WO2023226824A1 - 一种基于小动脉光电容积脉搏波的血压检测装置 - Google Patents

一种基于小动脉光电容积脉搏波的血压检测装置 Download PDF

Info

Publication number
WO2023226824A1
WO2023226824A1 PCT/CN2023/094497 CN2023094497W WO2023226824A1 WO 2023226824 A1 WO2023226824 A1 WO 2023226824A1 CN 2023094497 W CN2023094497 W CN 2023094497W WO 2023226824 A1 WO2023226824 A1 WO 2023226824A1
Authority
WO
WIPO (PCT)
Prior art keywords
blood pressure
arteriolar
pressure
pulse wave
photoplethysm
Prior art date
Application number
PCT/CN2023/094497
Other languages
English (en)
French (fr)
Inventor
周聪聪
王建军
Original Assignee
杭州兆观传感科技有限公司
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 杭州兆观传感科技有限公司 filed Critical 杭州兆观传感科技有限公司
Publication of WO2023226824A1 publication Critical patent/WO2023226824A1/zh

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
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
    • A61B5/6802—Sensor mounted on worn items
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems

Definitions

  • the present invention relates to the technical field of blood pressure detection, and in particular to a blood pressure detection device based on arteriolar photoelectric volume pulse wave.
  • Blood pressure is an important physiological parameter of the human body and can reflect the status of cardiovascular function.
  • High blood pressure increases a person's risk of a variety of cardiovascular diseases, including stroke, coronary artery disease, heart failure, atrial fibrillation, and peripheral vascular disease.
  • Monitoring blood pressure trends under different cardiovascular states is conducive to better understanding the pathogenesis of cardiovascular diseases and implementing effective management and control.
  • PPG signals When using PPG signals to predict blood pressure, according to the physiological structure of the human body, if you want to obtain a relatively stable PPG signal, it is generally necessary to measure it in a part of the human body with rich blood vessels. Currently, it is mainly measured on the earlobe or wrist, but the signal quality in these parts is poor. , signal-to-noise ratio is low.
  • the current conventional method is to directly measure the PPG signals to predict blood pressure.
  • the directly measured PPG signal contains many volume fluctuations, such as those of capillaries, arterioles, tissue fluid, and even veins, which causes greater signal interference and makes it difficult to measure the PPG signal. precise. Therefore, other PPG interference signals need to be eliminated when predicting blood pressure. Obtaining the photoplethysmogram at the arteriole and using it to predict blood pressure is the most direct and accurate.
  • the object of the present invention is to provide a blood pressure detection device based on arteriolar photoplethysmography to improve the accuracy of blood pressure prediction.
  • the present invention provides the following solutions:
  • a blood pressure detection device based on arteriolar photoplethysm pulse wave which is characterized in that it includes a finger ring structure, a pulse wave detection unit, a temperature sensing unit, a pressure sensing unit and a micro control unit arranged at the bottom of the ring structure, and a set of Spring control units on both sides inside the ring structure;
  • the pulse wave detection unit is used to detect the photoplethysm pulse wave at the user's fingertips
  • a pressure sensing unit used to detect the contact pressure value of the user's fingertip
  • Temperature sensing unit used to collect the temperature value of the user's fingertip
  • a microcontrol unit respectively connected to the pulse wave detection unit, the pressure sensing unit and the temperature sensing unit, for calculating arterioles based on the photoplethysm pulse wave, the pressure value and the temperature value.
  • Photoplethysmogram and perform blood pressure detection based on the arteriolar photoplethysm wave;
  • a spring control unit is connected to the pressure sensing unit and used to adjust the pressure sensing unit according to the pressure value.
  • the pulse wave detection unit includes a photoelectric sensor module.
  • the photoelectric sensor module is used to detect the photoplethysm of the user's fingertips and send the photoplethysm wave to the user through an analog front end or interface circuit.
  • micro control unit micro control unit
  • the pressure sensing unit includes a pressure sensor, the pressure sensor is used to detect the contact pressure value of the user's fingertip, and sends the pressure value to the micro control unit through an analog front end or interface circuit;
  • the temperature sensing unit includes a temperature sensor, which is used to collect the temperature value of the user's fingertip and send the temperature value to the microcontrol unit through an analog front end or interface circuit;
  • the blood pressure detection device also includes an IMU detection unit, which is connected to the pulse detection unit, the pressure sensing unit, the temperature sensing unit and the micro control unit respectively, for Detect whether there is interference data in the data detected by the pulse detection unit, the pressure sensing unit and the temperature sensing unit, and send the interference-removed data to the micro control unit.
  • IMU detection unit which is connected to the pulse detection unit, the pressure sensing unit, the temperature sensing unit and the micro control unit respectively, for Detect whether there is interference data in the data detected by the pulse detection unit, the pressure sensing unit and the temperature sensing unit, and send the interference-removed data to the micro control unit.
  • the spring control unit includes a precision motor, a ratchet structure and a constant pressure elastic module;
  • the constant pressure elastic module includes a constant pressure spring, a movable clamp and a fixed clamp;
  • the precision motor drives the spring according to the pressure value
  • the ratchet structure rotates to adjust the length of the constant pressure spring;
  • the movable clamp is set outside the constant pressure spring, and the fixed clamp is set on the upper part of the movable clamp;
  • the top end of the constant pressure spring Connected to the ratchet structure, the bottom end of the constant pressure bomb is connected to the bottom end of the finger ring structure;
  • the fixed clamp and the movable clamp are fixed in the finger ring structure.
  • the photoelectric sensor module includes a plurality of light-emitting LEDs and a plurality of photoelectric receivers; the plurality of light-emitting LEDs are arranged symmetrically about the center of the plurality of photoelectric receivers.
  • the photoelectric sensor module includes a plurality of light-emitting LEDs and a photoelectric receiver; a plurality of the light-emitting LEDs are arranged around the photoelectric receiver.
  • calculating the arteriolar photoplethysmogram based on the photoplethysmogram, the pressure value and the temperature value, and performing blood pressure detection based on the arteriolar photoplethysm pulse wave specifically includes:
  • a cardiovascular state preliminary screening model is used to determine the cardiovascular state of the user
  • a selected blood pressure prediction model is used for blood pressure detection.
  • the training process of the cardiovascular status preliminary screening model includes:
  • the training data set includes photoplethysmographic pulse wave and blood pressure data of different cardiovascular states, different genders, and different ages; the cardiovascular states include normal, atrial fibrillation, and atherosclerosis;
  • the machine learning model is trained through the various characteristic information and the cardiovascular state corresponding to the photoplethysm wave, and a preliminary cardiovascular state screening model is obtained.
  • the training process of the blood pressure prediction model group includes:
  • Multiple deep learning models are trained through the second set of training data and corresponding blood pressure data under different cardiovascular states to obtain a blood pressure prediction model group; the blood pressure prediction model group includes multiple blood pressure detection models, and the blood pressure detection models are used for Detection of blood pressure in cardiovascular states.
  • calculating the arteriolar photoplethysmogram according to the photoplethysmogram, the pressure value and the temperature value specifically includes:
  • An arteriolar photoplethysmogram is calculated based on the photoplethysmogram, the scaling factor, and a revised term of the scaling factor.
  • the present invention discloses the following technical effects:
  • the blood pressure detection device based on arteriolar photoplethysmography of the present invention adopts the form of a finger ring to measure at finger parts rich in blood vessels. It has a simple structure and easy operation.
  • the obtained signal has a high signal-to-noise ratio and high accuracy. , achieving continuous dynamic real-time collection of human aortic blood pressure, which has high medical value and broad market application prospects.
  • the present invention realizes PPG signal detection at different penetration depths by setting up a multi-wavelength photoelectric detection sensor module. It combines the motor, constant pressure spring and pressure sensor to realize the pressure control of the measurement part, and at the same time integrates the temperature of the detection part to remove It eliminates interference signals and realizes direct measurement and tracking of APPG signals.
  • Figure 1 is a schematic diagram of a blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • Figure 2 is a schematic structural diagram of a ring of a blood pressure detection device based on arteriolar photoplethysmography provided by an embodiment of the present invention
  • Figure 3 is a cross-sectional view of a ring structure of a blood pressure detection device based on arteriolar photoplethysmography provided by an embodiment of the present invention
  • Figure 4 is a schematic structural diagram of a constant pressure elastic module of a blood pressure detection device based on arteriolar photoplethysmography provided by an embodiment of the present invention
  • Figure 5 is a schematic diagram of feedback control of the constant pressure elastic module of the blood pressure detection device based on arteriolar photoplethysmography provided by an embodiment of the present invention
  • Figure 6 is a schematic diagram of the bottom end of the ring structure of the blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • Figure 7 is a schematic diagram of the implementation of arteriolar photoplethysmography pulse wave detection of the blood pressure detection device based on arteriolar photoplethysmography provided by an embodiment of the present invention
  • Figure 8 is a schematic diagram of the waveforms of the arteriolar photoplethysm, pressure, and temperature of the blood pressure detection device based on arteriolar photoplethysm provided by an embodiment of the present invention
  • Figure 9 is a block diagram of the implementation principle of the arteriolar photoplethysm pulse wave, pressure, and temperature detection device of the blood pressure detection device based on the arteriolar photoplethysm wave provided by the embodiment of the present invention.
  • Figure 10 is a block diagram of the APPG signal detection principle of the blood pressure detection device based on arteriolar photoplethysmography provided by an embodiment of the present invention
  • Figure 11 is a flow chart of group model learning and training of a blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • Figure 12 is a schematic diagram of feature extraction of a blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • Figure 13 is a flow chart for preliminary screening of cardiovascular status of the blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • Figure 14 is a schematic diagram of a group model construction of a blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • Figure 15 is a schematic diagram of the transfer learning model construction of a blood pressure detection device based on arteriolar photoplethysm pulse wave provided by an embodiment of the present invention
  • 1 Photoelectric sensor module
  • 1-1 LED light source with wavelength ⁇ 1
  • 1-2 LED light source with wavelength ⁇ 2
  • 1-3 Photoelectric receiver PD 1-1-1: The pulse wave received by the LED light with wavelength ⁇ 1 at 1-3
  • 1-2-1 The pulse wave received by the LED light with wavelength ⁇ 2 at 1-3
  • 1-12-1 Arteriolar pulse wave synthesized by LED light with wavelengths ⁇ 1 and ⁇ 2
  • 2-Temperature sensor 3-Pressure sensor
  • 4-1 Constant pressure spring
  • 4-2 Movable clamp
  • 4-3 Fixed fixture
  • 4-4 Top threaded hole
  • 4-5 Bottom threaded hole
  • 4-6 Top connection line
  • 4-7 Bottom connection line
  • 6-1 Threaded hole on the right side of the bottom of the ring
  • 6-2 Threaded hole on the left side of the bottom of the ring
  • 6-3 Micro control unit
  • 6-4 Electronic components
  • 6-5 Sensor
  • the object of the present invention is to provide a blood pressure detection device based on arteriolar photoplethysmography to improve the accuracy of blood pressure prediction.
  • a blood pressure detection device based on arteriolar photoplethysm pulse wave includes a ring structure, a pulse wave detection unit, a temperature sensing unit, a pressure sensing unit and a microcontroller arranged at the bottom of the ring structure. unit and spring control units arranged on both sides inside the ring structure.
  • the pulse wave detection unit is used to detect the photoplethysm of the user's fingertips.
  • the pressure sensing unit is used to detect the contact pressure value of the user's fingertip.
  • the temperature sensing unit is used to collect the temperature value of the user's fingertip.
  • the micro control unit is connected to the pulse wave detection unit, the pressure sensing unit and the temperature sensing unit respectively.
  • the microcontrol unit is used to calculate the arteriolar photoplethysmogram according to the photoplethysmogram, the pressure value and the temperature value, and to perform blood pressure detection based on the arteriolar photoplethysm pulse wave.
  • the micro-control unit contains the Bluetooth BLE low-power protocol stack. The micro-control unit is used to control each detection unit to collect signal data, analyze the signal, extract features and calculate blood pressure, and then send the blood pressure detection results to the terminal through the Bluetooth protocol stack.
  • the spring control unit is connected with the pressure sensing unit.
  • the spring control unit is used to adjust the pressure sensing unit according to the pressure value.
  • the blood pressure detection device also includes an IMU detection unit.
  • the IMU detection unit is connected to the pulse detection unit, pressure sensing unit, temperature sensing unit and micro control unit respectively; the IMU detection unit is used to detect the data detected by the pulse detection unit, pressure sensing unit and temperature sensing unit. Whether there is interference data, and the data after interference removal is sent to the micro control unit.
  • the blood pressure detection device also includes a power management unit and a storage unit.
  • the power management unit is used to supply power to power-consuming units; the storage unit is used to store local data.
  • the pulse wave detection unit includes a photoelectric sensor module 1 .
  • the photoelectric sensor module 1 is used to detect the photoplethysm wave at the user's fingertip, and send the photoplethysm wave to the micro control unit 6-3 through an analog front end or interface circuit.
  • the pressure sensing unit includes a pressure sensor 3 .
  • the pressure sensor 3 is used to detect the contact pressure value of the user's fingertip, and send the pressure value to the micro control unit 6-3 through an analog front end or interface circuit.
  • the temperature sensing unit includes a temperature sensor 2 .
  • the temperature sensor 2 is used to collect the temperature value of the user's fingertip, and send the temperature value to the micro control unit 6-3 through an analog front end or interface circuit.
  • the spring control unit includes a precision motor, a ratchet structure and a constant pressure elastic module 4.
  • the precision motor drives the ratchet structure to rotate according to the pressure value to adjust the length of the constant pressure spring 4-1;
  • the constant pressure elastic module 4 includes a constant pressure spring 4-1, a movable clamp 4-2 and a fixed clamp 4-3;
  • the movable clamp 4 -2 is set on the outside of the constant pressure spring 4-1;
  • the fixed clamp 4-3 is set on the upper part of the movable clamp 4-2;
  • the top of the constant pressure spring 4-1 is connected to the ratchet structure;
  • the bottom end is connected to the bottom end of the finger ring structure;
  • the fixed clamp 4-3 and the movable clamp 4-2 are fixed in the finger ring structure.
  • Each unit of the blood pressure detection device has a detachable structure and is assembled through a modular design.
  • the sensing, pressure control, power control and other parts can be separated by disassembly.
  • the pressure sensor 3, the photoelectric sensor module 1, and the temperature sensor 2 form a sensing detection module; the sensing detection module is located on the fingertip and passes through the threaded hole 6- on the right side of the bottom of the ring. 1.
  • the threaded hole 6-2 on the left side of the bottom of the ring and the threaded hole 4-5 at the bottom of the constant pressure elastic module are connected to the constant pressure elastic module 4.
  • the photoelectric sensor module 1 can be composed of multiple light-emitting LEDs and multiple photoelectric receivers PD. It can be multiple LEDs (such as 1-1, 1-2) forming a circular array around one PD (such as 1-3). , it can also be a structure in which LEDs such as (1-1, 1-2) are centrally symmetrical about PD (such as 1-3).
  • the top of the ring structure mainly includes a ratchet structure and a battery 5.
  • the finger ring structure is connected to the constant pressure elastic module 4 through the threaded holes on the left and right sides of the top and the threaded holes 4-4 on the top of the constant pressure elastic module.
  • the constant pressure elastic modules 4 are located on both sides of the ring, arranged symmetrically, and have the same structure on the left and right sides.
  • the constant pressure elastic module 4 is connected to the ratchet structure through the top connection line 4-6; the constant pressure elastic module 4 is connected to the sensor detection module through the bottom connection line 4-7.
  • the ratchet structure controls the length of the constant pressure spring 4-1 and adjusts the pressure value between the sensing detection module and the finger pulp.
  • the length of the sensor spring structure 6-5 will also change twice as the length of the constant pressure spring 4-1 changes. adjust.
  • the micro control unit 6-3 detects the pressure value measured at the pressure sensor 3 through the analog front end or interface circuit of the pressure sensing unit; at the same time, the micro control unit 6-3 detects the photoelectric sensor PPG signal measured at module 1.
  • the pressure value when the ring is just put on is recorded as F 0
  • the corresponding PPG signal at this time is recorded as PPG 0 .
  • the micro control unit 6-3 controls the precision motor to rotate, driving the ratchet structure to rotate clockwise N steps (N ⁇ 1), and the ratchet wheel
  • the structure stretches the constant pressure spring 4-1 through the top connection line 4-6, thereby stretching the entire sensor detection module.
  • the elastic potential energy of the sensor spring structure 6-5 increases, and the pressure value F 1 measured by the pressure sensor 3 at this time is recorded.
  • the corresponding PPG 1 compare the AC amplitudes AC_PPG 0 and AC_PPG 1 corresponding to PPG 0 and PPG 1 ; if AC_PPG 0 >AC_PPG 1 , the ratchet structure rotates counterclockwise M steps (N ⁇ M ⁇ 0), otherwise if AC_PPG 0 ⁇ AC_PPG 1 , then continue to rotate clockwise N steps.
  • this method confirm the appropriate position i of the ratchet structure, and record the value Fi of the pressure sensor at this time, so that AC_PPG i obtains the maximum value at the current position.
  • the spring control unit drives the ratchet structure to rotate through a precision motor to adjust the length of the constant pressure spring 4-1, thereby adjusting the sensor detection module, that is, stretching or compressing the pressure sensor 3 to adjust the pressure.
  • the storage unit, resistor, capacitor and other electronic components 6-4 in the blood pressure detection device are integrated at the bottom of the ring.
  • the blood pressure detection device based on arteriolar photoplethysmography of the present invention adopts the form of a finger ring to measure at finger parts rich in blood vessels. It has a simple structure and an easy operation method.
  • the obtained signal has a high signal-to-noise ratio and high accuracy, achieving The continuous dynamic real-time collection of human aortic blood pressure has high medical value and broad market application prospects.
  • the specific implementation process of the microcontrol unit in the arteriolar photoplethysm pulse wave blood pressure detection device provided by the present invention is as follows:
  • Step 1 Based on the photoplethysm wave, a cardiovascular state preliminary screening model is used to determine the cardiovascular state of the user.
  • Step 2 Select a corresponding blood pressure prediction model from the blood pressure prediction model group according to the user's cardiovascular status.
  • Step 3 Calculate the arteriolar photoplethysmogram according to the photoplethysm wave, the pressure value and the temperature value.
  • Step 4 Use the selected blood pressure prediction model to detect blood pressure based on the arteriolar photoplethysm pulse wave.
  • the training process of the cardiovascular status preliminary screening model in step 1 includes:
  • Step 11 Construct a training data set; the training data set includes photoplethysmographic pulse wave and blood pressure data of different cardiovascular states, different genders, and different ages; the cardiovascular states include normal, atrial fibrillation, and atherosclerosis.
  • Step 12 Divide the photoplethysmogram into a first group of training data and a second group of training data.
  • Step 13 Extract feature information from the first set of training data to obtain multiple feature information.
  • Step 14 Train the machine learning model through the various feature information and the cardiovascular status corresponding to the photoplethysm wave to obtain a preliminary cardiovascular status screening model.
  • PPG And the minimum length required by ABP is set to 5 minutes and all records of shorter length are deleted.
  • a fourth-order Butterworth bandpass filter is used to eliminate baseline drift and high-frequency noise in the PPG signal.
  • the passband frequency of the PPG signal is 0.5-8Hz.
  • a Hampel filter is used to remove spike noise in the ABP signal.
  • the photoplethysmogram (PPG) based on different cardiovascular states is constructed through public data sets; the public data sets include data subsets of cardiovascular state, gender (Gender), age (Age) and blood pressure (BP) ⁇ NORMAL ⁇ , ⁇ AF ⁇ , ⁇ CA ⁇ ;
  • the preprocessed PPG signal is divided into two types (i.e., the first set of training data and the second set of training data); the first one is used for feature engineering to extract features as a machine
  • the input of the learning model is used to train model parameters and output the preliminary screening classification results of cardiovascular status classification; the second is used as the input of the deep learning model for blood pressure prediction.
  • the feature information extracted in the feature engineering includes but is not limited to heart rate HR, peak and valley value Peak, rise time RT, fall time DT, waveform area PA, heart rate variability (HRV), etc., as shown in Figure 12(a), also Including frequency domain, time-frequency domain, nonlinear features such as sample entropy and other features.
  • Feature information is used to implement training, verification and testing of cardiovascular status classification. After the test accuracy exceeds 90%, the model parameters are solidified to achieve cardiovascular status screening.
  • existing research has proven that the intensity, shape, rhythm, rate and other characteristics of pulse waves will change with changes in the geometric shape and mechanical properties of blood vessels.
  • the pulse wave shape of a person with a normal cardiovascular state includes a relatively obvious descending gorge, with regular shapes, and there are obvious differences in the pulse wave morphology of people with abnormal cardiovascular status in terms of waveform area, amplitude, heart rate and heart rate variability.
  • different classification models can be designed for classification assessment of cardiovascular status.
  • Classification models include:
  • Bayesian classifier Bayesian theorem is the core of this type of algorithm, so it is collectively called Bayesian classification. Bayesian decision theory uses misjudgment loss to select the optimal category classification when the relevant probability is known.
  • Decision tree classifier The decision tree algorithm assigns multi-dimensional feature training set samples to different categories, which is equivalent to projecting the training set samples. The projected training set samples are assigned corresponding category labels.
  • the decision tree algorithm uses a recursive process to represent classification and regression. Each time, the optimal features in the feature set are selected, and the training set is classified and regressed based on the selected optimal features to ensure that the classification and regression processes experienced by each training set sample are consistent. is optimal. If the number of features of the training set samples is large, the features will be selected when starting to produce the decision tree, leaving features with strong ability to classify and regress the trained training set samples, and removing those features with weaker abilities. The entire recursive process looks like a tree, so it is called the decision tree algorithm.
  • the core of the decision tree algorithm is to construct a decision tree with high accuracy and low complexity.
  • the construction process of a decision tree includes two parts: decision tree generation and pruning.
  • the generation of a decision tree is to use training set samples to generate a decision tree.
  • the top node of the decision tree is the root node, and each branch is regarded as a decision node, representing the attributes of the object to be classified.
  • Each node represents the existence of a possible classification process.
  • the traversal process is the process of using decision trees to solve classification problems, using several attributes to classify training set samples.
  • Decision tree pruning is the process of testing and revising the generated decision tree. By decomplicating the decision tree, it can have better generalization ability and be able to adapt to more solutions.
  • Random Forest is an optimized version of Bagging based on the tree model. The generation of one tree is definitely not as good as multiple trees, so there is Random Forest to solve the generalization ability of decision trees. Weak characteristics. For the same batch of data, the same algorithm can only produce one tree. In this case, the Bagging strategy can produce different data sets.
  • the bagging strategy comes from bootstrap aggregation: resample and select Nb samples from the sample set (assuming N data points in the sample set) (sampling with replacement, the number of sample data points remains unchanged to N), on all samples , establish a classifier (ID3 ⁇ C4.5 ⁇ CART ⁇ SVM ⁇ LOGISTIC) for these N samples, repeat the above two steps m times, and obtain m classifiers; finally, based on the voting results of these m classifiers, decide whether the data belongs to Which category.
  • a classifier ID3 ⁇ C4.5 ⁇ CART ⁇ SVM ⁇ LOGISTIC
  • FIG. 13 a typical classification demonstration is shown in Figure 13, which uses multi-dimensional time domain, frequency domain, time-frequency domain, and nonlinear features such as sample entropy generated by feature engineering as the input layer.
  • a data synthesis strategy is used to add random Gaussian noise to the age feature and expand the data of the same age to the range of ⁇ 3 or ⁇ 5.
  • LSTM or Bi-LSTM is used to build a multi-layer feedback network, and the Densy layer and softmax layer are set to output the classification results.
  • the model layer can also be RF, SVM and other models.
  • the following table initially shows the average accuracy of the model divided into three categories of cardiovascular states: Normal, AF and CA.
  • the present invention uses a random forest classifier as a machine learning model for preliminary screening of cardiovascular status.
  • classifiers can be appropriately selected based on computing resources.
  • the training process of the blood pressure prediction model group in step 2 includes:
  • Multiple deep learning models are trained through the second set of training data and corresponding blood pressure data under different cardiovascular states to obtain a blood pressure prediction model group; the blood pressure prediction model group includes multiple blood pressure detection models, and the blood pressure detection models are used for Detection of blood pressure in cardiovascular states.
  • a blood pressure prediction model group is established based on the deep learning method.
  • the present invention develops a new model architecture, namely loop-attention neural network (LSTM-AT), based on long short-term memory, attention mechanism and fully connected neural network.
  • the input signal of the model is the pulse wave signal fragment within a certain time period.
  • the PPG fragment in the data subset is used as the input signal, and the BP is used as the label signal to train the deep learning blood pressure prediction model and build blood pressure prediction suitable for different data subsets.
  • the input of the blood pressure prediction model is a PPG data fragment.
  • the output vector After passing through the Bi-LSTM layer and adding the attention mechanism Attension layer to weight the important feature vectors, the output vector is combined with age, Demographic information such as gender is combined, and finally blood pressure prediction is performed through a fully connected layer.
  • Bahdanau attention is applied in the blood pressure prediction model: Among them, c i retains all the hidden state information output by the previous layer and is expressed as the weighted sum between the attention weight vector a i and h ti .
  • a i is normalized using the softmax function, and h ti represents h t at time i.
  • long short-term memory is a special RNN that can effectively solve the problem of gradient disappearance and explosion during training.
  • the structure diagram of the LSTM unit is shown in Figure 14.
  • LSTM has two transmission states: Ct (cell state) and ht (hidden state).
  • Ct changes very slowly, and ht will be different in different segments, corresponding to long-term memory and short-term memory respectively.
  • LSTM transfers information from one time step to another through forget gates, input gates, and output gates.
  • Transfer learning focuses on storing knowledge acquired in a source domain into another target domain, and the problem usually involves a small number of data samples to train the model.
  • This invention will be based on a new model architecture, namely loop-attention neural network (LSTM-AT), based on long short-term memory, attention mechanism and fully connected neural network, pre-trained on PPG and BP data obtained from large public data sets Parameters, using new data to adjust the parameters of the last layer of the individual model during individual use can significantly reduce the number of training sets required for new data, as shown in Figure 15.
  • LSTM-AT loop-attention neural network
  • step 3 specifically includes:
  • Step 31 Determine the adjustment amount of the spring control unit according to the pressure value.
  • Step 32 Obtain the photoplethysm wave when the LED wavelength is ⁇ 1 and ⁇ 2 under the adjustment amount of the spring control unit.
  • Step 33 Determine the scaling coefficient according to the photoplethysm pulse wave signal.
  • Step 34 Determine the revised term of the scaling coefficient according to the temperature value.
  • Step 35 Calculate the arteriolar photoplethysmogram according to the photoplethysm wave, the scaling coefficient and the revised term of the scaling coefficient.
  • the photoplethysmogram detected by the pulse wave detection unit includes: the photoplethysm wave when the LED wavelength is ⁇ 1 under the adjustment amount of the spring control unit, and the photoplethysm wave under the adjustment amount of the spring control unit.
  • the photoplethysm wave when the LED wavelength is ⁇ 2 .
  • the wavelength of LED1-1 is ⁇ 2 and the wavelength of LED1-2 is ⁇ 1 , where ⁇ 1 ⁇ ⁇ 2 , it can be seen that the light penetration depth of ⁇ 1 is deeper, and the photon propagation path contains the volume change signal of capillaries and arterioles.
  • the penetration depth of ⁇ 2 is shallow, and the photon propagation path mainly includes the volume change signal of capillaries. . Since photons propagate in skin tissue at this time mainly in the form of diffuse reflection, they are easily affected by fluctuations in tissue fluid and veins, making it difficult to accurately obtain the signal of a specific layer.
  • the PPG signal detected at the PD is usually a composite signal, where 1-1-1 is the pulse wave received by the LED light of wavelength ⁇ 1 at the photoelectric receiver 1-3; 1-2- 1 is the pulse wave received at the photoelectric receiver 1-3 by LED light with wavelength ⁇ 2 ; 1-12-1 is the arteriolar pulse wave synthesized by LED light with wavelength ⁇ 1 and ⁇ 2 .
  • the PPG signal is also affected by pressure and temperature signals.
  • pressure increases, the AC amplitude of the pulse wave first increases and then decreases.
  • the arterioles undergo a process from unloading to obstruction; as the temperature decreases, the AC amplitude of the pulse wave
  • the AC amplitude decreases, mainly because when the temperature decreases, the blood vessels shrink, the inner diameter decreases, the peripheral resistance increases, and the perfusion decreases.
  • the input signals are the value Fi of the pressure sensor, the temperature Ti , the superficial PPG, and the deep PPG, and the output is the small artery APPG signal.
  • the wavelength of LED1-2 is ⁇ 1
  • the photon penetration path of ⁇ 1 wavelength is simplified to include
  • its volume changes with time t can be expressed as According to Lambert-Beer's law, the penetration path volume change
  • the difference between the absorption coefficient and the background absorption coefficient is liquid
  • the change in light intensity at the photoreceiver PD can be expressed as
  • the change in light intensity of the photoreceiver in the part that penetrates the arteriole can be expressed as Considering that the proportion of photons penetrating the arterioles is affected by the skin structure and body temperature, the present invention introduces To correct the coefficient, the photoplethysm wave received by the light source with wavelength ⁇ 1 at the photoelectric receiver can be expressed by the following formula:
  • the wavelength of LED1-1 is ⁇ 2 ;
  • the photon penetration path of ⁇ 2 wavelength is simplified to include
  • its volume changes with time t can be expressed as According to Lambert-Beer's law, the penetration path volume change
  • the difference between the absorption coefficient and the background absorption coefficient is liquid, the change in light intensity received by the photoelectric receiver can be expressed as
  • due to its shallow penetration depth only a small part of the photons penetrates the arteriole, and the change in light intensity received at the photoelectric receiver can be expressed as
  • the photoplethysm wave received by the light source with wavelength ⁇ 2 at the photoelectric receiver can be expressed by the following formula:
  • ⁇ is the variance
  • n is the number of points
  • n is the average of n points.
  • the volume pulse wave signal at the small artery can be obtained, which can be used for subsequent cardiovascular status assessment and blood pressure prediction.
  • the present invention solidifies the group model (ie, blood pressure prediction model group) and coefficients into the microprocessing unit.
  • the microprocessing unit first detects the multi-channel pulse wave through the pulse wave detection unit, controls the ratchet structure through the motor, changes the length of the constant pressure spring and maintains a constant force Fi , so that at this Fi value, the AC value amplitude of the PPG Maximum, record the temperature value at this time, and calculate the cross-correlation coefficient between multi-channel PPG signals.
  • the cross-correlation coefficient > 0.5 and maintain stability for t (t > 3s) time confirm the scaling coefficient rat, according to Calculate APPG.
  • a calibration is performed for individual data, and the coefficients of the last layer of the blood pressure prediction model are updated through the transfer learning method to achieve personalized blood pressure detection. That is, cardiovascular status screening is first performed through feature engineering, which is divided into several different status s such as NORMAL, AF, and CA. Then, from the blood pressure prediction model group, the blood pressure prediction model coefficient w i corresponding to the different status s i is selected for blood pressure prediction. ;where s i represents the i-th cardiovascular state, w i represents the blood pressure prediction model parameters corresponding to the i-th cardiovascular state.
  • the coefficients of the last layer of the blood pressure prediction model are updated through input parameters such as the individual N seconds of PPG signal, age, gender, and past history.
  • An individual transfer learning model experiment shows that the blood pressure value predicted according to the model described in the present invention meets the Level A requirements of the BHS standard, as shown in Table 2 below.
  • the present invention realizes PPG signal detection at different penetration depths by setting up a multi-wavelength photoelectric detection sensor module. It combines a motor, a constant pressure spring and a pressure sensor to realize pressure control at the measurement part. At the same time, it integrates the temperature of the detection part and removes interference signals. , realizing direct measurement and tracking of APPG signals. Taking into account the shortcomings of traditional non-invasive blood pressure detection accuracy, the blood pressure model was improved, and a transfer learning model was designed to address the problem of large calculations of traditional deep learning algorithms. Through model training, APPG-based non-invasive heart detection was realized. Initial screening of vascular status and prediction of blood pressure.

Landscapes

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

Abstract

本发明公开了一种基于小动脉光电容积脉搏波的血压检测装置及方法。该装置包括指环结构、设置在指环结构底端的脉搏波检测单元、温度传感单元、压力传感单元和微控制单元以及设置在指环结构内部两侧的弹簧控制单元。在根据压力值确定的弹簧控制单元的调节量下，微控制单元根据脉搏波检测单元检测到的光电容积脉搏波、以及温度传感单元检测到的温度值去计算小动脉光电容积脉搏波，并根据小动脉光电容积脉搏波进行血压检测。本发明结构简单且操作方式简便，所获得的信号信噪比高，准确度高，能够实现APPG信号的直接测量和跟踪，并且通过多个模型的训练，实现了基于APPG的无创心血管状态初筛和血压预测。

Description

一种基于小动脉光电容积脉搏波的血压检测装置
本申请要求于2022年5月26日提交中国专利局、申请号为202210584491.3、发明名称为“一种基于小动脉光电容积脉搏波的血压检测装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及血压检测技术领域，特别是涉及一种基于小动脉光电容积脉搏波的血压检测装置。
背景技术
血压(BP，Blood Pressure)是人体重要的生理参数，能够反映心血管功能状况。高血压会增加人体患脑卒中、冠状动脉疾病、心力衰竭、房颤和周围血管病等多种心血管疾病的风险。监测不同心血管状态下的血压变化趋势，有利于更好的理解心血管疾病发病机制并实施有效管控。
在利用PPG信号预测血压时，根据人体生理结构，如果要获得比较稳定的PPG信号，一般要在人体血管丰富的部位进行测量，目前主要是在耳垂或者手腕上测量，但是这些部位的信号质量差、信噪比较低。
除此之外，在利用PPG信号进行血压预测时，当前的常规方法是直接测量PPG信号进而去预测血压。但是由于不同波长的穿透深度不一样，直接测得的PPG信号中包含很多容积波动，比如有毛细血管、小动脉、组织液、甚至是静脉的容积波动，使得信号干扰较大，PPG信号测量不准确。因此在预测血压时需要剔除掉其他PPG干扰信号，而获得小动脉处的光电容积脉搏波并利用其进行血压预测是最直接且最准确的。
发明内容
本发明的目的是提供一种基于小动脉光电容积脉搏波的血压检测装置，用以提高血压预测的精度。
为实现上述目的，本发明提供了如下方案：
一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，包括指环结构、设置在所述指环结构底端的脉搏波检测单元、温度传感单元、压力传感单元和微控制单元以及设置在所述指环结构内部两侧的弹簧控制单元；
脉搏波检测单元，用于检测用户指端的光电容积脉搏波；
压力传感单元，用于检测用户指端的接触压力值；
温度传感单元，用于采集用户指端的温度值；
微控制单元，分别与所述脉搏波检测单元、所述压力传感单元以及所述温度传感单元连接，用于根据所述光电容积脉搏波、所述压力值以及所述温度值计算小动脉光电容积脉搏波，并根据所述小动脉光电容积脉搏波进行血压检测；
弹簧控制单元，与所述压力传感单元连接，用于根据所述压力值对所述压力传感单元进行调节。
进一步的，所述脉搏波检测单元包括光电传感器模组，所述光电传感器模组用于检测用户指端的光电容积脉搏波，并通过模拟前端或接口电路将所述光电容积脉搏波发送至所述微控制单元；
所述压力传感单元包括压力传感器，所述压力传感器用于检测用户指端的接触压力值，并通过模拟前端或接口电路将所述压力值发送至所述微控制单元；
所述温度传感单元包括温度传感器，所述温度传感器用于采集用户指端的温度值，并通过模拟前端或接口电路将所述温度值发送至所述微控制单元；
进一步的，所述血压检测装置还包括IMU检测单元，所述IMU检测单元分别与所述脉搏检测单元、所述压力传感单元、所述温度传感单元以及所述微控制单元连接，用于检测所述脉搏检测单元、所述压力传感单元以及所述温度传感单元检测的数据是否存在干扰数据，并将去除干扰后的数据发送至所述微控制单元。
进一步的，所述弹簧控制单元包括精密电机、棘轮结构和恒压弹性模块；所述恒压弹性模块包括恒压弹簧、可移动夹具和固定夹具；所述精密电机根据所述压力值带动所述棘轮结构转动来调节所述恒压弹簧的长度；所述可移动夹具套设在所述恒压弹簧外部，所述固定夹具套设在所述可移动夹具的上部；所述恒压弹簧的顶端与所述棘轮结构连接，所述恒压弹的底端与所述指环结构的底端连接；所述固定夹具以及所述可移动夹具固定在所述指环结构内。
进一步的，所述光电传感器模组包括多个发光LED和多个光电接收器；多个所述发光LED关于多个所述光电接收器中心对称设置。
进一步的，所述光电传感器模组包括多个发光LED和1个光电接收器；多个所述发光LED围绕所述光电接收器设置。
进一步的，根据所述光电容积脉搏波、所述压力值以及所述温度值计算小动脉光电容积脉搏波，并根据所述小动脉光电容积脉搏波进行血压检测，具体包括：
根据所述光电容积脉搏波，采用心血管状态初筛模型，确定用户的心血管状态；
根据用户的心血管状态从血压预测模型组中选择对应的血压预测模型；
根据所述光电容积脉搏波、所述压力值以及所述温度值，计算小动脉光电容积脉搏波；
根据所述小动脉光电容积脉搏波，采用选择的血压预测模型进行血压检测。
进一步的，所述心血管状态初筛模型的训练过程包括：
构建训练数据集；所述训练数据集包括不同心血管状态、不同性别以及不同年龄的光电容积脉搏波和血压数据；所述心血管状态包括正常、心房颤动和动脉粥样硬化；
将所述光电容积脉搏波划分为第一组训练数据以及第二组训练数据；
对所述第一组训练数据进行特征信息提取，得到多种特征信息；
通过所述多种特征信息以及光电容积脉搏波对应的心血管状态对机器学习模型进行训练，得到心血管状态初筛模型。
进一步的，所述血压预测模型组的训练过程包括：
通过不同心血管状态下的第二组训练数据以及对应的血压数据训练多个深度学习模型，得到血压预测模型组；所述血压预测模型组包括多个血压检测模型，所述血压检测模型用于检测心血管状态下的血压。
进一步的，根据所述光电容积脉搏波、所述压力值以及所述温度值，计算小动脉光电容积脉搏波，具体包括：
根据压力值确定所述弹簧控制单元的调节量；
获取在所述弹簧控制单元调节量下的所述LED波长为λ1和λ2时的光电容积脉搏波；
根据所述光电容积脉搏波确定缩放系数；
根据温度值确定所述缩放系数的修订项；
根据所述光电容积脉搏波、所述缩放系数以及所述缩放系数的修订项计算小动脉光电容积脉搏波。
根据本发明提供的具体实施例，本发明公开了以下技术效果：
(1)本发明的基于小动脉光电容积脉搏波的血压检测装置，采用指环的形式在血管丰富的手指部位进行测量，结构简单且操作方式简便，所获得的信号信噪比高，准确度高，实现了人体主动脉血压的连续动态实时采集，具有较高的医用价值和广阔的市场应用前景。
(2)本发明通过设置多波长光电检测传感器模组实现了不同穿透深度的PPG信号检测，结合电机、恒压弹簧和压力传感器实现测量部位的压力控制，同时融合检测部位的温度，去除掉了干扰信号，实现了APPG信号的直接测量和跟踪。
说明书附图
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置示意图；
图2为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的指环结构示意图；
图3为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的指环结构剖面图；
图4为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的恒压弹性模块结构示意图；
图5为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的恒压弹性模块反馈控制示意图；
图6为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的指环结构底端示意图；
图7为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的小动脉光电容积脉搏波检测实施原理图；
图8为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的小动脉光电容积脉搏波、压力、温度波形示意图；
图9为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的小动脉光电容积脉搏波、压力、温度检测实施原理框图；
图10为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的APPG信号检测原理框图；
图11为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的群体模型学习训练流程图；
图12为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的特征提取示意图；
图13为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的心血管状态初筛流程图；
图14为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的群体模型构建示意图；
图15为本发明实施例提供的基于小动脉光电容积脉搏波的血压检测装置的迁移学习模型构建示意图；
其中，图中各符号标记：
1：光电传感器模组；
1-1：波长为λ1的LED光源；
1-2：波长为λ2的LED光源；
1-3：光电接收器PD
1-1-1：波长为λ1的LED光在1-3处接收到的脉搏波；
1-2-1：波长为λ2的LED光在1-3处接收到的脉搏波；
1-12-1：波长为λ1，λ2的LED光合成的小动脉脉搏波；
2-温度传感器；
3-压力传感器；
4-恒压弹性模块；
4-1：恒压弹簧；
4-2：可移动夹具；
4-3：固定夹具；
4-4：顶部螺纹孔；
4-5：底部螺纹孔；
4-6：顶部连接线；
4-7：底部连接线；
5-电池；
6-1：指环底部右侧螺纹孔；
6-2：指环底部左侧螺纹孔；
6-3：微控制单元；
6-4：电子元件；
6-5：传感器弹簧结构。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明的目的是提供一种基于小动脉光电容积脉搏波的血压检测装置，用以提高血压预测的精度。
为使本发明的上述目的、特征和优点能够更加明显易懂，下面结合附图和具体实施方式对本发明作进一步详细的说明。
如图1-图6所示，一种基于小动脉光电容积脉搏波的血压检测装置，包括指环结构、设置在指环结构底端的脉搏波检测单元、温度传感单元、压力传感单元和微控制单元以及设置在指环结构内部两侧的弹簧控制单元。
脉搏波检测单元用于检测用户指端的光电容积脉搏波。
压力传感单元用于检测用户指端的接触压力值。
温度传感单元用于采集用户指端的温度值。
微控制单元分别与所述脉搏波检测单元、所述压力传感单元以及所述温度传感单元连接。所述微控制单元用于根据所述光电容积脉搏波、所述压力值以及所述温度值计算小动脉光电容积脉搏波，并根据所述小动脉光电容积脉搏波进行血压检测。微控制单元包含蓝牙BLE低功耗协议栈，微控制单元用于控制各检测单元采集信号数据，并对信号进行分析、特征提取和血压计算，然后通过蓝牙协议栈向终端发送血压检测结果。
弹簧控制单元与所述压力传感单元连接。所述弹簧控制单元用于根据所述压力值对所述压力传感单元进行调节。
血压检测装置还包括IMU检测单元。所述IMU检测单元分别与脉搏检测单元、压力传感单元、温度传感单元以及微控制单元连接；所述IMU检测单元用于检测脉搏检测单元、压力传感单元以及温度传感单元检测的数据是否存在干扰数据，并将去除干扰后的数据发送至微控制单元。
血压检测装置还包括电源管理单元和存储单元。电源管理单元用于为耗电单元供电；存储单元用于存储本地数据。
其中，所述脉搏波检测单元包括光电传感器模组1。所述光电传感器模组1用于检测用户指端的光电容积脉搏波，并通过模拟前端或接口电路将所述光电容积脉搏波发送至所述微控制单元6-3。
所述压力传感单元包括压力传感器3。所述压力传感器3用于检测用户指端的接触压力值，并通过模拟前端或接口电路将所述压力值发送至所述微控制单元6-3。
所述温度传感单元包括温度传感器2。所述温度传感器2用于采集用户指端的温度值，并通过模拟前端或接口电路将所述温度值发送至所述微控制单元6-3。
弹簧控制单元包括精密电机、棘轮结构和恒压弹性模块4。精密电机根据压力值带动棘轮结构转动来调节恒压弹簧4-1的长度；恒压弹性模块4包括恒压弹簧4-1、可移动夹具4-2和固定夹具4-3；可移动夹具4-2套设在恒压弹簧4-1外部；固定夹具4-3套设在可移动夹具4-2的上部；恒压弹簧4-1的顶端与棘轮结构连接；恒压弹簧4-1的底端与指环结构的底端连接；固定夹具4-3以及可移动夹具4-2固定在指环结构内。
血压检测装置各个单元均为可拆卸结构，且通过模块化设计组装，可以通过拆卸的方式将传感、压力控制、电源控制等部分进行分离。
如图2、3、4、6所示，压力传感器3、光电传感器模组1、温度传感器2组成传感检测模块；该传感检测模块位于指腹部位，通过指环底部右侧螺纹孔6-1、指环底部左侧螺纹孔6-2以及恒压弹性模块的底部螺纹孔4-5与恒压弹性模块4连接。
光电传感器模组1可以由多个发光LED和多个光电接收器PD组成，可以是多个LED(如1-1、1-2)围绕1个PD(如1-3)形成圆形的阵列，也可以是LED如(1-1、1-2)关于PD(如1-3)中心对称的结构。
指环结构顶端主要包括棘轮结构和电池5。指环结构通过顶端左右两侧的螺纹孔和恒压弹性模块的顶部螺纹孔4-4与恒压弹性模块4连接。
恒压弹性模块4位于指环两侧，呈现对称排列，左右两侧的结构相同。恒压弹性模块4通过顶部连接线4-6与棘轮结构连接；恒压弹性模块4通过底部连接线4-7与传感器检测模块连接。精密电机带动棘轮结构转动时，棘轮结构会控制恒压弹簧4-1的长度，调节传感检测模块与指腹之间的压力值。此外，位于传感器检测模块下方的传感器弹簧结构6-5为了保持传感器检测模块与指腹的充分接触，传感器弹簧结构6-5的长度也会随着恒压弹簧4-1的长度变化进行二次调节。
压力调节功能的实现如图5所示，微控制单元6-3通过压力传感单元的模拟前端或接口电路检测压力传感器3处测量得到的压力值；同时，微控制单元6-3检测光电传感器模组1处测量的PPG信号。指环刚佩戴上时的压力值记为F0，此时对应的PPG信号记为PPG0，微控制单元6-3控制精密电机转动，带动棘轮结构顺时针旋转N步(N≥1)，棘轮结构通过顶部连接线4-6拉伸恒压弹簧4-1，进而拉伸整个传感器检测模块，此时传感器弹簧结构6-5的弹性势能增加，记录此时压力传感器3测量的压力值F1以及对应的PPG1，比较PPG0和PPG1对应的交流幅值AC_PPG0和AC_PPG1大小；如果AC_PPG0>AC_PPG1则棘轮结构逆时针旋转M步(N≥M≥0)，反之如果AC_PPG0≤AC_PPG1则继续顺时针旋转N步。通过循环操作这个方法，确认棘轮结构的合适位置i，记录此时压力传感器的值Fi，使得AC_PPGi在当前位置下取得最大值。
从而，根据压力传感器测量的压力值，弹簧控制单元通过精密电机带动棘轮结构转动调节恒压弹簧4-1的长度进而实现了对传感器检测模块的调节，即拉伸或压缩压力传感器3，调节压力传感器3与指腹之间的压力值。
血压检测装置中的存储单元、电阻、电容等电子元件6-4集成在指环底端。
本发明的基于小动脉光电容积脉搏波的血压检测装置，采用指环的形式在血管丰富的手指部位进行测量，结构简单且操作方式简便，所获得的信号信噪比高，准确度高，实现了人体主动脉血压的连续动态实时采集，具有较高的医用价值和广阔的市场应用前景。
本发明提供的小动脉光电容积脉搏波的血压检测装置中的微控制单元的具体实现过程如下：
步骤1、根据所述光电容积脉搏波，采用心血管状态初筛模型，确定用户的心血管状态。
步骤2、根据用户的心血管状态从血压预测模型组中选择对应的血压预测模型。
步骤3、根据所述光电容积脉搏波、所述压力值以及所述温度值，计算小动脉光电容积脉搏波。
步骤4、根据所述小动脉光电容积脉搏波，采用选择的血压预测模型进行血压检测。
其中，步骤1中的心血管状态初筛模型的训练过程包括：
步骤11：构建训练数据集；所述训练数据集包括不同心血管状态、不同性别以及不同年龄的光电容积脉搏波和血压数据；所述心血管状态包括正常、心房颤动和动脉粥样硬化。
步骤12：将所述光电容积脉搏波划分为第一组训练数据以及第二组训练数据。
步骤13：对所述第一组训练数据进行特征信息提取，得到多种特征特征信息。
步骤14：通过所述多种特征信息以及光电容积脉搏波对应的心血管状态对机器学习模型进行训练，得到心血管状态初筛模型。
在具体实施例中，如图11所示，首先选择同时包含ABP，PPG和心血管状态，年龄，性别的公开数据集，如MIMIC数据集或者桂林大学公布的数据集，进行数据预处理，PPG和ABP所需的最小长度设置为5分钟，并删除所有较短长度的记录。使用四阶巴特沃斯带通滤波器来消除PPG信号中的基线漂移和高频噪声，PPG信号的通频带频率为0.5-8Hz，利用Hampel滤波器去除ABP信号中的尖峰噪声。
通过公开数据集构建以不同心血管状态作为划分依据的光电容积脉搏波(PPG)；所述公开数据集包括心血管状态、性别(Gender)、年龄(Age)和血压(BP)的数据子集{NORMAL},{AF},{CA}；预处理后的PPG信号分为2种(即第一组训练数据和第二组训练数据)；第一种是用于特征工程提取特征，作为机器学习模型的输入，用于训练模型参数并输出心血管状态划分的初筛分类结果；第二种是作为深度学习模型的输入，用于血压预测。其中特征工程中提取的特征信息包括但不限于心率HR，峰谷值Peak，上升时间RT，下降时间DT，波形面积PA，心率变异性(HRV)等，如图12(a)所示，也包括频域、时频域，非线性特征如样本熵等特征。特征信息用于实现心血管状态分类的训练、验证和测试，测试准确率超过90％后固化模型参数，实现心血管状态筛查。同时，现有的研究证明，脉搏波的强度、形态、节律及速率等特征会随着血管几何形态和力学性质的改变而改变。不同的心血管状态对应的脉搏波形态有较为明显的差异；如图12(b)所示，正常心血管状态良好的人的脉搏波形态包括了较为明显的降中峡，形态规律，而存在心血管状态异常的人的脉搏波形态在波形面积，降中峡，心率及心率变异性上都存在明显的差异。有鉴于此，可设计不同的分类模型进行心血管状态的分类评估。
分类模型包括：
1)贝叶斯分类器：贝叶斯定理是这类算法的核心，因此统称为贝叶斯分类。贝叶斯决策论通过相关概率已知的情况下利用误判损失来选择最优 的类别分类。
2)决策树分类器：决策树算法将多维特征的训练集样本分配到不同类别中去，相当于给训练集样本做了一次投影，经过投影后的训练集样本被赋予相应的类别标签。决策树算法用递归流程来表示分类与回归，每次通过选择特征集中最优特征，根据选择的最优特征对训练集进行分类和回归，保证每个训练集样本所经历的分类和回归过程都是最优的。如果训练集样本的特征数量较多，则开始生产决策树时对特征进行选择，留下有较强能力对训练的训练集样本进行分类和回归的特征，去掉那些能力较弱的特征。整个递归过程形似一棵树，所以称为决策树(Decision Tree)算法。决策树算法核心是构造一个准确度高，复杂度小的决策树。决策树的构造过程包括决策树的生成与剪枝两个部分。决策树的生成即利用训练集样本生成决策树，决策树的顶端节点为根节点，每个分支都看作是一个决策节点，代表待分类对象的属性。每个节点都表示存在一种可能的分类过程。遍历决策树时，在每个决策节点的不同的选择都会导致输出不同的分支，最终抵达叶子节点进行输出。该遍历的过程就是利用决策树解决分类问题的过程，利用若干个属性来对训练集样本分类。决策树的剪枝就是对生成的决策树进行检验和修正的过程，通过对决策树去复杂化，使之具有更好的泛化能力，能够适应更多的解决方案。
3)随机森林分类器：Random Forest(随机森林)是一种基于树模型的Bagging的优化版本，一棵树的生成肯定还是不如多棵树，因此就有了随机森林，解决决策树泛化能力弱的特点。而同一批数据，用同样的算法只能产生一棵树，这时Bagging策略可以产生不同的数据集。Bagging策略来源于bootstrap aggregation：从样本集(假设样本集N个数据点)中重采样选出Nb个样本(有放回的采样，样本数据点个数仍然不变为N)，在所有样本上，对这N个样本建立分类器(ID3\C4.5\CART\SVM\LOGISTIC)，重复以上两步m次，获得m个分类器；最后根据这m个分类器的投票结果，决定数据属于哪一类。
此外，一种典型的分类示范例如图13所示，将特征工程产生的多维度时域、频域、时频域、及非线性特征如样本熵等特征作为输入层，为了避免过拟合，光电接收器年龄数据在作为心血管状态分类的特征数据训练 时，使用数据合成策略，对年龄特征加入随机高斯噪声，将相同年龄的数据扩充到±3或±5的范围内。在模型层利用LSTM或Bi-LSTM构建多层反馈网络，设置Densy层与softmax层输出分类的结果。其中模型层也可以是RF，SVM等其他模型。以下表格初步显示了模型分为Normal,AF和CA三类心血管状态的平均精度。
表一 分类精度结果
本发明采用随机森林分类器作为心血管状态初筛的机器学习模型。在实际应用中，可根据计算资源来适当选择分类器。
其中，步骤2中的血压预测模型组的训练过程包括：
通过不同心血管状态下的第二组训练数据以及对应的血压数据训练多个深度学习模型，得到血压预测模型组；所述血压预测模型组包括多个血压检测模型，所述血压检测模型用于检测心血管状态下的血压。
在具体实施例中，如图14所示，针对每一种心血管状态初筛模型的分类结果，基于深度学习的方法建立了血压预测模型组。本发明开发了一种新的模型架构，即循环-注意力神经网络(LSTM-AT)，基于长短期记忆，注意力机制和全连接神经网络。模型的输入信号为一定时间周期内的脉搏波信号片段，利用数据子集数据中的PPG片段作为输入信号，BP作为标签信号来训练深度学习血压预测模型，构建适用于不同数据子集的血压预测模型组，并固化血压预测模型的参数作为群体模型系数，血压预测模型的输入为PPG数据片段，经过Bi-LSTM层并加入注意力机制Attension层对重要特征向量进行加权，将输出向量与年龄、性别等人口统计学信息结合，最后通过一个全连接层进行血压预测。血压预测模型中应用了Bahdanau注意力:其中，ci保留前一层输出的所有隐藏状态信息并表示为注意权重向量ai和hti之间的加权和，ai使用softmax函数标准化，hti代表了i时刻的ht。
其中，长短时记忆(LSTM)是一种特殊的RNN，可以有效地解决训练中梯度消失和爆发的问题，LSTM单元的结构图如图14所示。与RNN相比，LSTM有两种传输状态：Ct(细胞状态)和ht(隐藏状态)。其中，Ct变化非常缓慢，在不同的节段，ht会有所不同，分别对应于长期记忆和短期记忆。在内部，LSTM通过遗忘门、输入门和输出门将信息从一个时间步传输到另一个时间步。
迁移学习侧重于将源领域获得的知识存储到另一个目标领域中，该问题通常包含少量的数据样本来训练模型。本发明将基于新的模型架构，即循环-注意力神经网络(LSTM-AT)，基于长短期记忆，注意力机制和全连接神经网络，从大型公开数据集中获取的PPG和BP数据来预先训练参数，在个体使用过程中利用新数据调节个体模型最后一层的参数，可以大幅减少新数据所需的训练集数量，如图15所示。
其中，步骤3具体包括：
步骤31：根据压力值确定所述弹簧控制单元的调节量。
步骤32：获取在所述弹簧控制单元调节量下的所述LED波长为λ1和λ2时的光电容积脉搏波。
步骤33：根据所述光电容积脉搏波信号确定缩放系数。
步骤34：根据温度值确定所述缩放系数的修订项。
步骤35：根据所述光电容积脉搏波、所述缩放系数以及所述缩放系数的修订项计算小动脉光电容积脉搏波。
其中，脉搏波检测单元检测到的光电容积脉搏波包括：在所述弹簧控制单元调节量下的所述LED波长为λ1时的光电容积脉搏波，和在所述弹簧控制单元调节量下的所述LED波长为λ2时的光电容积脉搏波。
在具体实施例中，以1个光电接收器1-3，LED1-1和LED1-2为例，LED1-1的波长为λ2，LED1-2的波长为λ1，其中λ1≥λ2，可以看到λ1的光穿透深度较深，光子在传播路径中包含了毛细血管和小动脉的容积变化信号，λ2的穿透深度较浅，光子传播路径主要包括毛细血管的容积变化。由于此时光子在皮肤组织中传播主要是以漫反射的形式存在，容易受到组织液、静脉波动的影响，难以准确获取某一特定层的信号。如图7所示，PD处检测得到的PPG信号通常是一个复合信号，其中1-1-1为波长λ1的LED光在光电接收器1-3处接收到的脉搏波；1-2-1为波长λ2的LED光在光电接收器1-3处接收到的脉搏波；1-12-1为波长为λ1，λ2的LED光合成的小动脉脉搏波。
此外，PPG信号除了受到光子传播路径的影响，还受压力、温度信号的影响。如图8所示，随着压力的增加，脉搏波的交流AC幅度先增加，后减少，在这个过程中，小动脉经历了去负荷到阻塞的过程；随着温度的降低，脉搏波的交流AC幅度减少，主要是温度降低时，血管收缩，内径减少，外周阻力增加，灌注减少。
如图9、图10所示，其输入信号为压力传感器的值Fi，温度Ti，浅层PPG，深层PPG，输出为小动脉APPG信号。
LED1-2的波长为λ1，λ1波长的光子穿透路径简化包含毛细血管部分，其容积随时间t的变化可表示为根据朗伯比尔定律，穿透路径容积变化吸收系数与背景吸收系数差值为的液体，在光电接收器PD处光强的变化可以表示为同样，由于其穿透深度较深，穿透小动脉的部分在光电接收器光强的变化可以表示为考虑到穿透小动脉部分光子的比例收到皮肤结构和体温的影响，本发明引入来修正系数，波长为λ1的光源在光电接收器处接收到的光电容积脉搏波具体可以由以下公式表述：
同理，LED1-1的波长为λ2；λ2波长的光子穿透路径简化包含毛细血管部分，其容积随时间t的变化可表示为根据朗伯比尔定律，穿透路径容积变化吸收系数与背景吸收系数差值为的液体，在光电接收器接收到的光强变化可以表示为同样，由于其穿透深度较浅，只有小部分光子穿透小动脉，在光电接收器接收到的光强变化可以表示为考虑到穿透小动脉部分光子的比例收到皮肤结构和体温的影响，本发明引入来修正系数，波长为λ2的光源在光电接收器处接收到的光电容积脉搏波可以由以下公式表述：
在计算系数时，可以设置λ1＞＞λ2，λ1的光穿透深度较深，光子在传播路径中包含了毛细血管和小动脉的容积变化信号；λ2的穿透深度较浅，光子传播路径主要包括毛细血管的容积变化。此时 的值均为常数，为了简化模型计算，忽略掉常数项，同时使得可以得到以下公式：
可以看出对不同波长的光源求差分可以得到APPG；因此，需要明确缩放系数rat(t)的计算方法。考虑到缩放系数rat(t)的计算与t时刻不同光源穿透深度，动脉、静脉及组织的收缩和扩张相关，具体确定流程如下：
1)rat取值0-10，以0.01为增量增加，rat每取一个值计算一组信号矩阵。
2)计算每一组PPG与之间的互相关系数，计算公式如下：
其中，σ为方差，n为点数，为n点的平均值。当互相关系数>0.5且维持t(t>3s)时间的稳定，确认当前数据对应的rat为缩放系数rat的取值。
3)通过多组实验，评估温度对PPG幅度的影响，保持心率等因素恒定，记录PPG幅度随着温度变化的趋势，以υT表示温度对PPG幅度的修订项。
4)通过以上步骤2)确认的rat，步骤3)确认的υT，通过以下公式确定APPG的值。
其中，和分别是波长为λ1和λ2时测量得到的幅值，υT表示温度对PPG幅度的修订项，作为补偿可以得到更准确的动脉容积波波形。通过以上步骤，可以得到小动脉处的容积脉搏波信号，用于后续的心血管状态评估与血压预测。
本发明将群体模型(即血压预测模型组)及系数固化入微处理单元。微处理单元通过脉搏波检测单元先检测多通道脉搏波，通过电机控制棘轮结构，改变恒压弹簧的长度并维持恒定的作用力Fi，使得在该Fi值下，PPG的交流AC值幅度最大，记录此时的温度值，并计算多通道PPG信号之间的互相关系数，当互相关系数>0.5且维持t(t>3s)时间的稳定，确认缩放系数rat，根据计算APPG。针对个体的数据做一次校准，通过迁移学习方法，更新血压预测模型最后一层的系数，实现个性化血压检测。即先通过特征工程进行心血管状态筛查，分为NORMAL、AF、CA等几个不同状态s，然后从血压预测模型组中，选择不同状态si对应的血压预测模型系数wi进行血压预测；其中si代表第i种心血管状态，wi代表第i种心血管状态对应的血压预测模型参数。在个体模型(即血压预测模型)的建模过程中，通过个体N秒的PPG信号、年龄、性别和既往史等输入参数，更新血压预测模型最后一层的系数。一项针对个体的迁移学习模型实验表明，按照本发明阐述的模型预测的血压值符合BHS标准的A级要求，如下表二所示。
表二 迁移学习模型测试结果与BHS标准对比

本发明通过设置多波长光电检测传感器模组实现了不同穿透深度的PPG信号检测，结合电机、恒压弹簧和压力传感器实现测量部位的压力控制，同时融合检测部位的温度，去除掉了干扰信号，实现了APPG信号的直接测量和跟踪。并且考虑到传统无创血压检测准确度不理想的缺陷，对血压模型进行改进，并且针对传统的深度学习算法计算量大的问题设计了迁移学习模型，通过模型的训练，实现了基于APPG的无创心血管状态初筛和血压预测。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。
本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处。综上所述，本说明书内容不应理解为对本发明的限制。

Claims (10)

  1. 一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，包括指环结构、设置在所述指环结构底端的脉搏波检测单元、温度传感单元、压力传感单元和微控制单元以及设置在所述指环结构内部两侧的弹簧控制单元；
    脉搏波检测单元，用于检测用户指端的光电容积脉搏波；
    压力传感单元，用于检测用户指端的接触压力值；
    温度传感单元，用于采集用户指端的温度值；
    微控制单元，分别与所述脉搏波检测单元、所述压力传感单元以及所述温度传感单元连接，用于根据所述光电容积脉搏波、所述压力值以及所述温度值计算小动脉光电容积脉搏波，并根据所述小动脉光电容积脉搏波进行血压检测；
    弹簧控制单元，与所述压力传感单元连接，用于根据所述压力值对所述压力传感单元进行调节。
  2. 根据权利要求1所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，所述脉搏波检测单元包括光电传感器模组，所述光电传感器模组用于检测用户指端的光电容积脉搏波，并通过模拟前端或接口电路将所述光电容积脉搏波发送至所述微控制单元；
    所述压力传感单元包括压力传感器，所述压力传感器用于检测用户指端的接触压力值，并通过模拟前端或接口电路将所述压力值发送至所述微控制单元；
    所述温度传感单元包括温度传感器，所述温度传感器用于采集用户指端的温度值，并通过模拟前端或接口电路将所述温度值发送至所述微控制单元。
  3. 根据权利要求1所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，所述血压检测装置还包括IMU检测单元，所述IMU检测单元分别与所述脉搏检测单元、所述压力传感单元、所述温度传感单元以及所述微控制单元连接，用于检测所述脉搏检测单元、所述压力传感单元以及所述温度传感单元检测的数据是否存在干扰数据，并将去除干扰后的数据发送至所述微控制单元。
  4. 根据权利要求1所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，所述弹簧控制单元包括精密电机、棘轮结构和恒压弹性模块；所述恒压弹性模块包括恒压弹簧、可移动夹具和固定夹具；所述精密电机根据所述压力值带动所述棘轮结构转动来调节所述恒压弹簧的长度；所述可移动夹具套设在所述恒压弹簧外部，所述固定夹具套设在所述可移动夹具的上部；所述恒压弹簧的顶端与所述棘轮结构连接，所述恒压弹簧的底端与所述指环结构的底端连接；所述固定夹具以及所述可移动夹具固定在所述指环结构内。
  5. 根据权利要求2所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，所述光电传感器模组包括多个发光LED和多个光电接收器；多个所述发光LED关于多个所述光电接收器中心对称设置。
  6. 根据权利要求2所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，所述光电传感器模组包括多个发光LED和1个光电接收器；多个所述发光LED围绕所述光电接收器设置。
  7. 根据权利要求1所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，根据所述光电容积脉搏波、所述压力值以及所述温度值计算小动脉光电容积脉搏波，并根据所述小动脉光电容积脉搏波进行血压检测，具体包括：
    根据所述光电容积脉搏波，采用心血管状态初筛模型，确定用户的心血管状态；
    根据用户的心血管状态从血压预测模型组中选择对应的血压预测模型；
    根据所述光电容积脉搏波、所述压力值以及所述温度值，计算小动脉光电容积脉搏波；
    根据所述小动脉光电容积脉搏波，采用选择的血压预测模型进行血压检测。
  8. 根据权利要求7所述的基于光电容积脉搏波的血压检测装置，其特征在于，所述心血管状态初筛模型的训练过程包括：
    构建训练数据集；所述训练数据集包括不同心血管状态、不同性别以及不同年龄的光电容积脉搏波和血压数据；所述心血管状态包括正常、心房颤动和动脉粥样硬化；
    将所述光电容积脉搏波划分为第一组训练数据以及第二组训练数据；
    对所述第一组训练数据进行特征信息提取，得到多种特征信息；
    通过所述多种特征信息以及光电容积脉搏波对应的心血管状态对机器学习模型进行训练，得到心血管状态初筛模型。
  9. 根据权利要求8所述的基于光电容积脉搏波的血压检测装置，其特征在于，所述血压预测模型组的训练过程包括：
    通过不同心血管状态下的第二组训练数据以及对应的血压数据训练多个深度学习模型，得到血压预测模型组；所述血压预测模型组包括多个血压检测模型，所述血压检测模型用于检测心血管状态下的血压。
  10. 根据权利要求7所述的一种基于小动脉光电容积脉搏波的血压检测装置，其特征在于，根据所述光电容积脉搏波、所述压力值以及所述温度值，计算小动脉光电容积脉搏波，具体包括：
    根据压力值确定所述弹簧控制单元的调节量；
    获取在所述弹簧控制单元调节量下的所述LED波长为λ1和λ2时的光电容积脉搏波；
    根据所述光电容积脉搏波确定缩放系数；
    根据温度值确定所述缩放系数的修订项；
    根据所述光电容积脉搏波、所述缩放系数以及所述缩放系数的修订项计算小动脉光电容积脉搏波。
PCT/CN2023/094497 2022-05-26 2023-05-16 一种基于小动脉光电容积脉搏波的血压检测装置 WO2023226824A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210584491.3 2022-05-26
CN202210584491.3A CN114983364B (zh) 2022-05-26 2022-05-26 一种基于小动脉光电容积脉搏波的血压检测装置

Publications (1)

Publication Number Publication Date
WO2023226824A1 true WO2023226824A1 (zh) 2023-11-30

Family

ID=83028952

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/094497 WO2023226824A1 (zh) 2022-05-26 2023-05-16 一种基于小动脉光电容积脉搏波的血压检测装置

Country Status (2)

Country Link
CN (1) CN114983364B (zh)
WO (1) WO2023226824A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114983364B (zh) * 2022-05-26 2023-09-15 杭州兆观传感科技有限公司 一种基于小动脉光电容积脉搏波的血压检测装置
CN115770028A (zh) * 2022-12-16 2023-03-10 深圳市爱都科技有限公司 血压检测方法、系统、装置及存储介质
CN116028809A (zh) * 2022-12-27 2023-04-28 北京津发科技股份有限公司 一种连续血压的测量模型训练、测量方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20060081178A (ko) * 2005-01-07 2006-07-12 학교법인연세대학교 광혈류 측정신호를 이용한 이동형 혈압 감시장치
US20190209030A1 (en) * 2016-08-10 2019-07-11 Murata Manufacturing Co., Ltd. Blood pressure status measuring apparatus
CN110448287A (zh) * 2018-05-07 2019-11-15 香港中文大学 用于非侵入性毛细血管血压测量的装置和方法
CN114287892A (zh) * 2021-12-02 2022-04-08 中国科学院深圳先进技术研究院 一种外周血管阻力变化跟踪方法、系统、终端以及存储介质
CN114983364A (zh) * 2022-05-26 2022-09-02 杭州兆观传感科技有限公司 一种基于小动脉光电容积脉搏波的血压检测装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20060081178A (ko) * 2005-01-07 2006-07-12 학교법인연세대학교 광혈류 측정신호를 이용한 이동형 혈압 감시장치
US20190209030A1 (en) * 2016-08-10 2019-07-11 Murata Manufacturing Co., Ltd. Blood pressure status measuring apparatus
CN110448287A (zh) * 2018-05-07 2019-11-15 香港中文大学 用于非侵入性毛细血管血压测量的装置和方法
CN114287892A (zh) * 2021-12-02 2022-04-08 中国科学院深圳先进技术研究院 一种外周血管阻力变化跟踪方法、系统、终端以及存储介质
CN114983364A (zh) * 2022-05-26 2022-09-02 杭州兆观传感科技有限公司 一种基于小动脉光电容积脉搏波的血压检测装置

Also Published As

Publication number Publication date
CN114983364A (zh) 2022-09-02
CN114983364B (zh) 2023-09-15

Similar Documents

Publication Publication Date Title
WO2023226824A1 (zh) 一种基于小动脉光电容积脉搏波的血压检测装置
Zhao et al. EmotionSense: Emotion recognition based on wearable wristband
Slapničar et al. Continuous blood pressure estimation from PPG signal
CN106413534B (zh) 连续血压测量装置、测量模型建立方法和系统
CN102488503B (zh) 连续血压测量装置
CN102429649B (zh) 连续血压测量装置
CN102397064B (zh) 连续血压测量装置
Wang et al. Photoplethysmography-based blood pressure estimation combining filter-wrapper collaborated feature selection with LASSO-LSTM model
US20170238858A1 (en) Depression assessment system and depression assessment method based on physiological information
Maqsood et al. A survey: From shallow to deep machine learning approaches for blood pressure estimation using biosensors
TWI500411B (zh) 脈波與體質健康風險評估系統與方法
US20180228383A1 (en) Blood pressure measuring device, blood pressure measurement method and blood pressure measurement program
CN110037668A (zh) 脉搏信号时空域结合模型判断年龄、健康状态及恶性心律失常识别的系统
Roy et al. On-device reliability assessment and prediction of missing photoplethysmographic data using deep neural networks
CN109394203A (zh) 精神障碍康复期情绪监测与干预方法
Kanawade et al. Photoplethysmography based arrhythmia detection and classification
Pandey et al. Design and implementation of a photoplethysmography acquisition system with an optimized artificial neural network for accurate blood pressure measurement
CN104644151B (zh) 一种基于光电容积脉搏信号的压力脉搏波波形传播预测方法
CN114587310A (zh) 一种基于光电容积脉搏波实现有创血压波形估计的方法
Chakraborty et al. PPG-based automated estimation of blood pressure using patient-specific neural network modeling
KR20240015709A (ko) 기계학습을 이용한 비침습적 당화혈색소 또는 혈당 추정 방법 및 장치
Gupta et al. In-vivo estimation of glucose level using PPG signal
Shintomi et al. Effectiveness of the heartbeat interval error and compensation method on heart rate variability analysis
Priyadarshini et al. Review of PPG signal using machine learning algorithms for blood pressure and glucose estimation
Leitner et al. Personalized blood pressure estimation using photoplethysmography and wavelet decomposition

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23810897

Country of ref document: EP

Kind code of ref document: A1