difference in its basic wavelet selection: classical wavelet, orthogonal wavelet and bi-orthogonal wavelet.

Among classical wavelets are commonly used: haar wavelets, mexican Hat wavelets, gaussian wavelets, and the like. The wavelet is defined as:

the Haar wavelet waveform is shown in fig. 7. Haar waves are characterized by a tight support at the time domain level, have the advantage of removing phase distortions arising in the system, and have the disadvantage that they are not continuous wavelets, subject to certain limitations in handling practical signal problems. The common classes of orthogonal wavelets can be divided into the following classes: daubechies wavelet, coifles wavelet, meyer wavelet, symmetrical wavelet, etc., such wavelets are represented by scaling functions (Scalling function) multiplied by different weights and added.

Daubechies wavelet, dbN wavelet for short, N being the order of the wavelet. The characteristic is that the larger the order N is, the larger the vanishing moment (Vanishing Moments) is, the more concentrated the energy distribution is, and the more obvious the frequency band dividing effect is. However, the longer the corresponding filter length is, the lower the supportability is, the operation amount is obviously increased, the time consumption of the algorithm is prolonged, and the real-time performance is reduced.

Energy characteristics: according to the pasival Theorem (Parseval's Theore), the target signal contains energy identical to the sum of the energy contained in the components into which the signal is decomposed in the perfect orthogonal function set. The difference of the energy of the brain electrical signals among different individuals is very large, but the characteristic of the energy ratio of each frequency band of the brain electrical signals can reflect the state of the brain of the same study object in different periods, so the energy ratio of each frequency band of the brain electrical signals can be used as a key characteristic of sleep stage.

The energy definition of each frequency band of the brain electricity is as follows:

wherein x is _i (t) is the time domain signal corresponding to different frequency bands, E _i Is the energy of the frequency band.

The energy of each frequency band of the brain electricity is as follows:

the calculated features include: a ratio of alpha energy to total energy, a ratio of beta energy to total energy, a ratio of delta energy to total energy, a ratio of theta energy to total energy.

An electroencephalogram signal is a nonlinear, non-stationary, chaotic weak electrical signal, and thus some nonlinear characteristics can be used to characterize a subject's brain activity state, including: KS Entropy (Kolmogorov-sina, KS), approximate Entropy (Approximate Entropy, apEn) and Sample Entropy (Sample Entropy, sampEn, et al).

Both the approximate entropy and the sample entropy are methods of quantifying the temporal complexity of the sequence, essentially by detecting the size of the probability of generating a new sub-sequence in the time sequence. The lower the entropy value, the higher the self-similarity of the signal is characterized, and the higher the entropy value, the lower the similarity of the signal sequence is characterized, namely the more the sequence tends to be chaotic and complex. The calculation of the approximate entropy has the step of calculating the self-matching value, which can generate errors, and the improved sample entropy is abandoned in the calculation, so that part of errors are eliminated. The sample entropy is chosen to be used as a key feature of sleep stage.

The sample entropy is calculated specifically as follows:

(1) Defining an original sequence u=u { u (1), u (2), …, u (N) } with N data sources, and time-sequentially arranging the original sequence into a set of m-dimensional vectors X _m (1),X _m (2),…X _m (N-m+1), wherein: x is X _i (i)＝[u(i),u(i+1),…,u(i+m-1)],(1≤i≤N＝m+1`)。

The vector represents the values of consecutive m sequences u starting from the i-th point;

(2) Defining two m-dimensional vectors X _m (i) And X _m (j) The distance between them is d X _m (i),X _m (j)]To indicate that the two vectors are differenced with respect to each other, wherein the maximum value of the difference is the distance between the two vectors, namely:

d[X _m (i),X _m (j)]＝max{u(i+k)-u(j+k)}

(0≤k≤m-1,1≤N-m+1,1≤j≤N-m+1,i≠i)

(3) Given a threshold r (r > 0), traversing all i values, counting dX _m (i),X _m (j)]Number N less than threshold r ^m (i) Calculating the ratio of the data below the threshold to the total number N-m, noted as:

(4) Calculation ofAverage for all i:

(5) Substituting m+1 for m, repeating steps (1) - (4), and constructing C _i ^m+1 (r)、φ ^m+1 (r)。

(6) The sample entropy is defined as follows:

m represents the embedding dimension, r is the tolerance threshold, and SD is defined as the standard deviation of the original sequence. The value of SampEn is related to m and r, the common value of m is 1 or 2, and the value of r is related to SD, and the common value range is. And finally, the embedding dimension and the tolerance threshold are selected through comparison of multiple experimental results. The data length N selected during sample entropy calculation is required to be 100-5000[66], the length N is selected to be 100, the step length is 10, and the filling mode is minimum filling. And calculating the average value and variance of all sample entropies through calculating a series of sample entropy values corresponding to the electroencephalogram of different frequency bands after 30 seconds of electroencephalogram wavelet decomposition, and realizing the feature extraction of the sample entropies of the electroencephalogram of different frequency bands.

The electroencephalogram signal features extracted by the research have correlation, so that in order to eliminate information redundancy, the data processing speed is improved, and feature dimension reduction is performed while the current information is fully reserved. The study uses principal component analysis (Principal Component Analysis, PCA) algorithm to perform feature dimension reduction. PCA is a dimension reduction method for extracting main characteristic components of data, and a group of linearly related variables are changed into independent new variables by positive-negative conversion. The main idea is to reduce the data dimension on the premise of maximally reserving data information, enable all main components to be mutually independent, and finally order the data according to the contribution of the main components to the data difference so as to facilitate the subsequent feature screening. The PCA algorithm calculation flow is divided into the following four steps:

(1) Let the original eigenvector be X { X } ₁ ,x ₂ ,…,x _m M represents the feature dimension, n is the number of sample sets, x _ij The i-th feature vector of the i-th sample is characterized. For x _ij Normalized to obtain

And S is _j The sample mean and standard deviation of the jth feature vector are shown.

(2) Calculate the correlation coefficient matrix r= (R) _ij ) _m×m ：

r _ij Is the correlation coefficient between the ith and jth eigenvectors, where r _ii ＝1，r _ij ＝r _ji 。

(3) Calculating the characteristic value lambda of the correlation coefficient matrix ₁ ≥λ ₂ ≥…λ _m 0 and eigenvector u ₁ ,u ₂ ,…,u _m Definition u _j ＝(u _1j ,u _2j ,…,u _mj ) ^T The original feature vector composition m new principal component vectors can be expressed as:

wherein y is _m Representing the mth principal component vector.

(4) Calculating a characteristic value lambda _j Information contribution ratio and cumulative contribution ratio of (j=1, 2, …, m). Definition of principal component y _j Information contribution ratio b of (2) _j The method comprises the following steps:

when alpha is _p When the value is close to 1, the first p new feature vectors are selected as the principal components to replace the original m feature vectors, so that the p principal components are analyzed to obtain the comprehensive evaluation value Z as follows:

the support vector machine (Support Vector Machine, SVM) is a machine learning algorithm generated based on statistical learning, and the algorithm combines model complexity and learning capacity, so that a global optimal solution can be obtained by solving in a small-scale sample. SVM is often used in the field of nonlinear signal processing and classification, particularly in sleep stage problems based on electroencephalogram signals. The study uses an SVM classification model for sleep state identification.

Taking the linear separable classification problem as an example, two classification samples (x _i ,y _i ) Wherein x is _i ∈R ^d ，y _i E.+ -. 1, i-1,2, …, N. d is the feature dimension, X, of each sample _i Represents the ith feature vector, y _i And the label corresponding to the true category of the sample is represented. Constructing a hyperplane, g (x) = (w.x+b) = 0, classifies the N samples correctly. The distance from each sample point to be classified to the hyperplane is defined as a geometric interval, and the formula is as follows:

the minimum value is selected from all geometric intervals:

and defining positive and negative samples with positions closest to the hyperplane on the interval boundary as support vectors, wherein the minimum value of the geometric interval obtained by solving in the following formula is the distance from the support vectors to the hyperplane. According to the machine learning optimization problem, the conditions are equivalent to:

the classification concept is illustrated by the graph shown in fig. 8, where the solid line represents the decision boundary, the dashed line represents the interval boundary, and the light-colored dots represent the support vectors:

the problem is converted into an inequality constraint quadratic programming problem, and the Lagrangian multiplier is introduced to convert the problem into an unconstrained problem. Weight vectors are available according to the formula:

wherein alpha is _i (1, 2, …, l) is the Lagrangian multiplier, selected α ^* Is a component alpha of (a) ^* _j So that it satisfies the condition 0<α ^* _j <C, calculating b ^* ：

Finally, deriving a linear separable classification decision function as:

if the sample data belongs to the problem of nonlinearity in the feature space, the sample data is converted into a linear separable problem under the higher-dimensional feature space. Selecting the appropriate kernel function at the time of conversion corresponds the inner product operation in the inseparable space to the inner product operation in the inseparable space. The original features are mapped to the high-dimensional space to obtain updated new features, and the new features are recorded as a schematic diagram of the mapping process of the sample from the low dimension to the high dimension, as shown in fig. 9.

The nonlinear classification is substantially similar to the linear separable derivation of ideas, and finally a nonlinear separable classification decision function can be derived:

common kernel functions include linear kernel, polynomial kernel, radial basis kernel, sigmoid kernel, etc. Linear kernel:

K(x _i ，x)＝x _i ·x

a polynomial core:

K(x _i ，x)＝[(x _i ·x)+1] ^d

radial basis function:

sigmoid core:

all four kernel functions are applicable to the situation that the sample data are less and the dimension is lower. Radial basis functions and Sigmoid kernel functions are typically employed when the number of samples is very large. The radial basis function has a good recognition effect on the classification problem of higher sample data dimension, an example sample can be mapped from an original input space to an infinite dimension space, the generalization capability of the model is greatly enhanced, and the convergence domain can be maintained at a wider level.

In this embodiment, an apparatus for detecting an apnea-hypopnea syndrome is further provided, and the apparatus is used to implement the foregoing embodiments and preferred embodiments, and will not be described in detail. As used below, the term "module" may be a combination of software and/or hardware that implements a predetermined function. While the means described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

The present embodiment provides an apnea and hypopnea syndrome detection device, as shown in fig. 10, comprising:

an acquisition signal module 1001, configured to acquire an electroencephalogram signal, an abdomen displacement signal, and an oronasal airflow respiration signal;

a determining state module 1002, configured to perform sleep staging based on the electroencephalogram signal, and determine sleep state information;

a feature extraction module 1003 for extracting a feature vector space based on the abdomen displacement signal screening and wavelet transformation preprocessing;

the statistics module 1004 is configured to convert to a sleep state space according to the oral-nasal airflow respiration signal, the sleep state information, and the preset neural network model, and perform SVM statistics and respiration state classification.

Further functional descriptions of the above respective modules and units are the same as those of the above corresponding embodiments, and are not repeated here.

The apnea and hypopnea syndrome detection device in this embodiment is in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit ) circuit, a processor and memory executing one or more software or fixed programs, and/or other devices that may provide the above functions.

The embodiment of the invention also provides computer equipment, which is provided with the apnea and hypopnea syndrome detection device shown in the figure 10.

Referring to fig. 11, fig. 11 is a schematic structural diagram of a computer device according to an alternative embodiment of the present invention, as shown in fig. 11, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are communicatively coupled to each other using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions executing within the computer device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface. In some alternative embodiments, multiple processors and/or multiple buses may be used, if desired, along with multiple memories and multiple memories. Also, multiple computer devices may be connected, each providing a portion of the necessary operations (e.g., as a server array, a set of blade servers, or a multiprocessor system). One processor 10 is illustrated in fig. 11.

The processor 10 may be a central processor, a network processor, or a combination thereof. The processor 10 may further include a hardware chip, among others. The hardware chip may be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general-purpose array logic, or any combination thereof.

Wherein the memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform a method for implementing the embodiments described above.

The memory 20 may include a storage program area that may store an operating system, at least one application program required for functions, and a storage data area; the storage data area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, memory 20 may optionally include memory located remotely from processor 10, which may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

Memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk, or solid state disk; the memory 20 may also comprise a combination of the above types of memories.

The computer device also includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

The embodiments of the present invention also provide a computer readable storage medium, and the method according to the embodiments of the present invention described above may be implemented in hardware, firmware, or as a computer code which may be recorded on a storage medium, or as original stored in a remote storage medium or a non-transitory machine readable storage medium downloaded through a network and to be stored in a local storage medium, so that the method described herein may be stored on such software process on a storage medium using a general purpose computer, a special purpose processor, or programmable or special purpose hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, a solid state disk or the like; further, the storage medium may also comprise a combination of memories of the kind described above. It will be appreciated that a computer, processor, microprocessor controller or programmable hardware includes a storage element that can store or receive software or computer code that, when accessed and executed by the computer, processor or hardware, implements the methods illustrated by the above embodiments.

Although embodiments of the present invention have been described in connection with the accompanying drawings, various modifications and variations may be made by those skilled in the art without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope of the invention as defined by the appended claims.

Claims

1. A method for detecting apnea and hypopnea syndrome, said method comprising:

screening and preprocessing wavelet transformation based on the abdomen displacement signals to extract a feature vector space;

and converting to a sleep state space according to the oral-nasal airflow respiratory signal, the sleep state information and a preset neural network model, and carrying out SVM statistics and respiratory state classification.

2. An apnea and hypopnea syndrome detection device, said device comprising:

the feature extraction module is used for extracting a feature vector space based on the abdomen displacement signal screening and wavelet transformation preprocessing;

and the statistics module is used for converting into a sleep state space according to the oral-nasal airflow respiration signals, the sleep state information and a preset neural network model, and carrying out SVM statistics and respiration state classification.

3. A computer device, comprising:

a memory and a processor, said memory and said processor being communicatively coupled to each other, said memory having stored therein computer instructions, said processor executing said computer instructions to perform the apnea and hypopnea syndrome detection method according to claim 1.

4. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the apnea and hypopnea syndrome detection method according to claim 1.