CN116250823A

CN116250823A - Sleep breathing abnormality early warning method based on MSET and real-time dynamic baseline

Info

Publication number: CN116250823A
Application number: CN202111503075.8A
Authority: CN
Inventors: 张翼英; 柳依阳; 王德龙; 马彩霞; 王鹏凯
Original assignee: Tianjin University of Science and Technology
Current assignee: Tianjin University of Science and Technology
Priority date: 2021-12-10
Filing date: 2021-12-10
Publication date: 2023-06-13

Abstract

The invention relates to a sleep breathing abnormality early warning method based on MSET and a real-time dynamic baseline, which is mainly technically characterized in that: and training a sleep breathing state health state evaluation model by using a large amount of data, evaluating the sleep breathing state according to the deviation of the model, and mining potential abnormal information of the patient. And calculating a real-time dynamic baseline by adopting a support vector regression prediction method, so that the judgment of the sleep respiratory state is realized and the future sleep respiratory state can be predicted. Is more accurate than the fixed threshold method. According to the contribution rate of the respiratory monitoring index, abnormal information of the respiratory monitoring index is further mined, and more accurate information is provided for doctors. The method can accurately find abnormal body states in early stage, and remind people to take effective measures in time. The scheme of the invention aims to monitor the sleep breathing state in real time, early warn the abnormal state and predict the trend, thereby providing precious treatment time for patients and improving the early warning accuracy.

Description

Sleep breathing abnormality early warning method based on MSET and real-time dynamic baseline

Technical Field

The invention relates to a sleep breathing abnormality early warning method based on an MSET and a real-time dynamic baseline, which is used for carrying out real-time monitoring on sleep breathing based on the MSET, carrying out early warning on abnormal states by utilizing the real-time dynamic baseline and carrying out trend prediction, providing precious treatment time for patients and improving early warning accuracy.

Background

The conventional sleep quality detection method focuses on the detection of information such as brain waves and heartbeat during sleep, and the information such as respiration plays an important role. Some commercial devices collect data from headphones or bracelets worn by the user for analysis of sleep quality or use a set of pressure sensitive sensors to monitor sleep disturbances in the mattress. But these devices are uncomfortable to wear and are not suitable for everyday use.

PSG is one of the standard methods commonly used for diagnosing sleep disordered breathing. PSG monitoring is a sleep monitoring of a patient using multiple sensors, which has certain limitations. Therefore, the method adopts MSET and a real-time dynamic baseline method, and can better master the sleep respiratory state, track the abnormal development process in real time, and discover the sleep respiratory abnormal state and main monitoring index abnormal information in advance.

Disclosure of Invention

The invention is based on the extraction of abnormal characteristic quantity, firstly, a model of the normal sleep breathing state of the tested person is established by using the MSET, the sleep breathing state of the tested person is optimally estimated, and the estimated value and the sleep breathing of the tested person have a certain correlation. And constructing a real-time dynamic baseline according to the deviation, and judging the sleep breathing state. The method can accurately find abnormal body states in early stage, and remind people to take effective measures in time.

In order to achieve the above object, the present invention adopts the following technical steps:

step A: data preprocessing

Four characteristic values of heart beat frequency, respiratory frequency, snore and limb jump are selected. Data preprocessing operations were performed on the 4 sleep respiratory indices. The invention performs maximum-minimum normalization: mapping the data between [0,1] is convenient for comprehensive analysis.

And (B) step (B): historic memory matrix construction

Since the data of all health states is too large, the history memory matrix is too large, which results in a large calculation amount. Therefore, it is necessary to select a representative state vector to construct a history memory matrix on the basis of ensuring coverage of all health states. The time interval sampling method is selected herein, and the time interval between each sampling point is fixed. By way of example with respiratory rate, the normal respiratory rate of an adult is 16-18 times/minute, and the respiratory rate is reduced during sleep. And the normal breathing rate of infants is a little faster.

The equally spaced sampling process is shown in fig. 1.

Step C: sleep respiratory state estimation

Based on training a health state estimation model, 4 sleep respiratory indexes are subjected to state estimation by adopting a multi-element state estimation method. And selecting the most representative respiratory monitoring index for state and deviation analysis, comparing the observed value with the model estimated deviation in time sequence, and indicating that the model has stronger abnormality identification capability when the deviation of all variables trained by the health state estimation model is near 0. The respiratory rate estimated value in a certain period is assumed to have larger deviation, larger fluctuation and larger fluctuation of respiratory rate, and other respiratory indexes have small change, so that the monitoring index is possibly abnormal.

The sleep breathing state estimation process is shown in fig. 2.

Step D: real-time state monitoring and early warning

The method defines the multi-dimensional integral deviation degree, and adopts a fixed threshold value and a real-time dynamic baseline to carry out real-time on-line monitoring and early warning on the sleep health state of the tested person. And taking the absolute residual error mean value as a measurement standard of the two-dimensional vector deviation degree, setting an early warning coefficient k=20%, and solving a fixed threshold value of the whole deviation degree so as to determine all abnormal moments. Whether the comparison deviation exceeds an early warning threshold value or not is judged, whether the monitored person has abnormal signs or not is judged, and therefore abnormal early warning is achieved.

Compared with the prior art, the invention has the following advantages and positive effects:

the invention utilizes MSET method and real-time dynamic baseline method to monitor the sleep respiratory state in real time with multiple indexes and extremely high accuracy. Not only can the sleep breathing state be monitored in real time, but also trend prediction can be carried out on each breathing monitoring index. The abnormal state information is timely and accurately sent in early stage of abnormal state occurrence, so that precious treatment time is won for patients.

Drawings

FIG. 1 is an equally spaced sampling process

FIG. 2 is a sleep breathing state estimation process

FIG. 3 is a flow chart of a sleep breathing state determination process

Detailed Description

Step A: constructing a history memory matrix according to the history health data:

assuming n respiratory monitoring indicators during sleep respiratory monitoring, at t _k The observation vector at the moment is noted as:

X(t _k )＝[x ₁ (t _k )，x ₂ (t _k )，…，x _n (t _k )] ^T (1)

note that: x in the formula _n (t _k ) Indicated as nth respiration monitoring index at t _k Monitoring data of time. The history matrix D can be expressed in the form of n x m:

and (B) step (B): generating a dynamic baseline:

the statistical index for judging the health state of the monitored person based on the deviation sequence composition can be defined as the overall deviation degree of the system, and the statistical index is expressed by a formula:

wherein: x is the observation vector and Y is the estimation vector. s reflects the degree of deviation of the current sleep-breathing state from the normal state.

Under the condition of real-time monitoring, each interval t ₀ Obtaining a newly added observation matrix in a time period, wherein the acquisition frequency is P ₀ . The newly added observation matrix is as follows:

the whole deviation degree is calculated by the observation vectors in the newly added observation matrix one by one, and a deviation degree sequence can be obtained:

S _t ＝[s ₁ ，s ₂ ，…，s _m ] (5)

step1: take t nearest to the current time ₁ A time period. Calculate each t ₀ Mean N of the departure sequences over a period of time _i Sum of variances R _i (i=1, 2, …, t1/t 0), constituting an N sequence and an R sequence.

Step2: predicting the mean value sequence by using a support vector regression algorithm, wherein the predicted value is recorded as follows:

step3: dynamic baseline for future t+1 time periods.

If the overall deviation s at a certain moment is less than or equal to N _y The situation is normal at this moment; if the overall deviation s is greater than N _y An abnormal situation occurs.

Step C: sleep breathing state analysis:

when sleep breathing is abnormal, it is often the case that some respiratory monitoring indicators deviate to a large extent, resulting in an overall degree of deviation that is greater than the dynamic baseline. In summary, to determine the cause of the abnormality requires finding respiratory monitoring indicators that deviate significantly, and can be determined by comparing the contribution rates.

In abnormal situations, respiration monitoring index contribution rate:

wherein: cnp (i) is expressed as the contribution rate of the ith respiratory monitoring index, X _obs (i, j) represents the observed vector of the ith respiratory marker at j, X _est (i, j) represents the estimated vector of the ith respiratory index at the moment j, and m is the number of observation vectors.

Step D: maximum-minimum normalized pretreatment operations were performed on 4 sleep respiratory indices.

Step E: real-time state monitoring and early warning:

by adopting a support vector regression prediction method, the dynamic range of the overall deviation of each respiratory monitoring index is obtained through analyzing the dynamic deviation of data in a certain period, and the dynamic baseline and the overall deviation in a certain period in the future are subjected to nonlinear comparison, so that trend prediction is carried out on each respiratory monitoring index.

A flow chart of the sleep breathing state determination process is shown in fig. 3.

The foregoing description is only of the preferred embodiments of the invention, and all changes and modifications that come within the scope of the invention as defined by the appended claims are intended to be embraced therein.

Claims

1. The sleep breathing abnormality early warning method based on the MSET and the real-time dynamic baseline is characterized by comprising the following steps of:

(1) Acquiring sleep respiratory health data;

(2) Determining sleep respiration monitoring indexes;

(3) Collecting sleep breathing data;

(4) Constructing a sleep breathing index prediction model;

(5) And constructing an evaluation index to judge the abnormal sleep respiratory state.

2. The method of claim 1, wherein the sleep respiration monitoring metric comprises: heart beat frequency, respiratory frequency, snoring and limb jump.

3. A method according to claims 1-2, characterized in that the data related to sleep respiration monitoring is obtained from a sensor terminal, including heart beat frequency, respiratory frequency, snoring and limb jump.

4. A method according to claims 1-3, characterized in that the MSET is used to build a prediction model of the monitoring index in the normal state of sleep breathing based on the acquired sleep breathing health data, and the deviation is calculated for the real-time monitored sleep breathing data.

5. The method of claim 4, wherein the determination of sleep breathing state is achieved by constructing a real-time dynamic baseline using a support vector regression prediction method based on the obtained deviation.

6. The method according to claims 1-5, wherein for the limitation of a single sleep respiration monitoring index and a fixed threshold value, a method for predicting sleep respiration state according to a dynamic baseline is provided, and the dynamic baseline is calculated in real time by learning historical health data, so that the real-time monitoring and abnormality analysis of sleep respiration state are realized.

7. The sleep disordered breathing method based on the MSET and the real-time dynamic baseline according to claim 1, wherein the method comprises the following steps: the system design principle is simple, easy to realize, high in execution capacity, high in accuracy and capable of supporting a large amount of data; the MSET method is utilized to monitor the sleep breathing state in real time with a plurality of indexes and extremely high accuracy; by using a real-time dynamic baseline method, not only can the current state be monitored, but also the future state can be predicted.