CN114947782A - System and method for central arterial pressure waveform reconstruction based on PPG signal - Google Patents

System and method for central arterial pressure waveform reconstruction based on PPG signal Download PDF

Info

Publication number
CN114947782A
CN114947782A CN202210648435.1A CN202210648435A CN114947782A CN 114947782 A CN114947782 A CN 114947782A CN 202210648435 A CN202210648435 A CN 202210648435A CN 114947782 A CN114947782 A CN 114947782A
Authority
CN
China
Prior art keywords
waveform
central
model
pressure
pressure waveform
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210648435.1A
Other languages
Chinese (zh)
Inventor
肖汉光
宋旺旺
刘畅
彭波
夏清玲
姜彬
彭滔
蒋鑫
朱秘
李艳梅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Technology
Original Assignee
Chongqing University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Technology filed Critical Chongqing University of Technology
Priority to CN202210648435.1A priority Critical patent/CN114947782A/en
Publication of CN114947782A publication Critical patent/CN114947782A/en
Pending legal-status Critical Current

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
    • 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
    • A61B5/6813—Specially adapted to be attached to a specific body part
    • A61B5/6825—Hand
    • A61B5/6826—Finger
    • 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
    • A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2562/00—Details of sensors; Constructional details of sensor housings or probes; Accessories for sensors
    • A61B2562/02—Details of sensors specially adapted for in-vivo measurements
    • A61B2562/0233—Special features of optical sensors or probes classified in A61B5/00
    • A61B2562/0238—Optical sensor arrangements for performing transmission measurements on body tissue

Abstract

The invention discloses a system and a method for central arterial pressure waveform reconstruction based on PPG signals, and relates to the technical field of artificial intelligence and medical instrument research and development. The device comprises a data acquisition control module, a data processing module, a radial artery pressure measuring module, a central artery pressure measuring module and a data display module; the data acquisition control module consists of a fingerstall type photoelectric sensor, a lead and a physiological signal acquisition circuit, and is used for controlling other modules to carry out sequential measurement and processing, and radial pulse signals reflect and convert optical signals into PPG signals through the fingerstall type photoelectric sensor. Through a series of improvements, characteristics do not need to be extracted manually, an intermediate simulation model does not need to be established and parameter estimation thereof are not needed when the model is used, an end-to-end reconstruction model from peripheral blood pressure PPG signals to radial artery pulse waves and then to central artery pressure is established, and reconstruction precision of central artery pressure waveforms is effectively improved.

Description

System and method for central arterial pressure waveform reconstruction based on PPG signal
Technical Field
The invention belongs to the technical field of artificial intelligence and medical instrument research and development, and particularly relates to a system and a method for central arterial pressure waveform reconstruction based on PPG signals.
Background
Cardiovascular diseases are systemic vasculopathy or the manifestation of systemic vasculopathy in the heart, are the important reasons of death or disability of human beings at present, and in recent years, with the acceleration of the life rhythm of people, various stresses are prominent, bad working habits are developed, and the risk of suffering from cardiovascular diseases is continuously increased; the health condition of the cardiovascular system can be accurately mastered, and the incidence and mortality of cardiovascular events can be effectively reduced; the central artery pressure refers to the lateral pressure born by the blood vessel at the root of the ascending aorta and is the root of blood perfusion of important organs; the central arterial pressure waveform is a curve of the ascending aorta root blood pressure changing along with time, contains rich physiopathological information, and the extracted central arterial pressure waveform parameters can accurately reflect important parameter indexes such as systolic pressure, diastolic pressure, pulse pressure, enhancement index, ejection time and the like; research shows that compared with peripheral blood pressure such as brachial artery and the like, central arterial pressure can more directly and accurately reflect load conditions of left ventricle, coronary artery and cerebral vessels, and has independent and stronger prediction value of cardiovascular diseases and related complications; therefore, focusing and reducing central arterial pressure will help prevent cardiovascular events, and waveform measurements for central arterial pressure are of great value in clinical applications;
with the continuous development of computer artificial intelligence technology, deep updating is taken as an important branch in the field of artificial intelligence, and the method is widely applied to multiple fields of voice recognition, machine vision, image processing, signal processing, natural language processing and the like; in order to improve the reconstruction precision and generalization capability of the central arterial pressure, students such as Huttunen try to use a machine to update training data to predict the aortic pulse wave conduction rate and the blood pressure pulse transmission time for the first time, so that a good prediction effect is obtained;
photo Plethysmography (PPG) is based on an LED light source and a detector, measures attenuation light after reflection and absorption of blood vessels and tissues of a human body, records the pulsation state of the blood vessels and measures pulse waves; a great deal of research shows that the PPG signal has high correlation with the blood pressure signal in the time domain and the frequency domain, and a blood pressure waveform can be reconstructed through the PPG signal, wherein a blood pressure reconstruction method by a photoplethysmography method is divided into three methods, namely Pulse Transit Time (PTT), Pulse Wave Velocity (PWV) and Pulse Wave Analysis (PWA); wherein PTT is the calculation of the velocity of blood pressure propagation in the artery from the PPG and ECG signals; PWV is PPG through two artery monitoring points to calculate the pulse wave transit time between them; since PWV only requires the PPG signal, clinical performance tends to be better than PTT; the PWA is a regression algorithm for training PPG signals and blood pressure signals through a neural network, which is receiving more and more attention; therefore, researches on detecting blood pressure signals by utilizing PPG signals updated by a machine are more and more, the blood pressure signals are often complex and specific due to the influence of various factors, and the traditional blood pressure characteristic extraction mode designed manually is gradually difficult to meet the requirements of modern medical treatment on blood pressure detection and diagnosis and treatment; to this end, we have devised a system and method for central arterial pressure waveform reconstruction based on PPG signals.
Disclosure of Invention
The present invention is directed to a system and a method for central arterial pressure waveform reconstruction based on PPG signals, so as to solve the problems mentioned in the above background art.
In order to solve the technical problems, the invention is realized by the following technical scheme:
the invention relates to a system and a method for central arterial pressure waveform reconstruction based on PPG signals, which comprises a data acquisition control module, a data processing module, a radial arterial pressure measuring module, a central arterial pressure measuring module and a data display module;
the data acquisition control module consists of a fingerstall type photoelectric sensor, a lead and a physiological signal acquisition circuit and controls other modules to perform sequential measurement and processing;
the data processing module is responsible for receiving signals of the data acquisition module, preprocessing the signals and sending the data to the designated module for further processing;
the radial artery pressure measurement module receives the processed PPG signal sent by the data processing module and obtains a reconstructed waveform of the radial artery pressure through CBi-SAN neural network calculation;
the central arterial pressure measuring module receives the radial arterial pressure waveform data measured by the radial arterial pressure measuring module and obtains a reconstructed waveform of the central arterial pressure through CBi-SAN neural network calculation;
and the data display module receives and displays the reconstructed radial artery pressure waveform output by the radial artery pressure measurement module and the reconstructed central artery pressure waveform and the basic parameters output by the central artery pressure measurement module.
A method for central arterial pressure waveform reconstruction based on PPG signals, for the above item, the steps are as follows:
constructing a CBi-SAN model based on a plurality of groups of independent one-dimensional CNN, Bi-LSTM, a self-attention mechanism and a fully-connected neural network;
outputting a reconstructed radial artery pulse waveform by taking the PPG signal as input, outputting a reconstructed central artery pressure waveform component into a CBi-SAN model by taking the radial artery pulse waveform as input, and realizing end-to-end reconstruction from the PPG signal to the central artery pressure waveform;
the radial artery pressure waveform signal and the cardiac pulse pressure waveform signal are respectively used as the input and the output of the model, and the radial artery pressure signal is sent into a plurality of layers of mutually independent one-dimensional convolution units Conv and then is processed by batch normalization Bn and an activation function Sule;
then sending the data to a bidirectional long-time memory network to update the global characteristics; the self-attention mechanism further strengthens the updating capability of the model to the local features through the adjustment of the weight; and finally outputting the predicted central arterial pressure waveform data through the full-connection layer.
Furthermore, in the central arterial pressure model based on the CBi-SAN, the convolution neural network is adopted to improve the capacity of the model for updating local features, the bidirectional long-time memory network is adopted to improve the capacity for updating global features, and the model updating capacity is strengthened through a self-attention mechanism.
Further, a central arterial pressure waveform is reconstructed, and the convolutional neural network can be updated from the blood pressure waveform of a long time period; in each pulse period of the central arterial pressure, the blood pressure waveform provides a large number of physiological characteristics, including important characteristics of central arterial systolic pressure, diastolic pressure, a second peak value and ejection time, and a circulating neural network cannot effectively update the central arterial systolic pressure, the diastolic pressure, the second peak value and the ejection time, and a one-dimensional convolutional neural network can not only update global characteristics from the central systolic pressure, but also enhance the processing capability of a network model on the local characteristics, grasp the correlation between adjacent positions of a time sequence and effectively improve the waveform reconstruction effect;
the feature extractor of the convolutional neural network is composed of a convolutional layer and a sub-sampling layer, wherein the convolutional layer realizes the updating of input features and the mapping of data from a low-dimensional space to a high-dimensional space; the sub-sampling layer reduces the dimensionality and redundancy of data through down-sampling, thereby compressing characteristic dimensionality, realizing the reduction of the complexity of a model and reducing the operation time;
the blood pressure waveform as a continuous time sequence; the Bi-LSTM is used as a bidirectional long-time memory network and is combined with the forward-propagation LSTM and the backward-propagation LSTM, and characteristic values in two directions can be obtained; the output h of the Bi-LSTM therefore includes characteristic information of the context of the waveform.
Further, the characteristic sequence X ═ X (X) for the convolutional neural network output 1 ,x 2 ,...,x N ) X belongs to N, wherein N is the length of the input characteristic sequence; information C of last moment t-1 ，C∈[0,1]And output h t-1 Input x at the same time t Inputting into LSTM cell to form new cell state C t And output h t (ii) a The calculation process is as follows:
C t ＝σ(W f ·[h t-1 ,x t ]+b f )*C t-1 +σ(W i ·[h t-1 ,x t ]+b i )*tanh(W C ·[h t-1 ,x t ]+b C )
h t ＝σ(W o ·[h t-1 ,x t ]+b o )*tanh(C t )
w and b are weights and bias vector corresponding to the hidden layer, cell states at the time t are updated through an activation function sigmoid and tanh, and output h at the time t is determined t 。
Further, the calculation of the self-attention mechanism is divided into two steps: firstly, calculating attention distribution on all input information, and then calculating weighted average of the input information according to the attention distribution; the self-attention model is a model for obtaining the output of the network layer by dynamically generating weights of different connections between the input and the output of the same network layer by using a self-attention mechanism; in a CBi-SAN network model, important information is searched, meanwhile, a self-attention mechanism model is used for establishing a long-distance dependency relationship in a sequence, weights of different connections are dynamically generated for different sections of input data, the number and the size of the weights are variable, and the problems that a cyclic neural network is poor in long-sequence data memory and the like are effectively improved.
Furthermore, the CBi-SAN neural network is formed by connecting a plurality of groups of independent one-dimensional CNN convolution layers, Bi-LSTM, a self-attention mechanism network and a full-connection layer in series; input X of CBi-SAN neural network t Respectively obtaining a PPG signal and a radial artery pressure waveform signal in sequence, obtaining a feature tensor containing signal waveform local features through processing of batch normalization Bn and an activation function Sule through multi-group parallel one-dimensional CNN convolution operation, and obtaining waveform feature X 'through tensor superposition realized by a Concat function' m (ii) a Then sent to Bi-LSTM to update the global characteristics of the blood pressure waveform, and thenFurther strengthening the updating capability of the model to local characteristics by adjusting the weight through a self-attention mechanism, and finally obtaining a reconstructed radial artery pressure reconstruction waveform and a reconstructed central artery pressure reconstruction waveform Y through a full connection layer t 。
The invention has the following beneficial effects:
1. through a series of improvements, characteristics do not need to be extracted manually, an intermediate simulation model does not need to be established and parameter estimation thereof is not needed when the model is used, an end-to-end reconstruction model from peripheral blood pressure PPG signals to radial artery pulse waves and then to central artery pressure is established, the radial artery waveform is convenient to extract, and the reconstruction precision of the central artery pressure waveform is effectively improved.
2. According to the invention, the artificial neural network structure is improved, the network model is optimized, the updating capability of the network on waveform characteristics is improved, the CBi-SAN neural network updates and acquires the front and back correlation information in the waveform sequence by using the bidirectional long-time memory neural network, the problems of poor memory of the network on long sequence data and the like are improved by using the self-attention mechanism module, the performance of a deep updating model is effectively improved, compared with other artificial neural networks, the network does not need to extract a radial artery signal first, can directly act on a PPG signal, and has stronger updating capability on a blood pressure waveform, higher generalization capability, higher precision improvement and better reconstruction effect on the central artery pressure.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a block diagram of a device for measuring central blood pressure according to the present invention;
FIG. 2 is a structural diagram of central arterial pressure reconstruction based on CBi-SAN according to the present invention;
FIG. 3 is a graph showing the effect of the central blood pressure waveform measurement of the present invention;
FIG. 4 is a schematic diagram of the internal structure of XX of the present invention;
FIG. 5 is a graph showing the effect of the central blood pressure waveform measurement of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The first embodiment is as follows:
referring to fig. 1-5, the present invention is a system and method for central artery pressure waveform reconstruction based on PPG signals.
A system for central arterial pressure waveform reconstruction based on PPG signals.
The system comprises a data acquisition control module, a data processing module, a radial artery pressure measuring module, a central artery pressure measuring module and a data display module, wherein the structural block diagram of the central artery pressure measuring module is shown in figure 1;
the data acquisition control module consists of a fingerstall type photoelectric sensor, a lead and a physiological signal acquisition circuit and controls other modules to carry out sequential measurement and processing, wherein the radial pulse signal reflects an optical signal through the fingerstall type photoelectric sensor and is converted into a PPG signal (electric signal), the filtering, amplification and other processing of the PPG signal are completed through a low-pass filter and an operational amplifier by a blood pressure signal processing circuit, and the processed signal is transmitted to the data processing module through the data acquisition control module;
the data processing module is responsible for receiving signals of the data acquisition module and preprocessing the signals, denoising and preprocessing the PPG signals collected by the data acquisition module through a filter, and transmitting the signals to the radial artery pressure measurement module and the data display module;
the radial artery pressure measurement module receives the processed PPG signal sent by the data processing module and obtains a reconstructed waveform of the radial artery pressure through CBi-SAN neural network calculation; sending the reconstructed radial artery pressure waveform data to a central artery pressure measuring module and a data display module;
the central arterial pressure measuring module receives the radial arterial pressure waveform data measured by the radial arterial pressure measuring module, and the reconstructed waveform of the central arterial pressure is obtained through CBi-SAN neural network calculation; calculating parameters such as central arterial systolic pressure, central arterial diastolic pressure, mean cardiac pulse pressure and the like, and then transmitting the reconstructed central arterial pressure waveform and basic parameters to a data display module;
the data display module receives and displays a reconstructed radial artery pressure waveform output by the radial artery pressure measurement module and a reconstructed central artery pressure waveform and basic parameters output by the central artery pressure measurement module;
example two:
a method for central arterial pressure waveform reconstruction based on PPG signals comprises the following steps:
outputting a reconstructed radial artery pulse waveform by taking a PPG signal as input, outputting a reconstructed central artery pressure waveform component into a CBi-SAN model by taking the radial artery pulse waveform as input, and automatically realizing end-to-end reconstruction from the PPG signal to the central artery pressure waveform;
the CBi-SAN model is shown in figure 2 and comprises a plurality of groups of independent one-dimensional CNN, Bi-LSTM, a self-attention mechanism and a fully-connected neural network, and can be divided into an input layer, a feature updating layer, a weight adjusting layer and a waveform reconstruction layer according to functions;
the radial artery pressure waveform signal and the cardiac pulse pressure waveform signal are respectively used as the input and the output of the model, after the radial artery pressure signal is sent into a plurality of layers of mutually independent one-dimensional convolution units Conv, the radial artery pressure signal is subjected to batch normalization Bn and activation function Sule, and finally, all data are superposed on channels, so that the updating of local characteristics is realized;
then, sending the data to a bidirectional long-time memory network to update the global characteristics; the self-attention mechanism further strengthens the updating capability of the model to the local features through the adjustment of the weight; finally, outputting predicted central arterial pressure waveform data through the full-connection layer;
in a central arterial pressure model based on CBi-SAN, the invention adopts a convolution neural network to improve the capability of the model for updating local characteristics, adopts a bidirectional long-time memory network to improve the capability of updating global characteristics, and strengthens the updating capability of the model through a self-attention mechanism;
for reconstructing the central arterial pressure waveform, the convolutional neural network may update features of interest from the blood pressure waveform for a long period of time; in each pulse period of the central arterial pressure, the blood pressure waveform provides a large number of physiological characteristics, including important characteristics such as central arterial systolic pressure, diastolic pressure, a second peak value, ejection time and the like, and the circulating neural network cannot effectively update the central arterial systolic pressure, while the one-dimensional convolutional neural network can not only update global characteristics from the central arterial pressure waveform, but also strengthen the processing capability of a network model on the local characteristics, grasp the correlation between adjacent positions of a time sequence and effectively improve the waveform reconstruction effect;
the convolutional neural network is an end-to-end supervised updating network, and the characteristic updating capability of the network is enhanced through a unique characteristic extractor; the feature extractor of the convolutional neural network is composed of a convolutional layer and a sub-sampling layer, wherein the convolutional layer realizes the updating of input features and the mapping of data from a low-dimensional space to a high-dimensional space; the sub-sampling layer can reduce the dimensionality and redundancy of data through down-sampling, thereby compressing the characteristic dimensionality, realizing the reduction of the complexity of a model and reducing the operation time; structurally, the convolutional neural network reduces the number of parameters in a local connection and weight sharing mode to improve the training speed, and updates the depth characteristics by utilizing multi-core combination, thereby improving the classification and regression performance;
the blood pressure waveform is taken as a continuous time sequence, and the fluctuation change of the waveform is closely related to the context; the invention adopts a Bi-directional long-short time memory network Bi-LSTM as the combination of forward-propagating LSTM and backward-propagating LSTM, and characteristic values in two directions can be obtained; therefore, the output h of the Bi-LSTM comprises the characteristic information of the context of the waveform, the problem that the forward LSTM network can only obtain the characteristic information is solved, and the updating effect on the time sequence is better;
the structure of the LSTM cell is shown in fig. 3:
characteristic sequence X ═ X (X) for convolutional neural network output 1 ,x 2 ,...,x N ) X belongs to N, wherein N is the length of the input characteristic sequence; information C of last moment t-1 ，C∈[0,1]And output h t-1 Input x at the same time t Inputting into LSTM cell to form new cell state C t And output h t (ii) a The calculation process is as follows:
C t ＝σ(W f ·[h t-1 ,x t ]+b f )*C t-1 +σ(W i ·[h t-1, x t ]+b i )*tanh(W C ·[h t-1, x t ]+b C ) (1)
h t ＝σ(W o ·[h t-1 ,x t ]+b o )*tanh(C t ) (2)
w and b are weight and bias vector corresponding to the hidden layer, cell state at the time t is updated through an activation function sigmoid and tanh, and output h at the time t is determined t ；
The self-attention mechanism focuses on information which is more critical to the current task in a plurality of input information, reduces the attention degree to other information, even filters out irrelevant information, can solve the problem of information overload, and improves the efficiency and the accuracy of task processing; the calculation of the attention value can be divided into two steps: firstly, calculating attention distribution on all input information, and then calculating weighted average of the input information according to the attention distribution; the self-attention model is a model for obtaining the output of the network layer by dynamically generating weights of different connections between the input and the output of the same network layer by using a self-attention mechanism; in the CBi-SAN network model, while searching important information, a self-attention mechanism model is used for establishing a long-distance dependency relationship in a sequence, weights of different connections are dynamically generated in different intervals of input data, the number and the size of the weights are variable, and the problems that a cyclic neural network is poor in long-sequence data memory and the like are effectively improved;
the final reconstitution effects are shown in fig. 4, the central dynamic pressure reconstitution effects of normal blood pressure and abnormal blood pressure, hypotension and hypertension, blood pressure at different frequencies, stable blood pressure and unstable blood pressure are shown;
the CBi-SAN neural network is formed by connecting a plurality of groups of independent one-dimensional CNN convolution layers, Bi-LSTM, a self-attention mechanism network and a full-connection layer in series; input X of CBi-SAN neural network t Respectively obtaining a PPG signal and a radial artery pressure waveform signal in sequence, obtaining a feature tensor containing signal waveform local features through processing of batch normalization Bn and an activation function Sule through multi-group parallel one-dimensional CNN convolution operation, and obtaining waveform feature X 'through tensor superposition realized by a Concat function' m (ii) a Then sending the data to Bi-LSTM to update the global characteristics of the blood pressure waveform, further enhancing the updating capability of the model to the local characteristics by adjusting the weight through an attention mechanism, and finally obtaining the reconstructed radial artery pressure reconstruction waveform and the reconstructed central artery pressure reconstruction waveform Y through a full connection layer in sequence t . In order to verify the effectiveness of CBi-SAN in radial artery reconstruction of a central artery model, the invention carries out comparison experiments on the model, a traditional method and other depth updating models under the same data set as well as the model of the invention, and three indexes of Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Spearman Correlation Coefficient (SCC) are adopted to evaluate the reconstruction effect of the central artery pressure.
TABLE 1 Performance comparison with conventional methods
Figure BDA0003686926740000111
As can be seen from Table 1, compared to the conventional method ARX, central arterial pressure waveform reconstruction based on the CBi-SAN model reduced the MAE from 2.47 + -26 mmHg to 2.23 + -0.11 mmHg, reduced the RMSE from 4.25 + -75 mmHg to 2.21 + -0.07 mmHg, reduced the RMSECASP from 4.66 + -2.47 to 2.94 + -0.48, reduced the RMSECADP from 3.28 + -2.30 to 1.96+0.06, increased the SCC coefficient from 0.71 to 0.99, and the predicted effect on Central Arterial Pressure (CAP) was superior to that of the ARX model. Compared with the conventional method NPMA, the RMSECASP is also reduced from 3.06 +/-0.32 to 2.94 +/-0.48, and the prediction effect on the Central Arterial Systolic Pressure (CASP) is better than that of the NPMA model.
TABLE 2 comparison of the Effect of the CBi-SAN model with different depth update methods
Figure BDA0003686926740000121
The CBi-SAN model shows excellent reconstruction effects in comparison of various depth update methods. Upon reconstruction of the central arterial pressure waveform, RMSE CBi-SAN Minimum value, SCC CBi-SAN Maximum, MAE CBi-SAN Although slightly higher than MAECNN-BilSTM, the difference is not large. This shows that the CBi-SAN model is slightly superior to the CNN-BilSTM model in reconstruction of central arterial pressure waveform, and the performance is not greatly improved. But in the detection of the systolic pressure and the diastolic pressure of the central artery, the CASPCbi-SAN and the CASDPCBi-SAN are both far lower than other depth updating models, and the reconstruction performance is obviously improved.
Example three:
a method for using a system for central artery pressure waveform reconstruction based on a PPG signal, the method being used in the first embodiment and the second embodiment as follows:
the first step is as follows: after a tester lies and has a rest for 5 minutes, the data acquisition control module starts and controls the fingerstall type photoelectric sensor to emit LED parallel light, and the receiving end converts received light signals which penetrate through human tissues and are absorbed into electric signals to complete measurement, filtering and amplification of radial artery blood pressure PPG signals;
the second step is that: the data processing module is started and controlled through the data acquisition control module, and denoising and drifting removal processing of the radial artery PPG signals is completed;
the third step: starting a radial artery pressure measurement module to enable a PPG signal to be processed through a CBi-SAN neural network and finish the reconstruction waveform of the radial artery pressure;
the fourth step: starting a central arterial pressure measuring module to enable the reconstructed radial arterial waveform to be processed through a CBi-SAN neural network, completing the reconstructed waveform of the central arterial pressure, and calculating basic parameters of the central arterial blood pressure waveform;
the fifth step: and displaying the PPG signal, the radial artery pressure waveform, the central artery pressure waveform and relevant parameters on a display screen, and simultaneously generating a measurement report.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims (7)

1. A system for central artery pressure waveform reconstruction based on PPG signals is characterized in that: the device comprises a data acquisition control module, a data processing module, a radial artery pressure measuring module, a central artery pressure measuring module and a data display module;
the data acquisition control module consists of a fingerstall type photoelectric sensor, a lead and a physiological signal acquisition circuit and controls other modules to perform sequential measurement and processing;
the data processing module is responsible for receiving signals of the data acquisition module, preprocessing the signals and sending the data to the designated module for further processing;
the radial artery pressure measurement module receives the processed PPG signal sent by the data processing module and obtains a reconstructed waveform of the radial artery pressure through CBi-SAN neural network calculation;
the central arterial pressure measuring module receives the radial arterial pressure waveform data measured by the radial arterial pressure measuring module, and a reconstructed waveform of the central arterial pressure is obtained through CBi-SAN neural network calculation;
and the data display module receives and displays the reconstructed radial artery pressure waveform output by the radial artery pressure measurement module and the reconstructed central artery pressure waveform and the basic parameters output by the central artery pressure measurement module.
2. A method for central arterial pressure waveform reconstruction based on PPG signals is characterized by comprising the following steps:
constructing a CBi-SAN model based on a plurality of groups of independent one-dimensional CNN, Bi-LSTM, a self-attention mechanism and a fully-connected neural network;
outputting a reconstructed radial artery pulse waveform by taking a PPG signal as input, outputting a reconstructed central artery pressure waveform component into a CBi-SAN model by taking the radial artery pulse waveform as input, and realizing end-to-end reconstruction from the PPG signal to the central artery pressure waveform;
the radial artery pressure waveform signal and the cardiac pulse pressure waveform signal are respectively used as the input and the output of the model, and the radial artery pressure signal is sent into a plurality of layers of mutually independent one-dimensional convolution units Conv and then is processed by batch normalization Bn and an activation function Sule;
then sending the data to a bidirectional long-time memory network to update the global characteristics; the self-attention mechanism further strengthens the updating capability of the model to the local features through the adjustment of the weight; and finally outputting the predicted central arterial pressure waveform data through the full-connection layer.
3. The method for central arterial pressure waveform reconstruction based on the PPG signal according to claim 2, wherein in the CBi-SAN based central arterial pressure model, a convolutional neural network is adopted to improve the capability of the model to update local features, a bidirectional long-time and short-time memory network is adopted to improve the capability to update global features, and the model updating capability is enhanced through a self-attention mechanism.
4. The method for central arterial pressure waveform reconstruction based on PPG signal according to claim 2, wherein the central arterial pressure waveform is reconstructed, and the convolutional neural network can be updated from the blood pressure waveform of a long time period; in each pulse period of the central arterial pressure, the blood pressure waveform provides a large number of physiological characteristics, including important characteristics of central arterial systolic pressure, diastolic pressure, a second peak value and ejection time, and a circulating neural network cannot effectively update the central arterial systolic pressure, the diastolic pressure, the second peak value and the ejection time, and a one-dimensional convolutional neural network can not only update global characteristics from the central systolic pressure, but also enhance the processing capability of a network model on the local characteristics, grasp the correlation between adjacent positions of a time sequence and effectively improve the waveform reconstruction effect;
the feature extractor of the convolutional neural network is composed of a convolutional layer and a sub-sampling layer, wherein the convolutional layer realizes the updating of input features and the mapping of data from a low-dimensional space to a high-dimensional space; the sub-sampling layer reduces the dimensionality and redundancy of data through down-sampling, thereby compressing characteristic dimensionality, realizing the reduction of the complexity of a model and reducing the operation time;
the blood pressure waveform as a continuous time sequence; the Bi-LSTM is used as a bidirectional long-time memory network and is combined with the forward-propagation LSTM and the backward-propagation LSTM, and characteristic values in two directions can be obtained; the output h of the Bi-LSTM therefore includes characteristic information of the context of the waveform.
5. The method for central arterial pressure waveform reconstruction based on PPG signal according to claim 4, wherein the characteristic sequence output by the convolutional neural network is X ═ X (X) 1 ,x 2 ,...,x N ) X belongs to N, wherein N is the length of the input characteristic sequence; information C of last moment t-1 ，C∈[0,1]And output h t-1 Input x at the same time t Inputting into LSTM cell to form new cell state C t And output h t (ii) a The calculation process is as follows:
C t ＝σ(W f ·[h t-1 ,x t ]+b f )*C t-1 +σ(W i ·[h t-1 ,x t ]+b i )*tanh(W C ·[h t-1 ,x t ]+b C )
h t ＝σ(W o ·[h t-1 ,x t ]+b o )*tanh(C t )
w and b are weight and bias vector corresponding to the hidden layer, cell state at the time t is updated through an activation function sigmoid and tanh, and output h at the time t is determined t 。
6. The method for central arterial pressure waveform reconstruction based on PPG signal according to claim 2, wherein the calculation of the self-attention mechanism is divided into two steps: firstly, calculating attention distribution on all input information, and then calculating the weighted average of the input information according to the attention distribution; the self-attention model is a model for obtaining the output of the network layer by dynamically generating weights of different connections between the input and the output of the same network layer by using a self-attention mechanism; in a CBi-SAN network model, important information is searched, meanwhile, a self-attention mechanism model is used for establishing a long-distance dependency relationship in a sequence, weights of different connections are dynamically generated for different sections of input data, the number and the size of the weights are variable, and the problems that a cyclic neural network is poor in long-sequence data memory and the like are effectively improved.
7. The method for central arterial pressure waveform reconstruction based on PPG signal according to claim 2, wherein the CBi-SAN neural network is formed by connecting multiple independent sets of one-dimensional CNN convolution layer, Bi-LSTM, self-attention mechanism network and full connection layer in series; input X of CBi-SAN neural network t Respectively obtaining a PPG signal and a radial artery pressure waveform signal in sequence, obtaining a feature tensor containing signal waveform local features through processing of batch normalization Bn and an activation function Sule through multi-group parallel one-dimensional CNN convolution operation, and obtaining waveform feature X 'through tensor superposition realized by a Concat function' m (ii) a Then sent to Bi-LSTM to update the global characteristics of the blood pressure waveform, and then sent to the Bi-LSTM to update the global characteristics of the blood pressure waveformThe self-attention mechanism further strengthens the updating capability of the model on local characteristics by adjusting the weight, and finally obtains a reconstructed radial artery pressure reconstruction waveform and a reconstructed central artery pressure reconstruction waveform Y through a full connection layer t 。
CN202210648435.1A 2022-06-09 2022-06-09 System and method for central arterial pressure waveform reconstruction based on PPG signal Pending CN114947782A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210648435.1A CN114947782A (en) 2022-06-09 2022-06-09 System and method for central arterial pressure waveform reconstruction based on PPG signal

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210648435.1A CN114947782A (en) 2022-06-09 2022-06-09 System and method for central arterial pressure waveform reconstruction based on PPG signal

Publications (1)

Publication Number Publication Date
CN114947782A true CN114947782A (en) 2022-08-30

Family

ID=82961915

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210648435.1A Pending CN114947782A (en) 2022-06-09 2022-06-09 System and method for central arterial pressure waveform reconstruction based on PPG signal

Country Status (1)

Country Link
CN (1) CN114947782A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115990007A (en) * 2022-11-30 2023-04-21 未来穿戴健康科技股份有限公司 Central artery pressure waveform fitting method, monitoring device and watch equipment

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115990007A (en) * 2022-11-30 2023-04-21 未来穿戴健康科技股份有限公司 Central artery pressure waveform fitting method, monitoring device and watch equipment
CN115990007B (en) * 2022-11-30 2024-01-23 未来穿戴健康科技股份有限公司 Central artery pressure waveform fitting method, monitoring device and watch equipment

Similar Documents

Publication Publication Date Title
Abbas et al. Phonocardiography signal processing
Tanveer et al. Cuffless blood pressure estimation from electrocardiogram and photoplethysmogram using waveform based ANN-LSTM network
US11298029B2 (en) Blood pressure measuring apparatus, blood pressure measuring method, electronic device, and computer readable storage medium
CN106413534B (en) Continuous blood pressure measuring device, measuring model establishing method and system
CN102429649B (en) Continuous blood pressure measuring device
CN102397064B (en) Continuous blood pressure measuring device
CN102488503B (en) Continuous blood pressure measurer
CN106821356B (en) Cloud continuous BP measurement method and system based on Elman neural network
CN113143230B (en) Peripheral arterial blood pressure waveform reconstruction system
WO2023185873A1 (en) Non-cuff blood pressure measuring apparatus based on multi-stage multi-modal learning and method
Argha et al. A novel automated blood pressure estimation algorithm using sequences of Korotkoff sounds
CN111839488B (en) Non-invasive continuous blood pressure measuring device and method based on pulse wave
CN110840428B (en) Noninvasive blood pressure estimation method based on one-dimensional U-Net network
CN114947782A (en) System and method for central arterial pressure waveform reconstruction based on PPG signal
CN113040738B (en) Blood pressure detecting device
Bhattacharjee Cuff-less blood pressure estimation from electrocardiogram and photoplethysmography based on VGG19-LSTM network
KR20220105092A (en) Continuous blood pressure measurement method by inputting the difference between electrocardiogram and the photoplethysmography signal into artificial neural network
WO2023240739A2 (en) Intelligent blood pressure prediction method based on multi-scale residual network and ppg signal
CN113499048B (en) Central arterial pressure waveform reconstruction system and method based on CNN-BilSTM
CN115836847A (en) Blood pressure prediction device and equipment
CN116138755A (en) Method for constructing model for noninvasive blood pressure monitoring and wearable device
CN115240845A (en) Coronary heart disease classification method, system and device based on external counterpulsation
CN112914528B (en) Model generation method, device and computer readable medium for cuff-free blood pressure measurement
Kim et al. 1-Dimensional Convolutional Neural Network Based Blood Pressure Estimation with Photo plethysmography Signals and Semi-Classical Signal Analysis
CN116649932A (en) Wearable continuous blood pressure measurement method and system based on removing age confusion factor

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