CN116803340B - Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network - Google Patents

Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network Download PDF

Info

Publication number
CN116803340B
CN116803340B CN202310885999.1A CN202310885999A CN116803340B CN 116803340 B CN116803340 B CN 116803340B CN 202310885999 A CN202310885999 A CN 202310885999A CN 116803340 B CN116803340 B CN 116803340B
Authority
CN
China
Prior art keywords
data
ppg
blood pressure
layer
graph
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
Application number
CN202310885999.1A
Other languages
Chinese (zh)
Other versions
CN116803340A (en
Inventor
史大威
卢一
刘一村
王军政
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.)
Beijing Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
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 Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN202310885999.1A priority Critical patent/CN116803340B/en
Publication of CN116803340A publication Critical patent/CN116803340A/en
Application granted granted Critical
Publication of CN116803340B publication Critical patent/CN116803340B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • A61B5/02—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
    • A61B5/02141—Details of apparatus construction, e.g. pump units or housings therefor, cuff pressurising systems, arrangements of fluid conduits or circuits
    • 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
    • A61B5/72—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
    • A61B5/02—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
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate
    • A61B5/02416—Detecting, measuring or recording pulse rate or heart rate using photoplethysmograph signals, e.g. generated by infrared radiation
    • 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
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate
    • A61B5/0245—Detecting, measuring or recording pulse rate or heart rate by using sensing means generating electric signals, i.e. ECG signals

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Cardiology (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Veterinary Medicine (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Physics & Mathematics (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Physiology (AREA)
  • Vascular Medicine (AREA)
  • Signal Processing (AREA)
  • Power Engineering (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Abstract

The invention discloses a noninvasive blood pressure detection method based on multi-source data fusion and a graph neural network, which comprises the following steps: s1, establishing a physiological data acquisition module and acquiring human PPG data and ECG data; s2, performing physiological signal compensation based on motion pseudo-noise suppression of measurement points and skin color difference suppression of wave bands; s3, constructing graph structure data after extracting waveform characteristics, performing model training according to a graph convolution neural network, and finally realizing accurate estimation of blood pressure data. The noninvasive blood pressure detection method based on the multi-source data fusion and the graph neural network improves the estimation accuracy of the blood pressure change trend and the robustness of long-term detection, and supports the long-term detection of blood pressure.

Description

Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network
Technical Field
The invention relates to the technical field of medical engineering fusion, in particular to a noninvasive blood pressure detection method based on multisource data fusion and a graph neural network.
Background
Blood pressure is an important indicator of cardiovascular health, and by detecting blood pressure, many cardiovascular diseases can be discovered and prevented early.
Invasive and noninvasive blood pressure detection algorithms are two different methods for measuring human blood pressure. Invasive blood pressure detection algorithms directly measure blood pressure by inserting a sensor in the arterial or venous vessel of the patient. Are commonly used in intensive care units and operating rooms to obtain high accuracy blood pressure data. The algorithm uses a pressure sensor to measure the pressure in the blood vessel and transmits the results to a monitor or computer for analysis and display. While invasive blood pressure detection algorithms provide accurate and real-time blood pressure data, the use of this method requires insertion into a blood vessel, possibly with the risk of infection and other complications.
The noninvasive blood pressure detection technology is widely applied to the scenes such as wards, operating rooms, emergency rooms, home health care and the like, and provides a convenient and quick blood pressure detection means for medical system personnel and individual users. Traditional noninvasive blood pressure detection algorithms calculate systolic and diastolic pressures by analyzing blood pressure signal changes based on cuff inflation and deflation processes. The noninvasive blood pressure detection algorithm is convenient, safe and easy to use, but cannot be detected in real time and is inconvenient to measure. To overcome these limitations, sleeveless blood pressure detection techniques have evolved to measure blood pressure using advanced sensor techniques and algorithms without the need for a cuff. In general, a device such as a photoelectric sensor, a pressure sensor or an acoustic sensor is used to detect a weak change caused by blood flow and acquire blood pressure data.
Sleeveless blood pressure detection using PPG (photoplethysmogram) and ECG (electrocardiogram) is a representative of this technique. The PPG sensor detects a change in blood flow by measuring a change in reflected light intensity of the skin surface using a photoelectric measurement principle; the ECG sensor records the electrical activity of the heart, including the information of systole and diastole, and by detecting the PPG and ECG signals simultaneously and combining with a new generation of AI processing algorithm, the real-time analysis and interpretation of blood pressure data can be realized. From the aspect of algorithm classification, the method is mainly divided into two major categories of traditional signal processing algorithm and machine learning/deep learning algorithm, and equipment products combining the two major categories are also available. Taking freeRTOS-based continuous noninvasive blood pressure measurement system (CN202110247131. X) as an example, the patent of Shanghai LiHetai medical science and technology Co., ltd, calculating the blood pressure of a person to be measured in real time according to an optimized traditional PWV calculation method and a regression equation generated by considering heart rate factors after data acquisition; in contrast, the patent of Beijing aviation aerospace university (CN 202211609290.0) only uses PPG data and combines with a new generation deep learning transducer algorithm, so that a part of estimation accuracy and the degree of association between various data are sacrificed in exchange for convenience of data acquisition.
The non-invasive blood pressure detection related equipment basically integrates the non-invasive blood pressure detection and has certain continuous detection capability. However, the problems of single input data, low detection precision and the like are not solved.
In summary, the following problems are presented in the conventional detection methods by understanding the blood pressure detection technology.
1. The current cuff-free blood pressure detection equipment is difficult to realize the design of use without sense or low sense in use;
2. On the basis of the blood pressure detection of the PPG and the ECG, the problems that the PPG acquisition is easily influenced by factors such as motion pseudo noise, skin color difference and the like and the ECG has insufficient data acquisition precision caused by insufficient input impedance of acquisition equipment exist;
3. The existing sleeveless blood pressure detection mostly adopts a curve fitting method based on pulse wave transmission time or an estimation method based on deep learning, and has the defects of poor detection precision, obvious individual difference, poor interpretability and the like;
4. In the long-term management of blood pressure, the current blood pressure detection equipment has the problems of inconvenient long-term data statistics, inaccurate blood pressure trend change, deep information of long-term blood pressure data to be mined and the like.
Disclosure of Invention
The invention aims to provide a noninvasive blood pressure detection method based on multi-source data fusion and a graph neural network, which improves the estimation accuracy of the blood pressure change trend and the robustness of long-term detection and supports the long-term detection of blood pressure.
In order to achieve the above purpose, the invention provides a noninvasive blood pressure detection method based on multi-source data fusion and a graph neural network, which comprises the following steps:
s1, establishing a physiological data acquisition module and acquiring human PPG data and ECG data;
S2, performing physiological signal compensation based on motion pseudo-noise suppression of measurement points and skin color difference suppression of wave bands;
s3, constructing graph structure data after extracting waveform characteristics, performing model training according to a graph convolution neural network, and finally realizing accurate estimation of blood pressure data.
Preferably, in step S1, the step of acquiring PPG data is as follows: firstly, an embedded multiband LED light source irradiates a tested part, each tissue of the tested part is absorbed to different degrees, the light absorption quantity changes along with the change of a substance, and a reflected transmitted light intensity signal is received by a plurality of photoelectric conversion sensors at points and converted into an electric signal, namely PPG data of different points at different wavebands is obtained;
the acquisition of ECG data requires the attachment of electrode pads to designated locations of the body.
Preferably, in step S2, the specific flow of the motion artifact suppression algorithm based on the measurement point location is as follows:
(1) For the PPG data (PPG A、PPGB、PPGC、PPGD) of 4 points (A, B, C, D is assumed to be four points) in the same wave band, the acquisition area is smaller, the difference of 7 human skin colors and the like in the measurement area is small, the correlation of different point data on the time domain and morphological characteristics is assumed to be high, in ideal case, the voltage waveform of the PPG data consists of reflected light, ambient light and instrument noise, and the amplitude of the PPG is expressed by the following formula
VPPG＝V Reflected light +V Ambient light +V Instrument noise (1)
Wherein V Reflected light is the voltage generated by the reflected light of the corresponding wave band, V Ambient light is the voltage generated by the light of the corresponding wave band existing in the environment, and V Instrument noise is the voltage noise generated by the measuring device;
(2) When the exercise happens, V Reflected light can change due to local muscle change and loose wearing, meanwhile, the phenomena of light leakage and the like of a measurement area can also change V Ambient light , and PPG data are related to two exercise pseudo-sounds;
(3) When there is no motion artifact, then V Reflected light ,V Ambient light is assumed to be substantially similar, then V PPG is derived from the following equation:
as can be seen from equation (2), when there is no motion artifact, the noise error of the instrument is reduced according to this manner.
Preferably, in step S2, the influence of two motion artifacts on PPG data is as follows:
(1) When the motion pseudo-noise 1 occurs, in the above formula (1), due to light leakage, the light intensity of the LED is reduced, and then V Reflected light is reduced, and simultaneously, the enhancement of the ambient light increases V Ambient light , and the parameter with the smallest front-back variation is reduced as the PPG acquisition value of the band;
(2) When motion artifact 2 occurs, V PPG of the four points is V A+ΔVA、VB+ΔVB、VC+ΔVC、VD+ΔVD, where V A、VB、VC、VD is V PPG of no motion artifact, and since Δv A<ΔVB、ΔVC<ΔVD exists in a linear relationship between the physical positions of a and B, C and D, and there is a functional relationship f between the waveform voltage change of the existence point O, PPG and the point between a and D, so that:
VO＝f((VB-VA),(VC-VD),VD,VA) (3)
The virtual point V O is PPG data of the wave band, and the reduction of motion pseudo noise is realized through the mapping relation.
Preferably, in step S2, the band-based skin color difference suppression algorithm is based on the difference in the penetration capability of different band lights to different tissue layers of the skin:
Wherein V Reflected light ,Vgreen、Vred、Vir_1、Vir_2 for the four bands is represented by the following formula:
wherein, V green is green light, V red is red light, V ir_1 and V ir_2 represent voltages generated by infrared light in different wave bands, when proper wearing and other measurement conditions are ideal, V Skin color is constant, and V Surface layer 、V Superficial layer in skin 、V Deep layer in skin 、V Deep layer of skin consists of tissue constant and vascular pulsation of the layer, namely:
V Layer(s) ＝V Layer structure +V blood vessel
And obtaining light reflection voltages of different tissue layers through a data difference algorithm, so as to achieve the effect of skin color difference inhibition.
Preferably, in step S3, the graph convolution neural network blood pressure estimation algorithm mainly comprises an input layer, a GCN layer, and a connection layer, and performs feature extraction on physiological data when the input layer, and performs model training by using GCN after constructing graph structure data by using the features, and sequentially performs GCN, pooling layer, and full connection layer processing on the graph structure data, specifically as follows:
(1) GCN layer: updating the characteristic representation of the node by calculating the weighted average value of the node and the neighbor node thereof;
(2) Pooling layer: the method is used for reducing the number of nodes in the graph convolution neural network, and keeping important information, and a pooling operation is used for aggregating a group of nodes into one node, so that the scale of the graph is reduced;
(3) Full tie layer: the full-connection layer is used for carrying out final feature processing and output prediction in the graph roll neural network, each node in the full-connection layer is connected with all nodes of the previous layer, each connection has weight, the full-connection layer carries out further feature extraction and combination on the outputs of the graph roll layer and the pooling layer to generate a final prediction result, and the full-connection layer comprises a ReLU nonlinear activation function to increase the expression capacity and nonlinear modeling capacity of the network.
Preferably, in step S3, the feature representation of the node is updated by calculating a weighted average of the node and its neighboring nodes, and the specific steps are as follows:
(a) The GCN layer is a core component of the graph convolution neural network and is used for carrying out feature propagation and learning on graph structure data;
(b) Feature propagation is performed by using neighbor information of nodes, so that the features of each node are affected by neighboring nodes;
Specifically, the output of the GCN layer is an updated node feature matrix, wherein the features of each node are updated to a new representation that comprehensively considers neighbor node information, wherein the formula of the GCN layer is:
where σ is the activation function, H l is the node feature matrix of the first layer, In the form of an adjacency matrix a plus a self-loop,Is in the form of a degree matrix D plus a self-loop, W l is the layer i weight matrix.
Preferably, in step S3, feature extraction of physiological data mainly includes: and extracting characteristic points of PPG of different wave bands for PPG data, wherein the characteristic points comprise wave crests and wave troughs, the wave form rises by 0.1 times of the wave crests, and the ECG signals are used for identifying P waves, Q waves, R waves, S waves and T waves.
Preferably, the physiological data acquisition module is additionally provided with high input impedance in the ECG acquisition circuit, so that the robustness of the acquisition module is enhanced, and the accurate acquisition of ECG data is realized.
Therefore, the non-invasive blood pressure detection method based on multi-source data fusion and graph neural network has the following technical effects:
(1) The measuring module integrally adopts a wristband type structure, and is worn for a long time by a user; the acquisition part is designed to resist movement pseudo noise, is tightly connected with the wrist at a plurality of photoelectric conversion positions, can be suitable for wearable equipment such as a bracelet and the like, and has small interference to normal life.
(2) The invention adopts a technical route of simultaneously collecting 4-band PPG signals and single-lead ECG signals. According to the invention, data fusion is carried out according to the data of different points, so that motion artifact resistance is realized; according to the principle that light of different wave bands has different penetrability to skin, difference fusion is carried out on multi-wave band data, and inhibition of skin color influence is achieved. And a high input impedance design is added to a traditional ECG acquisition circuit, so that the ECG acquisition precision is improved.
(3) In the implementation of blood pressure estimation, the method of adopting the multi-modal data and the graph neural network algorithm is adopted, and compared with the single-channel PPG and single-lead ECG adopted in traditional sleeveless blood pressure detection for calculating PWTT and then fitting an empirical formula or an estimation method based on deep learning, the technical route of the invention can better dig possible dependency relationship among multiple physiological data and improve detection precision.
(4) According to the method, the information is transmitted to the cloud end while the blood pressure data is estimated, the data mining algorithm is deployed in the cloud end, and an experience playback mechanism is utilized, so that the accuracy of estimating the blood pressure change trend and the robustness of long-term detection can be improved, and the long-term detection of the blood pressure is supported.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
FIG. 1 is a hardware diagram of a physiological data acquisition module;
FIG. 2 is a graph of a blood pressure estimation algorithm based on the neural network of the graph;
FIG. 3 is a top view of a physiological data acquisition module hardware acquisition architecture;
FIG. 4 is a side view of a physiological data acquisition module hardware acquisition architecture;
FIG. 5 is a technical route of a non-invasive blood pressure detection method based on multi-source data fusion and a graph neural network;
FIG. 6 is a motion artifact generation scenario;
FIG. 7 is a physiological signal staging and generation principle;
Fig. 8 is diagram structural data.
Detailed Description
The technical scheme of the invention is further described below through the attached drawings and the embodiments.
Unless defined otherwise, technical or scientific terms used herein should be given the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Furthermore, it should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a separate embodiment, and that this description is provided for clarity only, and that the disclosure is not limited to the embodiments described in detail below, and that the embodiments described in the examples may be combined as appropriate to form other embodiments that will be apparent to those skilled in the art. Such other embodiments are also within the scope of the present invention.
It should also be understood that the above-mentioned embodiments are only for explaining the present invention, the protection scope of the present invention is not limited thereto, and any person skilled in the art should be able to cover the protection scope of the present invention by equally replacing or changing the technical scheme and the inventive concept thereof within the scope of the present invention.
Techniques, methods, and apparatus known to one of ordinary skill in the relevant art may not be discussed in detail, but are intended to be considered part of the specification where appropriate.
The disclosures of the prior art documents cited in the present specification are incorporated by reference in their entirety into the present invention and are therefore part of the present disclosure.
Example 1
As shown in the figure, the invention provides a noninvasive blood pressure detection method based on multi-source data fusion and a graph neural network, which comprises the following steps:
Step one, establishing a physiological data acquisition module and acquiring physiological data
The step of collecting data of the physiological data collection module is shown in figure 1, the measured individual collects PPG and ECG physiological parameters, firstly, the embedded LED light source irradiates to the measured part (four points of the wrist) for collecting PPG data, each tissue of the measured part is absorbed to different degrees, the light absorption quantity changes along with the change of substances, and the reflected transmitted light intensity signals are received by the photoelectric conversion sensors of the four points and converted into electric signals, so that PPG data of different points in the same wave band can be obtained, and the MCU time-sharing drives the LEDs to generate light in the wave bands of 550nm, 660nm, 880nm and 960nm, so that PPG data of different wave bands and different points can be obtained; the acquisition of ECG data requires that electrode pads are connected to specified positions of a body, common ECG acquisition modes comprise 12, 5, 3 leads and the like, but the acquisition is complex, consumable materials exist in wet electrodes, and the ECG acquisition is not suitable for real-time detection; in addition, due to the reasons of low amplitude, high human body impedance and the like, the ECG acquisition is easily influenced by factors such as noise, low input impedance of an acquisition device and the like, and in order to solve the problems, the invention integrates an ECG acquisition circuit comprising an amplifying and holding circuit, a filter circuit, an AD conversion circuit and the like and simultaneously carries out high input impedance design, thereby enhancing the robustness of an acquisition module and realizing accurate acquisition of ECG data.
The hardware acquisition structure of the physiological data acquisition module is shown in the anti-motion artifact designs shown in fig. 3 and 4, wherein '1' and '2' respectively represent photoelectric conversion sensors and multiband LED lamps, 4 photoelectric conversion sensors are projected to be contacted with skin more stably, and 4 band LEDs are recessed to be designed, so that the radiation area of LED light is wider, the inhibition of motion artifact is realized, and the resistance of the device to the motion artifact is improved.
The physiological data acquisition module adopts a wristband type design, and when a user performs blood pressure detection, the physiological data acquisition can be realized by only wearing the device with one hand and touching the electrode plate on the device with the other hand.
The physiological data acquisition module supports the expansion of functions such as data storage and 4G communication, can store and transmit a series of data such as multi-mode physiological data and blood pressure estimated values in a cloud, and supports the subsequent long-term management of blood pressure and deep information mining.
The physiological data of the user is continuously acquired in real time, wherein the physiological data mainly comprises non-invasive measurement of multiple physiological parameters by photoplethysmography and single-lead ECG.
Collecting human PPG data and ECG data, wherein the main component of the PPG signal is pulse waveform caused by heart pulsation, and the conditions of heart contraction and relaxation can be reflected indirectly by analyzing the characteristics of pulse wave, such as the amplitude, rise time, fall time and the like of the pulse wave; the ECG records the electrical activity of the heart. By analyzing characteristics such as morphology, duration, and interval of the ECG signal, the rhythm and function of the heart can be assessed.
Step two, physiological signal compensation is carried out based on motion pseudo-noise suppression of measurement points and band skin color difference suppression
The PPG algorithm adopts an optical sensing principle, but in the acquisition process, waveform distortion of PPG is often caused by factors such as motion pseudo noise, sensor attachment, ambient light, skin color difference and the like, and the partial algorithm performs PPG data quality optimization based on multi-point multi-band PPG data, so that the accuracy and reliability of the PPG data are improved.
The physiological signal compensation algorithm consists of a motion artifact suppression algorithm based on measurement points and a band-based skin color difference suppression algorithm.
The motion pseudo noise suppression algorithm based on the measurement points comprises the following specific flow:
(1) For the PPG data (PPG A、PPGB、PPGC、PPGD) of 4 points (A, B, C, D is assumed to be four points) in the same wave band, the acquisition area is smaller, the difference of 7 human skin colors and the like in the measurement area is small, the correlation of different point data on the time domain and morphological characteristics is assumed to be high, in ideal case, the voltage waveform of the PPG data consists of reflected light, ambient light and instrument noise, and the amplitude of the PPG is expressed by the following formula
VPPG＝V Reflected light +V Ambient light +V Instrument noise (1)
Wherein V Reflected light is the voltage generated by the reflected light of the corresponding wave band, V Ambient light is the voltage generated by the light of the corresponding wave band existing in the environment, and V Instrument noise is the voltage noise generated by the measuring device itself.
(2) When motion occurs, V Reflected light will change due to local muscle change and loose wearing, and meanwhile, the phenomena of light leakage in the measurement area will also cause V Ambient light to change, and specific motion artifacts are shown in fig. 6:
(a) When the motion artifact 1 occurs, in the above formula (1), due to light leakage, the light intensity of the LED is reduced, and then V Reflected light is reduced, and meanwhile, the ambient light is enhanced to raise V Ambient light , but as can be seen from fig. 6, two parameters of the point D have the smallest change, and at this time, PPG data of the point D is selected as a PPG acquisition value of the band;
(b) And when the situation shown in the motion artifact 2 of fig. 6 occurs, V PPG at four points is V A+ΔVA、VB+ΔVB、VC+ΔVC、VD+ΔVD, respectively, where V A、VB、VC、VD is V PPG when there is no motion artifact, respectively. As can be seen from the positional relationship of fig. 6, Δv A<ΔVB、ΔVC<ΔVD, but the physical positions of a and B, C and D have a linear relationship, and it is assumed that there is a functional relationship f between the waveform voltage change of the point O, PPG between a and D and the point, so:
VO＝f((VB-VA),(VC-VD),VD,VA) (2)
The virtual point V O is PPG data of the wave band, and the reduction of motion pseudo noise is realized through the mapping relation.
(3) When there is no motion artifact, then V Reflected light ,V Ambient light is assumed to be substantially similar, then V PPG is derived from the following equation:
as can be seen from equation (3), when there is no motion artifact, the noise error of the instrument is reduced according to this manner.
The band-based skin color difference suppression algorithm is based on the fact that the penetration capacities of different band lights on different tissue layers of skin are different:
The penetration depth of green light (wavelength 495-570 nm) at the skin surface is relatively shallow, and it is mainly absorbed by blood vessels, hair and pigments at the skin surface;
the penetration depth of red light (about 620-700 nanometers) is relatively large, and can reach the middle and shallow layers of skin, so that the red light can penetrate the epidermis layer and the dermis layer, and has certain application value for evaluating skin tissues and blood components;
Infrared light (about wavelength 700 nm to 1 mm): the skin can be penetrated through the dermis layer and a part of subcutaneous tissue, and infrared light has a certain effect on observation and evaluation of some deep tissues.
Wherein V Reflected light ,Vgreen、Vred、Vir_1、Vir_2 for the four bands is represented by the following formula:
wherein, V green is green light, V red is red light, V ir_1 and V ir_2 represent voltages generated by infrared light in different wave bands, when proper wearing and other measurement conditions are ideal, V Skin color is constant, and V Surface layer 、V Superficial layer in skin 、V Deep layer in skin 、V Deep layer of skin consists of tissue constant and vascular pulsation of the layer, namely:
V Layer(s) ＝V Layer structure +V blood vessel
And obtaining light reflection voltages of different tissue layers through a data difference algorithm, so as to achieve the effect of skin color difference inhibition.
Thirdly, constructing graph structure data after extracting waveform characteristics, performing model training according to a graph convolution neural network, and finally realizing accurate estimation of blood pressure data
In order to realize deep mining of PPG and ECG data information and realize real-time and accurate detection of blood pressure indexes, the method comprises the steps of constructing graph structure data after waveform characteristics are extracted, performing model training according to a graph roll neural network (GCN), and finally realizing accurate estimation of blood pressure data.
The algorithm mainly comprises an input layer, a GCN layer and a connecting layer.
When the physiological data is input into the layer, firstly, feature extraction is carried out on the physiological data, and feature points of PPG of different wave bands shown in figure 7 are extracted mainly by adopting a Fourier curve fitting, an AMPD peak algorithm, a threshold method and the like on PPG data, wherein the features comprise wave crests, wave troughs, wave crests rising by 0.1 times and the like. The ECG signal mainly adopts wavelet transformation and self-adaptive sliding window algorithm to identify P wave, Q wave, R wave, S wave, T wave and the like.
The graph structure data is constructed using the above features, as shown in fig. 8. Wherein, the hollow node represents the characteristics of different wave bands PPG at different waveform characteristic points, and the black node is the ECG characteristic. The nodes from left to right are sequentially time sequence and amplitude parameters corresponding to different characteristic points, and the sides are corresponding relations of actual physiological activities of the heart.
After the graph structure data is established, model training is carried out by utilizing the GCN, and the graph structure data is sequentially processed by the GCN, the pooling layer and the full-connection layer, wherein the model training method comprises the following steps of:
1. GCN layer: the characteristic representation of the node is updated by calculating the weighted average value of the node and the neighbor node, and the specific steps are as follows:
(1) The GCN layer is a core component of the graph roll-up neural network for feature propagation and learning on graph structure data.
(2) Feature propagation is performed by using neighbor information of nodes, so that features of each node are affected by neighboring nodes.
Specifically, the output of the GCN layer is an updated node feature matrix, wherein the features of each node are updated to take into account new representations of neighbor node information. Wherein the formula of the GCN layer is:
where σ is the activation function, H l is the node feature matrix of the first layer, In the form of an adjacency matrix a plus a self-loop,In the form of a degree matrix D plus a self-loop, W l is a layer i weight matrix,
2. Pooling layer: used in graph convolutional neural networks to reduce the number of nodes in the graph and to retain important information. The pooling operation aggregates a set of nodes into one node, thereby reducing the size of the graph. Common graph pooling methods include maximum pooling, which chooses the largest eigenvalue in the aggregate node, and average pooling, which computes the characteristic average of the aggregate node.
3. Full tie layer: the fully connected layer is used for final feature processing and output prediction in the graph roll-up neural network. Each node in the fully connected layer has a connection with all nodes of the previous layer, each connection having a weight. The fully connected layer performs further feature extraction and combination of the outputs of the graph roll layer and the pooling layer to generate a final prediction result. The fully-connected layer includes a ReLU (RECTIFIED LINEAR Unit) nonlinear activation function to increase the expressive power and nonlinear modeling power of the network.
Through training and verification of the self-built data set, the error of the blood pressure estimation of the module is as follows:
Index (I) Absolute error/mmHg Relative error/%
Systolic Blood Pressure (SBP) ±9.1 7.8
Diastolic Blood Pressure (DBP) ±5.5 7.09
Therefore, the noninvasive blood pressure detection method based on the multi-source data fusion and the graph neural network improves the estimation accuracy of the blood pressure change trend and the robustness of long-term detection, and supports the long-term detection of blood pressure.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention and not for limiting it, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that: the technical scheme of the invention can be modified or replaced by the same, and the modified technical scheme cannot deviate from the spirit and scope of the technical scheme of the invention.

Claims (6)

1. A noninvasive blood pressure detection method based on multi-source data fusion and a graph neural network is characterized by comprising the following steps:
s1, establishing a physiological data acquisition module and acquiring human PPG data and ECG data;
S2, performing physiological signal compensation based on motion pseudo-noise suppression of measurement points and skin color difference suppression of wave bands;
S3, constructing graph structure data after extracting waveform characteristics, performing model training according to a graph convolution neural network, and finally realizing accurate estimation of blood pressure data;
In step S2, the specific flow of the motion artifact suppression algorithm based on the measurement point location is as follows:
(1) For 4 points in the same wave band, the four points are A, B, C, D, PPG data and PPG A、PPGB、PPGC、PPGD, the acquisition area is smaller, the difference of human skin color is small in the measurement area, the correlation of the data of different points in time domain and morphological characteristics is high, in ideal case, the voltage waveform of PPG data consists of reflected light, ambient light and instrument noise, and the amplitude of PPG is represented by the following formula
VPPG＝V Reflected light +V Ambient light +V Instrument noise (1)
Wherein V Reflected light is the voltage generated by the reflected light of the corresponding wave band, V Ambient light is the voltage generated by the light of the corresponding wave band existing in the environment, and V Instrument noise is the voltage noise generated by the measuring device;
(2) When motion occurs, V Reflected light can change due to local muscle change and loose wearing, meanwhile, the light leakage phenomenon of the measurement area can also change V Ambient light , and PPG data are related to motion artifact 1 and motion artifact 2;
(3) When there is no motion artifact, then V Reflected light ,V Ambient light is identical, and then V PPG is derived from the following equation:
as can be seen from the formula (2), when motion artifacts are not present, reducing the noise error of the instrument according to the formula (2);
In step S2, the influence of two kinds of motion artifacts on PPG data is as follows:
(1) When the motion pseudo-noise 1 occurs, in the above formula (1), due to light leakage, the light intensity of the LED is reduced, and then V Reflected light is reduced, and simultaneously, the enhancement of the ambient light increases V Ambient light , and the parameter with the smallest front-back variation is reduced as the PPG acquisition value of the band;
(2) When motion artifact 2 occurs, V PPG of the four points is V A+ΔVA、VB+ΔVB、VC+ΔVC、VD+ΔVD, where V A、VB、VC、VD is V PPG of no motion artifact, and since Δv A<ΔVB、ΔVC<ΔVD exists in a linear relationship between the physical positions of a and B, C and D, and there is a functional relationship f between the waveform voltage change of the existence point O, PPG and the point between a and D, so that:
VO＝f((VB-VA),(VC-VD),VD,VA) (3)
Wherein DeltaV A represents the noise voltage generated by motion artifact at the A point; deltaV B represents the noise voltage generated by motion artifact at point B; deltaV C represents the noise voltage generated by motion artifact at point C; deltaV D represents the noise voltage generated by motion artifact at point D; the virtual point V O is PPG data of the wave band, and the reduction of motion pseudo noise is realized through the mapping relation.
2. A non-invasive blood pressure detection method based on multi-source data fusion and graph neural network according to claim 1, wherein in step S1, the step of collecting PPG data is as follows: firstly, an embedded multiband LED light source irradiates a tested part, each tissue of the tested part is absorbed to different degrees, the light absorption quantity changes along with the change of a substance, and a reflected transmitted light intensity signal is received by a plurality of photoelectric conversion sensors at points and converted into an electric signal, namely PPG data of different points at different wavebands is obtained;
the acquisition of ECG data requires the attachment of electrode pads to designated locations of the body.
3. The non-invasive blood pressure detection method based on multi-source data fusion and graph neural network according to claim 1, wherein in step S3, the graph convolution neural network blood pressure estimation algorithm mainly comprises an input layer, a GCN layer and a connection layer, the physiological data is subjected to feature extraction when the input layer is used, the graph structure data is constructed by the features, then model training is performed by using the GCN, and the graph structure data is sequentially subjected to GCN, pooling layer and full connection layer processing, specifically as follows:
(1) GCN layer: updating the characteristic representation of the node by calculating the weighted average value of the node and the neighbor node thereof;
(2) Pooling layer: the method is used for reducing the number of nodes in the graph convolution neural network, and keeping important information, and a pooling operation is used for aggregating a group of nodes into one node, so that the scale of the graph is reduced;
(3) Full tie layer: the full-connection layer is used for carrying out final feature processing and output prediction in the graph roll neural network, each node in the full-connection layer is connected with all nodes of the previous layer, each connection has weight, the full-connection layer carries out further feature extraction and combination on the outputs of the graph roll layer and the pooling layer to generate a final prediction result, and the full-connection layer comprises a ReLU nonlinear activation function to increase the expression capacity and nonlinear modeling capacity of the network.
4. A non-invasive blood pressure detection method based on multi-source data fusion and graph neural network according to claim 3, wherein in step S3, the characteristic representation of the node is updated by calculating the weighted average of the node and its neighboring nodes, and the specific steps are as follows:
(a) The GCN layer is a core component of the graph convolution neural network and is used for carrying out feature propagation and learning on graph structure data;
(b) Feature propagation is performed by using neighbor information of nodes, so that the features of each node are affected by neighboring nodes;
Specifically, the output of the GCN layer is an updated node feature matrix, wherein the features of each node are updated to a new representation that comprehensively considers neighbor node information, wherein the formula of the GCN layer is:
where σ is the activation function, H l is the node feature matrix of the first layer, In the form of adjacency matrix A plus self-loop,/>Is in the form of a degree matrix D plus a self-loop, W l is the layer i weight matrix.
5. The non-invasive blood pressure detection method based on multi-source data fusion and graph neural network according to claim 1, wherein in step S3, feature extraction of physiological data mainly comprises: and extracting characteristic points of PPG of different wave bands for PPG data, wherein the characteristic points comprise wave crests and wave troughs, the wave form rises by 0.1 times of the wave crests, and the ECG signals are used for identifying P waves, Q waves, R waves, S waves and T waves.
6. The noninvasive blood pressure detection method based on multi-source data fusion and graph neural network according to claim 1, wherein the physiological data acquisition module is additionally provided with high input impedance in an ECG acquisition circuit, so that the robustness of the acquisition module is enhanced, and the accurate acquisition of ECG data is realized.
CN202310885999.1A 2023-07-19 2023-07-19 Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network Active CN116803340B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310885999.1A CN116803340B (en) 2023-07-19 2023-07-19 Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310885999.1A CN116803340B (en) 2023-07-19 2023-07-19 Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network

Publications (2)

Publication Number Publication Date
CN116803340A CN116803340A (en) 2023-09-26
CN116803340B true CN116803340B (en) 2024-06-14

Family

ID=88079508

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310885999.1A Active CN116803340B (en) 2023-07-19 2023-07-19 Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network

Country Status (1)

Country Link
CN (1) CN116803340B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015187732A1 (en) * 2014-06-03 2015-12-10 The Texas A&M University System Optical sensor for health monitoring
CN114145724A (en) * 2021-12-08 2022-03-08 四川北易信息技术有限公司 Method for dynamically monitoring blood pressure based on ECG (electrocardiogram) and PPG (photoplethysmography) multiple physiological characteristic parameters

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11534071B2 (en) * 2019-04-12 2022-12-27 Beijing Shunyuan Kaihua Technology Limited Blood pressure measurement
CN111493850A (en) * 2020-04-13 2020-08-07 中国科学院深圳先进技术研究院 Blood pressure measuring method and device based on deep neural network
WO2023055862A1 (en) * 2021-10-01 2023-04-06 Softserve, Inc. Systems and methods for platform-agnostic, real-time physiologic vital sign detection from video stream data
CN115956889A (en) * 2022-12-05 2023-04-14 出门问问信息科技有限公司 Blood pressure monitoring method and device and electronic equipment

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015187732A1 (en) * 2014-06-03 2015-12-10 The Texas A&M University System Optical sensor for health monitoring
CN114145724A (en) * 2021-12-08 2022-03-08 四川北易信息技术有限公司 Method for dynamically monitoring blood pressure based on ECG (electrocardiogram) and PPG (photoplethysmography) multiple physiological characteristic parameters

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于脉搏波的糖尿病和高血压诊断算法研究;汤宇飞;《中国矿业大学硕士学位论文》;20230501;第37页第1段至第52页第2段 *

Also Published As

Publication number Publication date
CN116803340A (en) 2023-09-26

Similar Documents

Publication Publication Date Title
CN109157202B (en) Cardiovascular disease early warning system based on multi-physiological signal deep fusion
US20220104775A1 (en) Processing of electrophysiological signals
WO2019161609A1 (en) Method for analyzing multi-parameter monitoring data and multi-parameter monitor
CA3097663A1 (en) Methods to estimate the blood pressure and the arterial stiffness based on photoplethysmographic (ppg) signals
CN107233087A (en) A kind of Woundless blood pressure measuring device based on photoplethysmographic feature
US11666229B2 (en) Monitoring device for monitoring of vital signs
CN110236508A (en) A kind of non-invasive blood pressure continuous monitoring method
CN108478203A (en) A kind of blood pressure measuring method monitoring radar based on single vital sign
Zhou et al. The noninvasive blood pressure measurement based on facial images processing
CN111714088B (en) Human body characteristic index detection method and system based on traditional Chinese medicine principle
Sel et al. ImpediBands: Body coupled bio-impedance patches for physiological sensing proof of concept
CN115299899A (en) Activity recognition and beat-to-beat blood pressure monitoring, analyzing and early warning system based on multiple sensors
CN110881967A (en) Non-invasive multi-segment peripheral arterial vessel elastic function detection method and instrument thereof
CN107582040B (en) Method and device for monitoring heart rhythm
Mehrgardt et al. Deep learning fused wearable pressure and PPG data for accurate heart rate monitoring
CN106618532A (en) Dressing device for collecting characteristic parameters of electrocardio and blood pressure and pulse
Wang et al. A novel approach to estimate blood pressure of blood loss continuously based on stacked auto-encoder neural networks
CN116869499A (en) Continuous blood pressure measurement method based on PPG and multi-order differential signals thereof
CN116803340B (en) Noninvasive blood pressure detection method based on multi-source data fusion and graph neural network
Kew et al. Wearable patch-type ECG using ubiquitous wireless sensor network for healthcare monitoring application
Alqudah et al. Multiple time and spectral analysis techniques for comparing the PhotoPlethysmography to PiezoelectricPlethysmography with electrocardiography
Yılmaz et al. Comparison of Electrode Configurations for Impedance Plethysmography Based Heart Rate Estimation at the Forearm
CN209733969U (en) Pulse wave propagation time measuring equipment
Lu et al. Uncertainties in the Analysis of Heart Rate Variability: A Systematic Review
Zhe et al. Design and research of continuous blood pressure monitoring system based on lora

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