CN113662520B - Wearable continuous blood pressure measuring system based on uncertainty quantification strategy - Google Patents
Wearable continuous blood pressure measuring system based on uncertainty quantification strategy Download PDFInfo
- Publication number
- CN113662520B CN113662520B CN202110985640.2A CN202110985640A CN113662520B CN 113662520 B CN113662520 B CN 113662520B CN 202110985640 A CN202110985640 A CN 202110985640A CN 113662520 B CN113662520 B CN 113662520B
- Authority
- CN
- China
- Prior art keywords
- model
- blood pressure
- theta
- probability
- parameter
- 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
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
- A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
- A61B5/02116—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics of pulse wave amplitude
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
- A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
- A61B5/02125—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics of pulse wave propagation time
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
- A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
- A61B5/6802—Sensor mounted on worn items
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
Abstract
The invention discloses a wearable continuous blood pressure measuring system based on an uncertainty quantification strategy, and belongs to the technical field of medical instruments. The signal acquisition and processing unit can be completed by a micro-control chip, can realize miniaturized wearable or undisturbed continuous monitoring, and can be conveniently applied to personal, family and clinical monitoring. According to the invention, bayesian combination prediction based on a Markov Monte Carlo simulation method is introduced into sleeveless blood pressure prediction for the first time, and the method has the following advantages: when different conditional probability distributions are used, the algorithm is not required to be changed; the Bayes combined prediction takes the posterior probability as weight to carry out weighted average on the selected single prediction model, and takes the posterior probability as the standard for judging the quality of the model, so that the parameter with high average correlation of the Bayes model can be processed, and meanwhile, the overall sample of the Bayes combined prediction contains the prediction of different single models.
Description
Technical Field
The invention belongs to the technical field of medical instruments, and particularly relates to a cuff-free continuous blood pressure measuring device based on uncertainty quantification.
Background
There are two methods for blood pressure measurement: direct measurement and indirect measurement. The direct measurement method is to directly measure the arterial blood pressure after the catheter is placed through arterial puncture, and has the advantages of directly and continuously measuring the intra-arterial pressure, accurate measurement value and the defect of needing professional technicians and being invasive, so the method is only suitable for critically ill and major surgery patients. Indirect measurement methods are widely used in clinical practice because of their non-invasive nature, and commonly employ cuff compression methods, including korotkoff sound-based auscultation, oscillography, radial artery applanation and arterial volume compensation. However, due to the limitation of technical principles, the methods have problems which are difficult to solve in practical application. For example, the auscultation method and the oscillometric method can only measure single-point blood pressure and cannot provide continuous blood pressure readings, and the inflation and deflation of the cuff can bring discomfort to the detected person; although the arterial applanation tension method and the arterial volume compensation method can be used for non-invasive continuous blood pressure measurement, they are still invasive, cause discomfort to the subject, and are complicated and bulky in apparatus, require a professional operator, and have limited application. Therefore, a new blood pressure measuring method is developed, non-invasive, undisturbed, continuous and real-time measurement of the blood pressure is realized, and the method has very important significance for early discovery, early diagnosis, early treatment and effective management of the hypertension. In recent years, with the development of wearable sensing and intelligent algorithms, a cuff-free blood pressure measurement method such as a Pulse Transit Time (PTT) based method provides a new method for non-invasive, continuous and undisturbed blood pressure monitoring due to the simple structure of the measurement equipment and no need of inflating the cuff. For cuff-free continuous blood pressure measurement, the measurement accuracy is one of the main performance indexes, and is represented as follows: 1) The measurement precision has universality among different groups and individuals; 2) The measurement accuracy is maintained for as long as possible after the initial calibration.
Currently, many advances have been made in the research on cuff-free blood pressure measurement (including PTT principle method), but it is still difficult to realize high-precision continuous blood pressure measurement in accordance with clinical medical standards in the prior art. The problems of the cuff-free blood pressure accurate measurement are mainly reflected in the following three aspects: (1) The cuff-less blood pressure estimation model form has uncertainty: researchers all hope to establish a high-precision model to accurately reflect noninvasive blood pressure change indexesThe relationship with the interaction and influence between the blood pressure so as to accurately estimate the blood pressure value. However, there are many factors that affect blood pressure regulation, and usually, researchers only consider the main factors and ignore the secondary factors to establish a blood pressure estimation model. In addition, in the modeling process, different researchers have different expression forms due to different knowledge reserves and cognitive levels. Existing models describing cuff-free blood pressure estimation include a PTT-based ln (PTT) model, a 1/PTT model, and an e k*PTT Models, and the like, and combined with PTT and other characteristic models for representing blood pressure change, such as PTT and PIR models, and the forms of the models have certain differences. (2) The cuff-less blood pressure estimation model parameters have uncertainty: the cardiovascular system has high dynamic characteristics, the performance state of the cardiovascular system is influenced by various factors, and the body surface characteristics obtained by the wearable device are influenced by the environment and a measuring system to change, so that the blood pressure change characteristics and the cardiovascular system have uncertainty. This uncertainty collectively affects the distribution of the model parameters, thereby making the model parameters uncertain. In addition, the different characteristics of the population and the physiological characteristics of individuals, including age, blood volume, vascular elasticity, cardiac function, and autonomic function regulation, lead to population and individual variability in blood pressure measurement. The dynamic characteristics of blood pressure determine the differences in the individual, such as diurnal difference in blood pressure and changes due to environmental factors such as individual behavior, mental state, body temperature, etc. (3) model estimation error has uncertainty: the estimation of blood pressure by using a mechanism or data-driven model causes uncertainty in the error between the estimation result of the model and the actual blood pressure because the actual blood pressure is affected by various disturbances (such as heart rate, elasticity of aorta, peripheral resistance, etc.), and this uncertainty is called model estimation error uncertainty. To solve the above problems well, not only improvement of data analysis and modeling means but also intensive research and improvement of cuff-less blood pressure measurement from physiological mechanisms are required.
Disclosure of Invention
The invention aims to provide a wearable cuff-free continuous blood pressure measuring system which aims at solving the problem that the existing noninvasive continuous blood pressure measurement is inaccurate and realizing the individualized and accurate measurement of cuff-free blood pressure.
The technical scheme of the invention is a wearable continuous blood pressure measuring system based on an uncertainty quantification strategy, which comprises: the blood pressure monitoring device comprises a data acquisition device, a cuff sphygmomanometer, a data preprocessing module, a feature extraction module, a blood pressure calculation module and a blood pressure display module;
the data acquisition unit acquires a reflected light signal or a pressure signal of a certain part of a human body and simultaneously acquires an electrocardiosignal; transmitting all the acquired signals to a data preprocessing module;
the cuff sphygmomanometer acquires the pulse blood pressure of a human body and transmits the blood pressure to the blood pressure calculation module;
the data preprocessing module is used for filtering and amplifying all signals collected by the data collector and then converting the obtained reflected light signals or pressure signals into pulse waves; transmitting the preprocessed electrocardiosignals and the pulse wave signals obtained by conversion to a feature extraction module;
the features calculated by the feature extraction module include: time interval PTT between the maximum rising edge of the pulse wave and the R wave peak value of the electrocardiosignal; pulse wave intensity ratio PIR of the ratio of the peak intensity to the trough intensity of the pulse wave; the time width PWHA corresponding to the pulse wave half pulse amplitude; then transmitting the calculated characteristics to a blood pressure calculation module;
the blood pressure calculation module comprises: a model memory, a model selection module,
The model memory stores 7 models, respectively:
wherein, SBP 0 For initial calibration of systolic pressure, DBP 0 For initial calibration of diastolic pressure, MBP 0 For initial calibration of mean blood pressure, PIR 0 PTT for initial calibration of pulse wave intensity ratio 0 For initial calibration of the pulse wave propagation time, gamma is a factor related to the state of the blood vessel, a Mi And b Mi I =1,2, \ 8230, 7 is a model coefficient corresponding to the models M1 to M7, respectively;
the parameter selection method in the model selection module comprises the following steps:
step 1: carrying out uncertainty quantification on each model parameter based on a Monte Carlo-Markov chain Metropolis-Hastings (M-H) sampling algorithm; the following operations were performed for each model:
step 1.1: initializing time T and sampling times T;
step 1.2: setting relevant model parameters u = (u) 1 ,u 2 ,...,u N ) And initializing the initial state theta of the parameter sequence (t) = U, where U is an N-dimensional value randomly generated from uniformly distributed U (min, max), and N is the number of uncertainty parameters in the model, e.g., model M4, where the model parameter to be determined is a M4 And b M4 I.e., N =2;
step 1.3: the following procedure was repeated:
i. let t = t +1;
for each dimension model parameter: i =1,2, \ 8230;, N;
known q (θ) i ) Obey a mean value of u i A normal distribution with variance of 1, with a probability density function of:
from q (theta) i |θ i (t-1) ) To generate a new candidate state theta i (*) ,q(θ i |θ i (t-1) ) Given a previous state theta i (t-1) Current state of i The condition distribution of (1); let the whole vector before update be:
calculating the probability of acceptance:wherein p (Θ) represents the probability of the updated parameter vector Θ;
v. generating a random value a from the uniform distribution U (0, 1);
if a ≦ α, accepting the newly generated value: theta.theta. i (t) =θ i (*) (ii) a Otherwise: theta i (t) =θ i (t-1) ;
Step 1.4: until T = T;
step 1.5: substituting the sampling parameter sequence theta obtained by sampling into a corresponding blood pressure estimation model, and selecting a parameter value generating the minimum estimation error as an optimal model parameter;
and 2, step: the cuff sphygmomanometer is used for acquiring the pulse blood pressure of a human body as initial calibration blood pressure, namely prior data, uncertainty quantification is carried out on a model form through model selection and a model combination theory based on a Bayesian theory, and the quantified model is a combination of a plurality of models; the specific method comprises the following steps:
step 2.1: randomly extracting N groups of sample data P from prior probability of K candidate models by adopting Monte Carlo method i =(P i1 ,P i2 ,...,P iK ) I =1,2, N, wherein P ik ∈[0,1]K =1, 2.., K, and satisfies
Step 2.2: the model prior probability P (M) generated in step 2.1 is used to determine for each set of samples D the model prior probability P (M) using Bayes's theorem k ) And updating to obtain the posterior probability of the model:
in particular, when setting the model prior probability P (M) k ) Obeying to uniform distribution, i.e. the model probabilities are all equal, the above equation can be simplified as:
wherein, P (D | M) k ) Is M k The model likelihood of (3) represents the degree of closeness of the model prediction result and the experimental data D, and the expression is as follows:
where L is the sample size, M is the length of a single sample data, d m Is a sample value, f (M) k ) For estimating model M k Predicting a mean of the distribution;
step 2.3: and 3, performing weighted average on the model by taking the posterior probability of the model obtained in the step 2.2 as weight to obtain an expression of the Bayesian combination model, and determining Bayesian combination prediction distribution:
whereinAn expression representing a bayesian combinatorial model,representation model M i Expression of (c), P (M) i D) represents the model M obtained with a known set of experimental data D i A posterior probability of (d);
step 2.4: calculating the average absolute error between the N groups of Bayes combined prediction distribution and the reference blood pressure, and taking the model or the model combination corresponding to the minimum group of prior probability in the N groups of average absolute errors obtained based on different prior probabilities as the current blood pressure calculation model;
the blood pressure calculating module comprises the following steps: calculating the current blood pressure according to the current blood pressure calculation model and the uncertainty parameters determined by the model selection module, the initial calibration blood pressure and the characteristics obtained by measuring the initial calibration blood pressure and the initial calibration blood pressure simultaneously; and outputting the calculated blood pressure to a blood pressure display module.
Further, the model selection method in step 2 may further include:
step 2.1: a total of M models are set, and for each model k, there is an unknown parameter vector theta k The model prior probability is:and isGiven a set of experimental data D, the likelihood function of the data D is f (D | theta [ ]) k M = k), parameter θ k Is a priori p (theta) k | M = k), pseudo-a priori p (θ) k |M≠k);
step 2.3: fromMiddle extractionRepresenting the model M (i.e., M) in the t-1 th iteration for given data D (t-1) ) And model parameters of models other than model k in the t-1 th iterationTime theta k A conditional probability distribution of (a);
step 2.4: from p (theta) k'≠k |M (t-1) Extraction in = k)p(θ k'≠k |M (t-1) = k) means that given the condition of model k in the t-1 iteration,the conditional probability distribution of (a);
step 2.6: repeating the above steps T times to obtain enough samples, and calculating the posterior estimate of the model M = k as follows:
wherein I (M) (t) = k) is the likelihood function for model k in this iteration, and the corresponding bayesian factor is estimated as:
step 2.7: and selecting a model k corresponding to the maximum value of the Bayesian factor estimation as a model of the current blood pressure estimation.
By adopting the technical scheme, the invention has the following beneficial effects:
the wearable continuous blood pressure monitoring device breaks through the restriction and the discontinuous measurement of the traditional inflatable cuff, and can realize the wearable continuous blood pressure monitoring.
The uncertainty quantification model provided by the invention has better applicability, robustness and accuracy.
In the model uncertainty quantification, the model selection and model combination method based on Bayes can be used for continuously optimizing the model according to the new blood pressure calibration value.
The signal acquisition and processing unit can be completed by a micro-control chip, can realize miniaturized wearable or undisturbed continuous monitoring, and can be conveniently applied to personal, family and clinical monitoring.
The Bayesian combined prediction based on the Markov Monte Carlo simulation method is introduced into the cuff-free blood pressure prediction for the first time, and the method has the following advantages: when different conditional probability distributions are used, the algorithm is not required to be changed; the Bayesian combined prediction takes the posterior probability as weight to carry out weighted average on the selected single prediction model, and takes the posterior probability as a standard for judging the quality of the model, so that parameters with high average relevance of the Bayesian model can be processed, and meanwhile, the overall sample of the Bayesian combined prediction contains the predictions of different single models.
Drawings
FIG. 1 is a block diagram of the system of the present invention;
FIG. 2 is a flow chart of a Bayesian combination prediction method;
FIG. 3 is a flow chart of a Bayesian combinatorial prediction method;
FIG. 4 is a flow chart of the Metropolis-Hastings sampling algorithm.
Detailed Description
In order to achieve the purpose, the wearable electric, optical and pressure sensing device is used for measuring the information of relevant heart and artery pulsation of a certain part of a human body to obtain electrocardio and photoplethysmography pulse waves and body surface arterial pressure signals, and characteristic parameters relevant to blood pressure change in the signals are measured; and calculating the blood pressure value of the fluctuation of each beat of pulse by adopting a blood pressure estimation model and an uncertainty quantification method thereof according to the characteristic parameters and the current real blood pressure value.
The heart and artery pulse signal measuring part comprises a radial artery, a carotid artery or a superficial temporal artery;
the characteristic parameters extracted from the signals comprise the Time interval between the maximum rising edge of the Pulse wave and the peak value of the electrocardio R wave, namely Pulse Transit Time (PTT);
the characteristic parameters extracted from the signals include a pulse wave Intensity Ratio (PIR) which is a Ratio of peak to trough intensities of pulse waves of photoplethysmography;
the characteristic parameters extracted from the signals comprise time Width (PWHA) corresponding to Half Pulse Amplitude of Pulse wave of photoplethysmography;
the characteristic parameters extracted from the signals comprise photoplethysmography Amplitude (AM);
the wearable cuff-free continuous blood pressure measuring method specifically comprises the following steps:
selecting a body surface artery of a detected object, such as a radial artery, as a detected object;
step two, adopting cuff type blood pressure to carry out conventional measurement on the blood pressure of the body surface artery selected by the examinee, and taking the numerical value of the two measurements as a blood pressure calibration value;
step three, wearing wearable measuring equipment to the surface of the skin of the artery on the other side of the same examinee while measuring cuff type blood pressure;
step four, starting the wearable equipment, and recording signals generated by the heart and artery pulsation of the examinee;
step five, preprocessing the signals acquired in the step four, including noise filtering and amplification;
step six, calculating the real-time blood pressure value of each time of pulsation based on a cuff-free continuous blood pressure uncertainty quantification method according to the processed signals and the calibrated blood pressure value;
step seven, displaying the real-time blood pressure value once every a period of time, wherein the value is the mean value of all the beating blood pressure values in the corresponding period of time;
and step eight, displaying the calculation result on a wearable device display, transmitting the calculation result to terminals such as a mobile phone in a wireless mode, and transmitting the calculation result to the family members or medical institutions of the examinees for reference decision making through the mobile phone.
Further, the cuff-free continuous blood pressure measuring method based on the uncertainty quantification architecture in the sixth step includes the following steps:
(1) A plurality of blood pressure estimation models are used for obtaining a plurality of groups of blood pressure estimation values, and the models are as follows:
wherein, SBP 0 、DB 0 、MBP 0 And PP 0 For initial calibration of blood pressure, PIR0 and PTT0 correspond to an initial calibration pulse wave intensity ratio and a pulse wave propagation coefficient; gamma is a coefficient related to the state of blood vessels, and the value of gamma is between 0.016mmHg-1 and 0.018 mmHg-1; a and b are the corresponding model coefficients, respectively. Applying a multi-model prediction and quantification method of model uncertainty, comprising hierarchical sampling and a model optimization method based on Bayesian theory, and quantifying uncertainty of model parameters and uncertainty of model forms in the step (1) respectively;
(2) And obtaining the estimated value of the continuous blood pressure of the individual artery by measuring the wearable body surface characteristics by using the obtained optimal model parameters and the optimal model or the combination of the optimal models.
Further, the blood pressure estimation model parameters and form uncertainty quantification in the step (2) specifically comprises the following steps:
(2.1) carrying out uncertainty quantification on the model parameters based on a Monte Carlo-Markov chain sampling algorithm;
(2.2) carrying out uncertainty quantification on the model form through model selection and model combination theory based on Bayesian theory;
further, in the step (2.1), uncertainty of the model parameters is quantified by using a component form Metropolis-Hastings (M-H) sampling method, the sampling algorithm samples one-dimensional parameters each time, the known distribution can adopt univariate distribution, and the specific algorithm flow is as follows:
(2.1.1) initializing T =1, sampling times T =1000;
(2.1.2) setting relevant model parameters u = (u) 1 ,u 2 ,...,u N ) And initializing the parameter initial state Θ (t) = U, where U is an n-dimensional value randomly generated from a uniform distribution U (min, max);
(2.1.3) repeating the following process:
i. let t = t +1;
for each dimension model parameter: i =1,2, \ 8230;, N;
from the known distribution q (θ) i |θ i (t-1) ) To generate a new candidate state theta i (*) Assume that the entire vector before no update is:the vector after update is:
wherein q is a standard normal density function:u i is the initial value generated in step 2), and σ i The variance corresponding to the ith dimension parameter;
v. generating a random value a from the uniform distribution U (0, 1);
if a ≦ α, accepting the newly generated value: theta i (t) =θ i (*) (ii) a Otherwise: theta i (t) =θ i (t-1) ;
(2.1.4) until T = T.
(2.1.5) substituting the sampling parameter sequence obtained by sampling into the corresponding blood pressure estimation model, and selecting the parameter value generating the minimum estimation error as the optimal model parameter.
Further, the uncertainty quantification of the model form based on the combined model method (i.e., the model averaging method) in step (2.2) is specifically performed as shown in fig. 2, and the specific steps are as follows:
(2.2.1) applying Monte Carlo method to K candidate models (such as M) 1 ,M 2 ,…,M K ) Randomly extracting N groups of sample data P according to the prior probability i =(P i1 ,P i2 ,...,P iK ),i=1,2,...,N,Wherein P is ik ∈[0,1]K =1, 2.., K, and satisfies
(2.2.2) updating the model prior probability generated in the step (2.2.1) to obtain the posterior probability of the model by using Bayesian theorem and aiming at each group of samples D:
in particular, when setting the model prior probability P (M) k ) Subject to uniform distribution, i.e. the model probabilities are all equal, the above equation can be simplified to:
wherein, P (D | M) k ) Is M k The model likelihood of (3) represents the degree of closeness of the model prediction result and the experimental data D, and the expression is as follows:
where L is the sample size, M is the single sample data length, d is the sample value, f (M) k ) To estimate model M k The mean of the distribution is predicted.
(2.2.3) carrying out weighted average on the model according to the posterior probability of the model obtained in the step (2.2.2) as the weight to obtain an expression of the Bayes combination model, and determining Bayes combination prediction distribution:
whereinRepresenting a Bayesian composite modelThe expression (c) of (a),representation model M i Expression of (c), P (M) i D) represents the model M obtained with a known set of experimental data D i A posterior probability of (d);
(2.2.4) calculating the average absolute error between the N groups of Bayes combined prediction distribution and the reference blood pressure, and taking the prior probability of the minimum group of N groups of average absolute errors obtained based on different prior probabilities as a preset value of the prior probability, so that the Bayes combined prediction effect obtained at the moment is more accurate.
Further, in the step (2.2), based on the model form uncertainty quantification of the model selection method, a bayesian factor is used to select an optimal model, and the specific process is as follows:
1) Carrying out uncertainty quantification on the model parameters based on a Gibbs sampling algorithm;
2) Carrying out uncertainty quantification on the model form based on a Bayesian factor;
assuming a total of M models, for each model k, there is an unknown parameter vector θ k The prior probability of the model isAnd isA set of experimental data D is known, the likelihood function of which is f (Y | theta) k M = k), parameter θ k Is a priori p (theta) k | M = k), pseudo-a priori p (θ) k I M ≠ k). Since theta is calculated k There is no relation to M ≠ k, so we can choose an arbitrary form as the pseudo-prior.
Further, the distribution sampling flow of the correlation parameters in step 1) is shown in fig. 3, and the specific algorithm steps are as follows:
P(M=k|θ (t) ,D)=P k (k=1,...,M)
Repeat the above steps to get enough samples, then a posteriori estimate of model M = k is
further, in the step (2.2), the uncertainty of the model form is quantified, and the model k corresponding to the maximum value of the Bayesian factor estimation is selected as the optimal model of the current blood pressure estimation.
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will fall within the scope of the appended claims after reading the present invention.
As shown in fig. 1: a cuff-free blood pressure measurement model and method based on uncertainty comprise the following steps:
the method comprises the following steps: analyzing the correlation difference between the non-invasive blood pressure change characteristics and the blood pressure in different individuals, and evaluating the influence of the blood pressure change caused by the group and individual differences on the blood pressure estimation precision through blood pressure variability indexes (such as variation coefficients); then, expressing the uncertainty of the relation between different blood pressure change characteristics and the blood pressure; finally, the uncertainty propagation of the blood pressure change characteristics to the estimated blood pressure due to the differences of the model form and the model parameters is analyzed.
Step two: carrying out uncertainty quantification on the model parameters based on a Monte Carlo-Markov chain sampling algorithm; the blood pressure estimation model parameter distribution is sampled by adopting a Metropolis-Hastings (M-H) sampling method, and the specific algorithm steps are shown in the attached figure 3.
Step three: and carrying out uncertainty quantification on the model form through a model combination theory based on a Bayesian theory. The method comprises the following specific steps:
step four: and obtaining the estimated value of the continuous blood pressure of the individual artery by measuring the wearable body surface characteristics by using the obtained optimal model parameters and the optimal model or the combination of the optimal models.
Claims (2)
1. A wearable continuous blood pressure measurement system based on an uncertainty quantification strategy, the system comprising: the blood pressure monitoring device comprises a data acquisition device, a cuff sphygmomanometer, a data preprocessing module, a feature extraction module, a blood pressure calculation module and a blood pressure display module;
the data acquisition unit acquires a reflected light signal or a pressure signal of a certain part of a human body and acquires an electrocardiosignal at the same time; transmitting all the acquired signals to a data preprocessing module;
the cuff sphygmomanometer acquires the pulse blood pressure of a human body and transmits the blood pressure to the blood pressure calculation module;
the data preprocessing module carries out noise filtering and amplification on all signals acquired by the data acquisition device, and then converts the obtained reflected light signals or pressure signals into pulse waves; transmitting the preprocessed electrocardiosignals and the pulse wave signals obtained by conversion to a feature extraction module;
the features calculated by the feature extraction module include: time interval PTT between the maximum rising edge of the pulse wave and the R wave peak value of the electrocardiosignal; pulse wave intensity ratio PIR of the ratio of the peak intensity to the trough intensity of the pulse wave; the time width PWHA corresponding to the pulse wave half pulse amplitude; transmitting the calculated characteristics to a blood pressure calculation module;
the blood pressure calculation module comprises: a model memory, a model selection module,
The model memory stores 7 models which are respectively:
wherein, SBP 0 For initial calibration of systolic pressure, DBP 0 For initial calibration of diastolic pressure, MBP 0 For initial calibrationMean blood pressure, PIR 0 PTT for initial calibration of pulse wave intensity ratio 0 For initial calibration of the pulse wave propagation time, gamma is a factor related to the state of the blood vessel, a Mi And b Mi I =1,2, \8230, 7 are model coefficients corresponding to the models M1 to M7 respectively;
the parameter selection method in the model selection module comprises the following steps:
step 1: carrying out uncertainty quantification on each model parameter based on a Monte Carlo-Markov chain Metropolis-Hastings sampling algorithm; the following operations are performed for each model:
step 1.1: initializing T and sampling times T;
step 1.2: setting relevant model parameters u = (u) 1 ,u 2 ,...,u N ) And initializing the initial state theta of the parameter sequence (t) = U, where U is an N-dimensional value randomly generated from a uniform distribution U (min, max), and N is the number of uncertainty parameters in the model;
step 1.3: the following process was repeated:
i. let t = t +1;
for each dimension model parameter: i =1,2, \ 8230;, N;
known q (θ) i ) Obey a mean value of u i A normal distribution with variance of 1, with a probability density function of:from q (theta) i |θ i (t-1) ) To generate a new candidate state theta i (*) ,q(θ i |θ i (t-1) ) Given a previous state theta i (t-1) Current state of i The condition distribution of (2); let the whole vector before update be:the updated vector is then:
iv.calculating the acceptance probability:wherein p (Θ) represents the probability of the updated parameter vector Θ;
v. generating a random value a from the uniform distribution U (0, 1);
if a ≦ α, accepting the newly generated value: theta.theta. i (t) =θ i (*) (ii) a Otherwise: theta i (t) =θ i (t-1) ;
Step 1.4: until T = T;
step 1.5: substituting the sampling parameter sequence theta obtained by sampling into a corresponding blood pressure estimation model, and selecting a parameter value generating the minimum estimation error as an optimal model parameter;
and 2, step: the cuff sphygmomanometer is used for acquiring the pulse blood pressure of a human body as initial calibration blood pressure, namely prior data, uncertainty quantification is carried out on a model form through model selection and a model combination theory based on a Bayesian theory, and the quantified model is a combination of a plurality of models; the specific method comprises the following steps:
step 2.1: randomly extracting N groups of sample data P from prior probability of K candidate models by adopting Monte Carlo method i =(P i1 ,P i2 ,...,P iK ) I =1, 2.., N, wherein P is ik ∈[0,1]K =1, 2.., K, and satisfies
Step 2.2: the model prior probability P (M) generated in step 2.1 is used to determine for each set of samples D the model prior probability P (M) using Bayes's theorem k ) And updating to obtain the posterior probability of the model:
when setting the prior probability P (M) of the model k ) Obeying to uniform distribution, i.e. the model probabilities are all equal, the above equation is simplified as:
wherein, P (D | M) k ) Is M k The model likelihood of (2) is expressed by the degree of closeness of the model prediction result to the experimental data D, and the expression is as follows:
where L is the sample size, M is the length of a single sample data, d m Is a sample value, f (M) k ) To estimate model M k Predicting a mean of the distribution;
step 2.3: carrying out weighted average on the model according to the posterior probability of the model obtained in the step 2.2 as the weight to obtain an expression of the Bayes combination model, and determining Bayes combination prediction distribution:
whereinAn expression representing a bayesian combinatorial model,representation model M i Expression of (c), P (M) i | D) represents the model M obtained with a known set of experimental data D i A posterior probability of (d);
step 2.4: calculating the average absolute error between the N groups of Bayes combined prediction distribution and the reference blood pressure, and taking the model or the model combination corresponding to the minimum group of prior probability in the N groups of average absolute errors obtained based on different prior probabilities as the current blood pressure calculation model;
the blood pressure calculation module comprises the following steps: calculating the current blood pressure according to the current blood pressure calculation model and the uncertainty parameters determined by the model selection module, the initial calibration blood pressure and the characteristics obtained by measuring the initial calibration blood pressure and the initial calibration blood pressure simultaneously; and outputting the calculated blood pressure to a blood pressure display module.
2. The wearable continuous blood pressure measurement system based on the uncertainty quantification strategy of claim 1, wherein the model selection module in the blood pressure calculation module is further selected by:
step 2.1: a total of M models are set, and for each model k, there is an unknown parameter vector theta k The model prior probability is:and is provided withGiven a set of experimental data D, the likelihood function of the data D is f (D | theta [ ]) k M = k), parameter θ k Is a priori p (theta) k | M = k), pseudo-a priori p (θ) k |M≠k);
Step 2.2: assigning an initial value: m (t-1) =k,θ k (t-1) ,Superscript t-1 represents the number of iterations;
step 2.3: fromMiddle extracting Representing given data D, model M in the t-1 st iteration (i.e., M) (t-1) ) And the mode of other models except the model k in the t-1 th iterationForm factorTime theta k The conditional probability distribution of (a);
step 2.4: from p (theta) k'≠k |M (t-1) Extraction in = k)p(θ k'≠k |M (t-1) = k) denotes given model k in the t-1 th iteration,the conditional probability distribution of (a);
step 2.6: repeating the above steps T times to obtain samples, calculating a posterior estimate of model M = k as:
wherein I (M) (t) = k) is the likelihood function for model k in this iteration, and the corresponding bayesian factor estimate is:
step 2.7: and selecting a model k corresponding to the maximum value of the Bayesian factor estimation as a model of the current blood pressure estimation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110985640.2A CN113662520B (en) | 2021-08-26 | 2021-08-26 | Wearable continuous blood pressure measuring system based on uncertainty quantification strategy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110985640.2A CN113662520B (en) | 2021-08-26 | 2021-08-26 | Wearable continuous blood pressure measuring system based on uncertainty quantification strategy |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113662520A CN113662520A (en) | 2021-11-19 |
CN113662520B true CN113662520B (en) | 2022-10-14 |
Family
ID=78546435
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110985640.2A Active CN113662520B (en) | 2021-08-26 | 2021-08-26 | Wearable continuous blood pressure measuring system based on uncertainty quantification strategy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113662520B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114617536A (en) * | 2022-01-14 | 2022-06-14 | 深圳技术大学 | Blood pressure prediction method and device and computer readable storage medium |
CN115633947B (en) * | 2022-12-26 | 2023-03-21 | 四川大学华西医院 | Wearable blood pressure monitoring device and blood pressure monitoring method |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102499658A (en) * | 2011-11-08 | 2012-06-20 | 中国科学院深圳先进技术研究院 | Central blood pressure waveform reconstruction module and reconstruction method |
CN104757955A (en) * | 2015-03-25 | 2015-07-08 | 华中科技大学 | Human body blood pressure prediction method based on pulse wave |
CN106037694A (en) * | 2016-05-13 | 2016-10-26 | 吉林大学 | Continuous blood pressure measuring device based on pulse waves |
CN106343976A (en) * | 2016-09-14 | 2017-01-25 | 京东方科技集团股份有限公司 | Method and device for establishing blood pressure model as well as method and device for determining blood pressure |
CN106618540A (en) * | 2015-11-03 | 2017-05-10 | 深圳市维亿魄科技有限公司 | Non-invasive blood pressure detection method and device |
WO2017167013A1 (en) * | 2015-03-31 | 2017-10-05 | Vita-Course Technologies (Hainan) Co., Ltd | System and method for physiological feature derivation |
CN107361754A (en) * | 2017-08-19 | 2017-11-21 | 上海矩点医疗科技有限公司 | The construction method of ambulatory blood pressure monitoring model |
CN107530016A (en) * | 2015-04-20 | 2018-01-02 | 深圳市长桑技术有限公司 | A kind of physiology sign information acquisition methods and system |
WO2019028448A1 (en) * | 2017-08-04 | 2019-02-07 | The Johns Hopkins University | An application for early prediction of pending septic shock |
CN109965861A (en) * | 2019-04-16 | 2019-07-05 | 重庆大学 | Continuous monitoring device when a kind of wearable non-invasive blood pressure of no cuff is long |
EP3760123A1 (en) * | 2019-07-05 | 2021-01-06 | IMEC vzw | A system for estimating a stress condition of an individual |
CN112274127A (en) * | 2020-10-28 | 2021-01-29 | 河北工业大学 | Noninvasive continuous blood pressure detection method and device based on one-way pulse wave |
CN112274126A (en) * | 2020-10-28 | 2021-01-29 | 河北工业大学 | Noninvasive continuous blood pressure detection method and device based on multiple pulse waves |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10575760B2 (en) * | 2014-03-26 | 2020-03-03 | GestureLogic Inc. | Systems, methods and devices for activity recognition |
US20190142286A1 (en) * | 2014-03-31 | 2019-05-16 | Sensogram Technologies, Inc. | Photoplethysmographic wearable blood pressure monitoring system and methods |
US10039455B2 (en) * | 2014-05-19 | 2018-08-07 | Qualcomm Incorporated | Continuous calibration of a blood pressure measurement device |
EP3033991B1 (en) * | 2014-12-15 | 2018-02-28 | Stichting IMEC Nederland | System and method for blood pressure estimation |
CN104856661A (en) * | 2015-05-11 | 2015-08-26 | 北京航空航天大学 | Wearable continuous blood pressure estimating system and method based on dynamic compensation of diastolic blood pressure |
WO2019185392A1 (en) * | 2018-03-30 | 2019-10-03 | Koninklijke Philips N.V. | System and method for non-invasive determination of blood pressure dip based on trained prediction models |
US20210100455A1 (en) * | 2018-04-13 | 2021-04-08 | Vita-Course Technologies Co., Ltd. | Systems and methods for determining blood pressure of subject |
CN108498089B (en) * | 2018-05-08 | 2022-03-25 | 北京邮电大学 | Noninvasive continuous blood pressure measuring method based on deep neural network |
-
2021
- 2021-08-26 CN CN202110985640.2A patent/CN113662520B/en active Active
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102499658A (en) * | 2011-11-08 | 2012-06-20 | 中国科学院深圳先进技术研究院 | Central blood pressure waveform reconstruction module and reconstruction method |
CN104757955A (en) * | 2015-03-25 | 2015-07-08 | 华中科技大学 | Human body blood pressure prediction method based on pulse wave |
WO2017167013A1 (en) * | 2015-03-31 | 2017-10-05 | Vita-Course Technologies (Hainan) Co., Ltd | System and method for physiological feature derivation |
CN107530016A (en) * | 2015-04-20 | 2018-01-02 | 深圳市长桑技术有限公司 | A kind of physiology sign information acquisition methods and system |
CN106618540A (en) * | 2015-11-03 | 2017-05-10 | 深圳市维亿魄科技有限公司 | Non-invasive blood pressure detection method and device |
CN106037694A (en) * | 2016-05-13 | 2016-10-26 | 吉林大学 | Continuous blood pressure measuring device based on pulse waves |
CN106343976A (en) * | 2016-09-14 | 2017-01-25 | 京东方科技集团股份有限公司 | Method and device for establishing blood pressure model as well as method and device for determining blood pressure |
WO2019028448A1 (en) * | 2017-08-04 | 2019-02-07 | The Johns Hopkins University | An application for early prediction of pending septic shock |
CN107361754A (en) * | 2017-08-19 | 2017-11-21 | 上海矩点医疗科技有限公司 | The construction method of ambulatory blood pressure monitoring model |
CN109965861A (en) * | 2019-04-16 | 2019-07-05 | 重庆大学 | Continuous monitoring device when a kind of wearable non-invasive blood pressure of no cuff is long |
EP3760123A1 (en) * | 2019-07-05 | 2021-01-06 | IMEC vzw | A system for estimating a stress condition of an individual |
CN112274127A (en) * | 2020-10-28 | 2021-01-29 | 河北工业大学 | Noninvasive continuous blood pressure detection method and device based on one-way pulse wave |
CN112274126A (en) * | 2020-10-28 | 2021-01-29 | 河北工业大学 | Noninvasive continuous blood pressure detection method and device based on multiple pulse waves |
Non-Patent Citations (6)
Title |
---|
基于Android平台的无创血压连续监测系统;游玲等;《中国生物医学工程学报》;20170820(第04期);116-120 * |
基于光电容积脉搏波和机器学习方法的血压预测系统研究;谭志坚等;《医疗装备》;20200615(第11期);34-36 * |
基于脉搏波传导时间和脉搏波特征参数的连续血压无创检测;赵彦峰等;《生物医学工程研究》;20180315(第01期);47-50 * |
基于脉搏波特征参数的无创连续血压监测;谢寒霜等;《中国医疗设备》;20171025(第10期);52-56 * |
无创血压测量技术的发展概况;于潇等;《广东医学》;20120810(第15期);2356-2359 * |
脉搏波波速法无创血压测量中多模量血压计算模型研究;顾亚雄等;《中国生物医学工程学报》;20161220(第06期);691-698 * |
Also Published As
Publication number | Publication date |
---|---|
CN113662520A (en) | 2021-11-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108498089B (en) | Noninvasive continuous blood pressure measuring method based on deep neural network | |
US5797395A (en) | Continuous cardiac output derived from arterial pressure waveform using pattern recognition | |
Dey et al. | InstaBP: cuff-less blood pressure monitoring on smartphone using single PPG sensor | |
CN102429649B (en) | Continuous blood pressure measuring device | |
Rong et al. | A blood pressure prediction method based on imaging photoplethysmography in combination with machine learning | |
CN113662520B (en) | Wearable continuous blood pressure measuring system based on uncertainty quantification strategy | |
CN112274126A (en) | Noninvasive continuous blood pressure detection method and device based on multiple pulse waves | |
US20100274102A1 (en) | Processing Physiological Sensor Data Using a Physiological Model Combined with a Probabilistic Processor | |
Natarajan et al. | Photoplethysmography fast upstroke time intervals can be useful features for cuff-less measurement of blood pressure changes in humans | |
CN109730663B (en) | Blood pressure evaluation method based on pulse wave conduction velocity nonlinear analysis | |
WO2018072175A1 (en) | Apparatus and method for determining a blood pressure of a subject | |
CN111631698A (en) | Wearable blood pressure monitoring and correcting method based on motion mode cascade constraint | |
Miao et al. | A wearable sensor for arterial stiffness monitoring based on machine learning algorithms | |
CN113303776B (en) | Non-contact blood pressure measurement method based on circulating neural network | |
Chakraborty et al. | Measurement of arterial blood pressure through single-site acquisition of photoplethysmograph signal | |
CN110558960A (en) | continuous blood pressure non-invasive monitoring method based on PTT and MIV-GA-SVR | |
Shao et al. | An optimization study of estimating blood pressure models based on pulse arrival time for continuous monitoring | |
CN112274120A (en) | Noninvasive arteriosclerosis detection method and device based on one-way pulse wave | |
Roy et al. | BePCon: A photoplethysmography-based quality-aware continuous beat-to-beat blood pressure measurement technique using deep learning | |
US20230172565A1 (en) | Systems, devices, and methods for developing a model for use when performing oximetry and/or pulse oximetry and systems, devices, and methods for using a fetal oximetry model to determine a fetal oximetry value | |
CN114145725B (en) | PPG sampling rate estimation method based on noninvasive continuous blood pressure measurement | |
CN115486824A (en) | Cuff-free continuous blood pressure estimation system based on uncertainty measurement | |
CN114587310A (en) | Method for realizing invasive blood pressure waveform estimation based on photoplethysmography | |
Ray et al. | DeepPulse: An Uncertainty-aware Deep Neural Network for Heart Rate Estimations from Wrist-worn Photoplethysmography | |
Forouzanfar et al. | Bayesian fusion algorithm for improved oscillometric blood pressure estimation |
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 |