Cuffless blood pressure monitoring system

Abstract

A cuffless blood pressure monitoring system, which relates to the technical field of blood pressure monitoring. The system comprises: a symmetrical dual-channel PPG collection apparatus (101), a feature extraction module (102), and a blood pressure prediction model (103), wherein the symmetrical dual-channel PPG collection apparatus (101) comprises a light-emitting diode, and a first photoelectric detector and a second photoelectric detector, which are symmetrically distributed at two sides of the light-emitting diode; the symmetrical dual-channel PPG collection apparatus (101) is configured to acquire a dual-channel pulse signal; an input end of the feature extraction module (102) is connected to an output end of the symmetrical dual-channel PPG collection apparatus (101); the feature extraction module (102) is configured to extract morphological features and calculation features from the dual-channel pulse signal; an input end of the blood pressure prediction model (103) is connected to an output end of the feature extraction module (102); and the blood pressure prediction model (103) is configured to predict diastolic pressure and systolic pressure according to the morphological features and the calculation features. The symmetrical dual-channel PPG collection apparatus (101) has a small size, low cost and high sensitivity, and a non-invasive, continuous and wearable cuffless blood pressure monitoring apparatus is achieved.

Description

A cuffless blood pressure monitoring system Technical Field

[0001] The embodiments of the present application relate to the technical field of blood pressure monitoring, and in particular to a cuffless blood pressure monitoring system. Background Art

[0002] Postoperative monitoring techniques for patients with intracranial aneurysms include transcranial Doppler ultrasound, medical angiography, blood pressure monitoring, etc. Compared with other complex examination methods, blood pressure monitoring has become the key to postoperative management of patients. From a clinical perspective, abnormal blood pressure fluctuations causing secondary rupture after surgery is also one of the important reasons for its high mortality rate, and therefore it is also a key aspect of clinical monitoring. In recent years, there have been many studies on secondary rupture after surgery caused by blood pressure fluctuations, and the research results have important reference value for clinical prevention and control and monitoring of treatment effects. Existing studies believe that hypertension is a risk factor for the rupture of intracranial aneurysms, especially for postoperative patients, who need to strictly control blood pressure fluctuations to avoid re-rupture. Technical issues

[0003] Existing blood pressure monitoring methods primarily include invasive and non-invasive methods. Invasive arterial blood pressure monitoring involves placing a pressure sensor inside an artery via a catheter to monitor arterial pressure. This method accurately reflects continuous blood pressure changes and can also reflect myocardial contractility, enabling timely adjustments to blood pressure control strategies. However, long-term use of invasive methods can lead to complications such as puncture site infection and localized skin necrosis. Furthermore, operating the equipment requires specialized skills, making it unsuitable for routine blood pressure management in patients undergoing intracranial aneurysm surgery. Non-invasive methods primarily use cuff-type blood pressure monitors, which require compressing the brachial artery with a cuff and recording Korotkoff sounds while releasing the cuff pressure. Traditional auscultatory blood pressure measurement methods, as well as electronic blood pressure monitors based on the same principle, are not suitable for continuous monitoring. If a patient's blood pressure suddenly rises and is not promptly and appropriately reduced, especially at night, the patient may be at risk of re-rupture. Therefore, they are also unsuitable for routine blood pressure management in patients undergoing intracranial aneurysm surgery. Technical Solutions

[0004] The embodiment of the present application provides a cuffless blood pressure monitoring system with the advantages of small size, low cost, high sensitivity, etc., and realizes a non-invasive, continuous, wearable cuffless blood pressure monitoring device.

[0005] To solve the above technical problems, an embodiment of the present application provides a cuffless blood pressure monitoring system, comprising: a symmetrical dual-channel PPG acquisition device, a feature extraction module, and a blood pressure prediction model; the symmetrical dual-channel PPG acquisition device comprises a light-emitting diode and a first photodetector and a second photodetector symmetrically distributed on both sides of the light-emitting diode; the symmetrical dual-channel PPG acquisition device is used to obtain a dual-channel pulse signal; the input end of the feature extraction module is connected to the output end of the symmetrical dual-channel PPG acquisition device; the feature extraction module is used to extract morphological features and calculation features from the dual-channel pulse signal; the input end of the blood pressure prediction model is connected to the output end of the feature extraction module; the blood pressure prediction model is used to predict diastolic pressure and systolic pressure based on the morphological features and calculation features.

[0006] In some exemplary embodiments, in a symmetrical dual-channel PPG acquisition device, the light-emitting diode is located at the center of the symmetrical dual-channel PPG acquisition device; the first photodetector and the second photodetector are symmetrically distributed on the left and right sides of the light-emitting diode; the first photodetector and the second photodetector receive reflected light to form a dual-channel PPG signal.

[0007] In some exemplary embodiments, the blood pressure prediction model is a CNN-LSTM prediction model based on an attention mechanism.

[0008] In some exemplary embodiments, the blood pressure prediction model includes a first convolution module, a second convolution module, a third convolution module and two LSTM modules; the first convolution module and the second convolution module each include a convolution layer and a SENet unit; the third convolution module includes a convolution layer, and each convolution layer is followed by a normalization layer and an activation function.

[0009] In some exemplary embodiments, the SENet unit includes a compression unit and an excitation unit, and the compression unit and the excitation unit are respectively represented as follows:

[0010] (1)

[0011] (2)

[0012] Among them, Squeeze and Excitation represent compression unit and excitation unit respectively; .

[0013] In some exemplary embodiments, a maximum pooling layer is provided between the third convolution module and the two LSTM modules; the LSTM module is used to eliminate the impact of nonlinear physiological feature changes on blood pressure prediction; the LSTM module includes an LSTM unit and a Dropout unit, and the blood pressure prediction model finally uses a fully connected layer to return the predicted value.

[0014] In some exemplary embodiments, the acquisition position of the symmetrical dual-channel PPG acquisition device is the radial artery outside the wrist joint, and the radial artery blood pressure is collected.

[0015] In some exemplary embodiments, the above-mentioned cuffless blood pressure monitoring system further includes: a preprocessing module arranged between the symmetrical dual-channel PPG acquisition device and the feature extraction module; the preprocessing module is used to preprocess the dual-channel pulse signal acquired by the symmetrical dual-channel PPG acquisition device.

[0016] In some exemplary embodiments, the preprocessing module includes a Butterworth filter and a sliding average filter; the preprocessing module uses the Butterworth filter to eliminate interference from the electromyographic signal in the dual-channel PPG signal, and uses the sliding average filter to eliminate the influence of baseline drift in the dual-channel PPG signal; and the amplitude-frequency response is brought close to an ideal state by adjusting the filter order; The amplitude and frequency of the Butterworth filter of order is expressed by formula (3):

[0017] (3)

[0018] The function transfer form is obtained by formula (3), as shown in formula (4):

[0019] (4)

[0020] Calculate the coefficients b0 to b1 in formula (4) according to the required cutoff frequency and filter order. N-1 , complete the design of Butterworth filter. Beneficial effects

[0021] The technical solution provided by the embodiments of the present application has at least the following advantages:

[0022] An embodiment of the present application provides a cuffless blood pressure monitoring system, comprising: a symmetrical dual-channel PPG acquisition device, a feature extraction module, and a blood pressure prediction model; the symmetrical dual-channel PPG acquisition device comprises a light-emitting diode and a first photodetector and a second photodetector symmetrically distributed on both sides of the light-emitting diode; the symmetrical dual-channel PPG acquisition device is used to acquire a dual-channel pulse signal; the input end of the feature extraction module is connected to the output end of the symmetrical dual-channel PPG acquisition device; the feature extraction module is used to extract morphological features and calculation features from the dual-channel pulse signal; the input end of the blood pressure prediction model is connected to the output end of the feature extraction module; the blood pressure prediction model is used to predict diastolic pressure and systolic pressure based on the morphological features and calculation features.

[0023] The cuffless blood pressure monitoring system provided by this application includes a symmetrical dual-channel PPG acquisition device with a high degree of integration, and proposes to use only PPG signals with strong anti-interference capabilities, so that it can be worn daily like a smart watch. At the same time, this application adds a lightweight module SENnet to the blood pressure prediction model, so that the neural network has the ability to focus on the input subset. The cuffless blood pressure monitoring system designed by this application is only 45.6 MB, and the operation process only takes up less storage resources and computing resources, and is completely portable to mobile terminals. Compared with existing cuffless blood pressure research, the cuffless blood pressure monitoring system provided by this application has greater potential for application in daily blood pressure monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] One or more embodiments are exemplarily described by the pictures in the corresponding drawings. These exemplifications do not constitute limitations on the embodiments. Unless otherwise stated, the pictures in the drawings do not constitute proportional limitations.

[0025] FIG1 is a schematic structural diagram of a cuffless blood pressure monitoring system provided in one embodiment of the present application;

[0026] FIG2 is a schematic diagram of a symmetrical dual-channel PPG acquisition principle according to an embodiment of the present application;

[0027] FIG3 is a schematic structural diagram of a cuffless blood pressure monitoring system provided by another embodiment of the present application;

[0028] FIG4 is a schematic diagram of a collection process framework of a cuffless blood pressure monitoring system according to an embodiment of the present application;

[0029] FIG5 is a schematic diagram of the operation flow of a cuffless blood pressure monitoring system provided in one embodiment of the present application. Best Mode for Carrying Out the Invention

[0030] As can be seen from the background art, existing invasive monitoring methods are not suitable for daily blood pressure management in patients undergoing intracranial aneurysm surgery. Non-invasive methods such as cuff-type blood pressure monitors, traditional auscultatory blood pressure measurement methods, and electronic blood pressure monitors are not suitable for continuous monitoring or daily blood pressure management in patients undergoing intracranial aneurysm surgery.

[0031] Existing studies suggest that hypertension is a risk factor that may lead to rupture of intracranial aneurysms, especially for postoperative patients, who need to strictly control blood pressure fluctuations to avoid re-rupture. This is because hypertensive patients often have conditions such as myocardial hypertrophy and left ventricular enlargement, which increase the burden on the heart, and the increased myocardial contraction force leads to increased impact force of blood flow on the vessel wall, further increasing the risk of re-rupture in patients. In addition, hypertension may also lead to microcirculation disorders, platelet aggregation and other conditions, which may also lead to worsening of the condition after surgery. The impact of blood flow on the blood vessel wall is the root cause of the formation, growth and rupture of intracranial aneurysms, and blood pressure, as a manifestation of the patient's overall blood flow status, is of great significance for understanding the effect of intervention and avoiding postoperative complications.

[0032] Studying mathematical modeling methods for blood pressure using electrocardiogram (ECG) and photoplethysmography (PPG) signals was an early research focus in cuffless blood pressure measurement. Among related studies, one method utilizes ECG-assisted PPG signals to continuously predict SBP and DBP. The proposed algorithm is not limited to simultaneous ECG and PPG measurements and is more robust than existing technologies. Another related technique proposes a new hybrid prediction model for continuous blood pressure estimation by using ECG-PPG signals as inputs to a recurrent neural network (RNN). The proposed method consists of two steps. Another related technique uses an artificial neural network to implement a user-specific nonlinear autoregressive BP prediction model (NARX) with exogenous inputs to estimate BP waveforms from ECG and PPG signals, establishing a cuffless blood pressure prediction method suitable for daily activities. Another related technique proposes a cuffless blood pressure monitoring system based on pulse transit time, which uses ECG and PPG signals to predict SBP in real time while a subject is seated. Domestic research has also explored methods for using raw PPG and ECG signals as inputs to deep learning models to predict blood pressure even when the raw signals are noisy or distorted. Hybrid deep learning models have also been constructed using body characteristics (age, height, weight, and gender). Related technologies have also developed machine learning-based SBP and DBP prediction models using ECG and PPG signals based on transmission speed and pulse waveform characteristics. During model construction, the average influence method was introduced to study the impact of each feature on the model, and a genetic algorithm was used to optimize parameters.

[0033] Existing invasive and non-invasive methods are unsuitable for routine blood pressure management in patients undergoing intracranial aneurysm surgery. With advances in sensors and algorithms in the medical industry, the demand for continuous cuffless BP monitoring technology continues to grow. Leveraging the pulse wave transmission velocity embedded in physiological signals provides an effective approach for cuffless BP monitoring.

[0034] To solve the above technical problems, an embodiment of the present application provides a cuffless blood pressure monitoring system, comprising: a symmetrical dual-channel PPG acquisition device, a feature extraction module, and a blood pressure prediction model; the symmetrical dual-channel PPG acquisition device includes a light-emitting diode and a first photodetector and a second photodetector symmetrically distributed on both sides of the light-emitting diode; the symmetrical dual-channel PPG acquisition device is used to obtain a dual-channel pulse signal; the input end of the feature extraction module is connected to the output end of the symmetrical dual-channel PPG acquisition device; the feature extraction module is used to extract morphological features and calculation features from the dual-channel pulse signal; the input end of the blood pressure prediction model is connected to the output end of the feature extraction module; the blood pressure prediction model is used to predict diastolic and systolic blood pressure based on morphological features and calculation features. By designing a dual-channel photoplethysmography acquisition device, an embodiment of the present application provides a cuffless blood pressure monitoring system with the advantages of small size, low cost, high sensitivity, etc., realizing a non-invasive, continuous, wearable cuffless blood pressure monitoring device.

[0035] The following detailed description of the various embodiments of the present application is provided in conjunction with the accompanying drawings. However, those skilled in the art will appreciate that many technical details are provided in the various embodiments of the present application to facilitate a better understanding of the present application. However, even without these technical details and the various variations and modifications based on the following embodiments, the technical solutions claimed in the present application can still be implemented.

[0036] 1 , an embodiment of the present application provides a cuffless blood pressure monitoring system, comprising: a symmetrical dual-channel PPG acquisition device 101, a feature extraction module 102, and a blood pressure prediction model 103; the symmetrical dual-channel PPG acquisition device 101 comprises a light-emitting diode (LED) and a first photodetector and a second photodetector symmetrically distributed on both sides of the LED; the symmetrical dual-channel PPG acquisition device 101 is configured to acquire a dual-channel pulse signal; an input end of the feature extraction module 102 is connected to an output end of the symmetrical dual-channel PPG acquisition device 101; the feature extraction module 102 is configured to extract morphological features and computational features from the dual-channel pulse signal; an input end of the blood pressure prediction model 103 is connected to an output end of the feature extraction module 102; and the blood pressure prediction model 103 is configured to predict diastolic and systolic pressures based on the morphological and computational features.

[0037] In some embodiments, referring to FIG2 , in a symmetrical dual-channel PPG acquisition device, a light-emitting diode is located at the center of the symmetrical dual-channel PPG acquisition device; a first photodetector and a second photodetector are symmetrically distributed on the left and right sides of the light-emitting diode; and the first photodetector and the second photodetector receive reflected light to form a dual-channel PPG signal.

[0038] The traditional PPG acquisition module mainly includes a light source and a photodetector, which need to be placed in different positions when measuring the pulse. It is inconvenient to wear daily, and the transmission distance is defined as a constant, which will introduce errors when calculating the pulse. The present invention proposes a cuffless blood pressure monitoring method based on a symmetrical dual-channel PPG acquisition device. As shown in Figure 2, the acquisition device designed in this application consists of a light emitting diode and two photodetectors (a first photodetector and a second photodetector). The light source is placed in the center, and the two photodetectors are symmetrically distributed on both sides. The light intensity is After the incident light passes through the skin, part of the light is absorbed by bones, skin, muscle tissue, etc., part of the light passes through the tissue into the body, and part of the light is absorbed by oxygenated hemoglobin ( ) is reflected in all directions. The symmetrically placed photodetectors receive light with an intensity of The reflected light forms a dual-channel PPG signal.

[0039] In some embodiments, the symmetrical dual-channel PPG acquisition device is located at the radial artery on the outside of the wrist joint to collect radial artery blood pressure.

[0040] Figure 4 shows the flowchart of the cuffless blood pressure monitoring system proposed in this application for daily postoperative management of intracranial aneurysms. After PPG signal acquisition, feature extraction and combination are performed, and then blood pressure is predicted. The radial artery, located just outside the wrist joint, is selected as the acquisition location. Besides being convenient to wear, the shallow position of the radial artery ensures high-precision PPG signals. Furthermore, the radial artery is constrained by the brachioradialis and flexor carpi radialis muscles to maintain a relatively straight position. The nearly linear flow of blood within the radial artery improves pulse calculation accuracy, making it an ideal acquisition location for this system.

[0041] In the symmetrical structure, the photodiode is a red light source with a wavelength of 660 nm. Because the two main components of arterial blood, oxyhemoglobin and reduced hemoglobin, have relatively stable absorption rates around 660 nm, the OPT101 photodetector was chosen. It is widely used in medical testing due to its excellent linearity in the visible light range. Two detectors are symmetrically located on either side of the light source, with a spacing of 0.5 cm and a sampling rate of 250 Hz, enabling efficient signal acquisition with high integration. The data processing unit uses the JFC101 chip, which features a multi-channel, high-precision analog-to-digital converter (ADC). After filtering, the raw signal is transmitted to a host computer program via a universal asynchronous receiver / transmitter (UART). The host computer program displays, saves, and extracts features. Finally, a CNN-LSTM blood pressure prediction model based on the attention mechanism was designed to explore wearable, cuffless blood pressure measurement methods for daily use.

[0042] In some embodiments, referring to FIG. 3 , the cuffless blood pressure monitoring system further includes: a preprocessing module 104 disposed between the symmetrical dual-channel PPG acquisition device 101 and the feature extraction module 102 ; the preprocessing module 104 is configured to preprocess the dual-channel pulse signal acquired by the symmetrical dual-channel PPG acquisition device 101 .

[0043] The collected raw signals are susceptible to interference from factors such as human motion and the external environment. These include high-frequency noise from power transmission lines, electromyographic signals, and electrical components, as well as baseline drift caused by the subject's breathing. Therefore, the raw signals must be preprocessed to obtain high-resolution PPG signals and improve the accuracy of subsequent blood pressure prediction. Common PPG signal preprocessing methods include finite impulse response (FIR) filters, wavelet transforms, impulse response (IIR) filters, and Butterworth filters. While FIR filters offer high stability, they also suffer from significant latency. Wavelet transforms also suffer from significant redundancy. While IIR filters are simple to use, their performance is unstable, especially when processing nonlinear phases. Therefore, neither is suitable for routine blood pressure monitoring systems after intracranial aneurysm surgery, which require strict timeliness. The Butterworth filter, on the other hand, offers three advantages: first, maximum flatness. Its maximum passband flattening effect can flatten the gain of all channels in the same channel; second, monotonically decreasing amplitude-frequency and phase-frequency characteristics; and third, the amplitude-frequency response can be adjusted to approach the ideal state by adjusting the filter order.

[0044] In some embodiments, the preprocessing module includes a Butterworth filter and a sliding average filter; the preprocessing module uses the Butterworth filter to eliminate interference from the electromyographic signal in the dual-channel PPG signal, and uses the sliding average filter to eliminate the influence of baseline drift in the dual-channel PPG signal; and the amplitude-frequency response is brought close to the ideal state by adjusting the filter order; The amplitude and frequency of the Butterworth filter of order is expressed by formula (3):

[0045] (3)

[0046] The function transfer form is obtained by formula (3), as shown in formula (4):

[0047] (4)

[0048] Calculate all coefficients b (b0 to b1) in formula (4) according to the required cutoff frequency and filter order. N-1 ,) to complete the design of the Butterworth filter. The present invention uses the Butterworth filter to eliminate interference such as electromyographic signals in the dual-channel PPG signal.

[0049] To address baseline drift in PPG signals, this application uses a sliding average filter. A sliding average filter is a commonly used digital filter that effectively suppresses noise and smooths variations in the input signal, thereby improving the signal-to-noise ratio. It operates by taking a weighted average of each input signal sample with a certain number of preceding samples to produce each sample of the output signal. The weighting coefficient of the sliding average filter is fixed and can be adjusted as needed to meet different application requirements.

[0050] After the symmetrical dual-channel PPG acquisition device 101 acquires the dual-channel pulse signal, the feature extraction module 102 extracts morphological and computational features from the dual-channel pulse signal. During the feature extraction process, the present application selects 100 synchronized and continuous heartbeat cycles at the ends of both channels, which are the signals most similar to those used for blood pressure measurement. The nine eigenvalues ​​of a single heartbeat cycle are then calculated for each channel. Compared to methods that input the complete waveform, this feature extraction significantly reduces computational complexity, making it promising for porting to wearable devices with limited processing power.

[0051] For the feature extraction task, this application extracts morphological and computational features of dual-channel PPG signals for blood pressure prediction. The following are the features from a certain number of synchronous and continuous heartbeat cycles:

[0052] (1) Heart rate (HR): the number of heart beats per minute.

[0053] (2) Root mean square (RMSSD): The root mean square of the difference between adjacent normal cardiac cycles.

[0054] (3) Very low-frequency (VLF): frequency range is 0.0033-0.04 Hz.

[0055] (4) Low frequency (LF): The power frequency range is 0.04-0.15 Hz.

[0056] (5) High frequency component (HF): high frequency signal with a bandwidth of 0.15~0.40hz.

[0057] (6) Pulse wave amplitude (Peak): The difference between the peak and the trough.

[0058] (7) Pressure constant (PK): The ratio of the maximum amplitude to the average amplitude over a period of time.

[0059] (5)

[0060] Where, and They represent the starting points of the PPG signals of the previous heart cycle and the next heart cycle respectively. express The PPG value at that moment.

[0061] (1) The PPG area (PA): The integrated area within a cardiac cycle, which reflects the changes in total peripheral resistance and vascular tension.

[0062] (2) Pulsatile heterogeneity height (PHH): Indicates the size of cardiac output and is calculated as follows:

[0063] (6)

[0064] in, is the rise time of the PPG signal, is the fall time of the PPG signal.

[0065] The first five are heart rate variability (HRV) features, and the last four are morphological features associated with a single cardiac cycle. Furthermore, six PTT features were obtained by calculating the temporal distance between the two acquired PTTA, PTTa, PPTB, PTTb, PTTC, and PTTc signals. Thus, this application collected 24 characteristic PPG signals from the symmetrical signals, including nine morphological and computational features extracted from each of the two sets of PPG signals, as well as six PTT features from the two symmetrical PPG signals.

[0066] The cuffless blood pressure monitoring system provided by this application is a dual-channel pulse wave measurement system. The measurement system is a host computer developed on the Qt Creator platform based on the Windows operating system. The software is used in conjunction with a symmetrically structured PPG signal acquisition module to measure the pulse wave at the radial artery. The system flow chart is shown in Figure 5. This application implements communication between modules based on QSerialPort. For example, to send an open command, the write() method is used to send the command from the computer to the specified serial port through URAT. After the software is turned on, the command 0x8A is sent again, and the measurement module will continue to send data packets containing pulse wave information. The readAll() method in the timer response signal onTimeOut() is used to read the data, and the toHex() method is used to convert the data into hexadecimal format and store it locally. This application also designs an analytical function to extract eigenvalues ​​for subsequent blood pressure prediction.

[0067] This software requires minimal storage resources, with the software itself occupying only 45.6MB. For pulse wave measurement, a typical computer can meet the storage requirements.

[0068] When predicting blood pressure using a blood pressure prediction model, pulse wave data is first collected. Based on the proposed dual-channel PPG measurement module, this application collected 260 sets of PPG signals. This application used the measurement values ​​of a Yuwell YE660F electronic blood pressure monitor as target values ​​to verify the performance of the prediction model. Considering that the cuff of an electronic blood pressure monitor can compress blood vessels, resulting in the inability to obtain valid PPG signals during and after the measurement, this application first collected dual-channel PPG signals and then used the electronic blood pressure monitor for measurement. When collecting data using the acquisition device designed in this invention, the acquisition device was worn at the left wrist joint to ensure that the radial artery pulse signal was collected. Each volunteer collected pulse waves for at least 5 minutes to ensure sufficient valid signals. Throughout the process, the volunteer's forearm was kept at heart height and all physical activity was minimized. The volunteers were asked to remove any thick clothing from their arms and wear the cuff on their left upper arm, with the cuff nozzle facing the lower arm and located on the inner side of the arm to ensure smooth pressure delivery. The lower edge of the cuff was also required to be 2-3 cm away from the elbow joint. Press the start button to start the measurement, and try to avoid any physical activity of the volunteers during the measurement. In order to avoid large fluctuations in blood pressure and reduce the gap between the two measurement methods, this application requires volunteers to sit quietly for 15 minutes before collection to stabilize their heart rate. In this stage, this application recruited 115 volunteers. In order to verify the effectiveness of this method for daily blood pressure fluctuations, this application collected data in multiple time periods such as 7 to 9 in the morning, 2 to 4 in the afternoon, and 6 to 8 in the afternoon. Finally, 260 valid data were collected for subsequent research, of which 80% were used as training sets and 20% as validation sets. A 10-fold cross-validation method was used to improve model performance. Each cross-validation randomly selected 10% of the samples from the training data for verification, and the remaining data was used for training. This chapter will also compare the proposed method with traditional machine learning methods and deep learning models in other studies to verify its robustness.

[0069] In some embodiments, the blood pressure prediction model is a CNN-LSTM prediction model based on the attention mechanism.

[0070] To improve BP prediction accuracy, we designed a CNN-LSTM-based prediction model. We extracted 100 consecutive pulse cycles from the symmetrical dual-channel PPG signal and extracted 24 eigenvalues ​​from each pulse cycle. This 24x100 feature vector was then fed into the model to extract hidden features for accurate DBP and SBP prediction.

[0071] Considering that the attention mechanism of the sequence excitation network can focus the model's vision on features that are more relevant to the results, this application adds a compression and excitation network SENet (Squeeze-and-Excitation) module to the constructed model. The convolution process of SENet is used as an example to introduce the optimization strategy of SENet. Indicates the convolution kernel, then for the output The formula is as follows:

[0072] (7)

[0073] Where * represents the convolution operation, and express The convolution kernel of a channel inputs the spatial features on a channel. The model can learn the spatial relationship of features, but because each channel is Operations, all feature relationships and the spatial relationships learned by the convolution kernel are mixed. SENet is designed to resolve this confusion, allowing the model to learn stronger feature relationships. SENet units include compression units Squeeze and excitation units Excitation, which can be applied to any mapping:

[0074]

[0075] (8)

[0076]

[0077] (9)

[0078]

[0079] (10)

[0080] Among them, Squeeze and Excitation represent compression unit and excitation unit respectively; .

[0081] In order to reduce the complexity of the model and improve the generalization ability, SENet adopts a bottleneck structure consisting of two fully connected layers, where the first fully connected layer plays the role of dimensionality reduction, and the dimensionality reduction coefficient is hyperparameters, then ReLU activation is used, and the original dimension is restored in the final fully connected layer.

[0082] In some embodiments, the blood pressure prediction model includes a first convolution module, a second convolution module, a third convolution module and two LSTM modules; the first convolution module and the second convolution module each include a convolution layer and a SENet unit; the third convolution module includes a convolution layer, and a normalization layer and an activation function are set after each convolution layer.

[0083] Specifically, the BP prediction model proposed in this invention consists of three convolutional modules and two LSTM modules. The first two convolutional modules consist of a convolutional layer and an SENet, while the last convolutional module is a single convolutional layer. Each convolutional layer is followed by a batch normalization (BN) layer and a rectified linear unit (ReLU). ReLU, also known as a rectified linear unit, is a commonly used activation function in artificial neural networks. A max pooling layer (Maxpool) is placed after the convolutional modules and before the LSTM modules. The LSTM modules consist of LSTM units and dropout. The LSTM modules are primarily used to eliminate the effects of nonlinear physiological changes on blood pressure prediction. The model finally uses a fully connected layer to return the predicted value. In this network, the number of CNN channels is 12, 24, and 1, respectively. The core size of the convolutional layer is set to 3×3, and the core size of the max pooling layer is set to 2×2. The number of neurons in the LSTM modules is 50 and 8, respectively, and the dropout learning rate is set to 0.5.

[0084] In summary, this application proposes a new cuffless blood pressure detection system for daily management of blood pressure after intracranial aneurysm surgery. It mainly consists of two parts. The first part is a dual-channel PPG signal acquisition module based on a symmetrical structure, which solves the problem that traditional methods are difficult to accurately define the pulse transmission distance, while having the advantages of low cost, high integration and wearability. The second part is a blood pressure prediction model. This application proposes a CNN-LSTM model based on an attention mechanism for predicting diastolic blood pressure DBP and systolic blood pressure SBP. Through the experiments of this application, the proposed method has good performance and reaches level A in the AAMI standard and the BHS standard. Industrial Applicability

[0085] Compared with the existing technology, the advantages of this application are: this application explores a lightweight blood pressure monitoring system so that it can meet the daily use environment. Starting from the design and development of the acquisition device, this application abandoned the idea of ​​fusing ECG signals and proposed to use only PPG signals with strong anti-interference capabilities to improve wearability. Unlike existing research, this application proposes a symmetrical dual-channel PPG acquisition device with high integration, so that it can be worn daily like a smart watch. For the research on blood pressure prediction models, this application also adheres to the high applicability model and adds a lightweight module SENnet to the blood pressure prediction model, so that the neural network has the ability to focus on the input subset. The designed blood pressure monitoring system is only 45.6 MB, and the running process only takes up less storage resources and computing resources, and is completely portable to mobile terminals. In general, compared with existing cuffless blood pressure research, the method of this application has more potential for application in daily blood pressure monitoring.

[0086] In order to verify the performance of BP prediction proposed in this application, this application analyzes the prediction performance of other machine learning and deep learning methods on the data set of this application. This application selects four traditional machine learning methods and four deep learning methods, including linear regression (LR), decision tree (DT), random forest (RF), Adaboost, CNN-LSTM, CNN and other methods. In RF, this application sets the estimator search range to 3~14 and 100~1000. For the AdaBoost method, this application uses a grid search method to train the base classifier in the range of 100-1000.

[0087] The mean absolute error (MAE) and standard deviation (STD) were used as evaluation metrics. For DBP and SBP, the proposed method achieved MAE and STD of 1.65±1.91 mmHg and 2.16±2.39 mmHg, respectively, outperforming other methods overall. Furthermore, it met the Association for the Advancement of Medical Instrumentation (AAMI) standard for sphygmomanometers (MAE <5 mmHg). Furthermore, the British Hypertension Society (BHS) categorizes blood pressure plans into four levels: AD. The blood pressure predicted by the cuffless blood pressure monitoring system proposed in this application achieved Class A performance for both DBP and SBP.

[0088] Based on the above technical solution, an embodiment of the present application provides a cuffless blood pressure monitoring system, including: a symmetrical dual-channel PPG acquisition device, a feature extraction module and a blood pressure prediction model; the symmetrical dual-channel PPG acquisition device includes a light-emitting diode and a first photodetector and a second photodetector symmetrically distributed on both sides of the light-emitting diode; the symmetrical dual-channel PPG acquisition device is used to obtain a dual-channel pulse signal; the input end of the feature extraction module is connected to the output end of the symmetrical dual-channel PPG acquisition device; the feature extraction module is used to extract morphological features and calculation features from the dual-channel pulse signal; the input end of the blood pressure prediction model is connected to the output end of the feature extraction module; the blood pressure prediction model is used to predict diastolic pressure and systolic pressure based on the morphological features and calculation features.

[0089] In the blood pressure prediction method proposed for the daily management of blood pressure in postoperative patients, the dual-channel photoplethysmography acquisition device designed in this application has the advantages of small size, low cost, and high sensitivity. It realizes a non-invasive, continuous, wearable cuffless blood pressure monitoring device, and uses a CNN-LSTM model based on the attention mechanism to predict diastolic and systolic blood pressure. The results show that the method proposed in this invention has good prediction performance.

[0090] The cuffless blood pressure monitoring system provided by this application includes a symmetrical dual-channel PPG acquisition device with a high degree of integration, and proposes to use only PPG signals with strong anti-interference capabilities, so that it can be worn daily like a smart watch. At the same time, this application adds a lightweight module SENnet to the blood pressure prediction model, so that the neural network has the ability to focus on the input subset. The cuffless blood pressure monitoring system designed by this application is only 45.6 MB, and the operation process only takes up less storage resources and computing resources, and is completely portable to mobile terminals. Compared with existing cuffless blood pressure research, the cuffless blood pressure monitoring system provided by this application has greater potential for application in daily blood pressure monitoring.

[0091] That is, those skilled in the art will understand that all or part of the steps in the above-mentioned method embodiments can be implemented by instructing the relevant hardware through a program. The program is stored in a storage medium and includes a number of instructions for causing a device (which may be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps in the above-mentioned method embodiments of the present application. The aforementioned storage medium includes: a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., various media that can store program code.

[0092] Those skilled in the art will appreciate that the above-described embodiments are specific examples for implementing the present application, and that in actual applications, various changes in form and detail may be made thereto without departing from the spirit and scope of the present application. Any person skilled in the art may make changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be subject to the scope defined in the claims.

Claims

1. A cuffless blood pressure monitoring system, characterized in that: include: Symmetrical dual-channel PPG acquisition device; The symmetrical dual-channel PPG acquisition device comprises a light emitting diode and a first photodetector and a second photodetector symmetrically distributed on both sides of the light emitting diode; the symmetrical dual-channel PPG acquisition device is used to acquire a dual-channel pulse signal; Feature extraction module; The input end of the feature extraction module is connected to the output end of the symmetrical dual-channel PPG acquisition device; the feature extraction module is used to extract morphological features and computational features from the dual-channel pulse signal; A blood pressure prediction model; the input end of the blood pressure prediction model is connected to the output end of the feature extraction module; the blood pressure prediction model is used to predict the diastolic pressure and the systolic pressure based on the morphological features and the calculation features.

2. The cuffless blood pressure monitoring system according to claim 1, characterized in that: In the symmetrical dual-channel PPG acquisition device, the light emitting diode is located at the center of the symmetrical dual-channel PPG acquisition device; the first photodetector and the second photodetector are symmetrically distributed on the left and right sides of the light emitting diode; The first photodetector and the second photodetector receive reflected light to form a dual-channel PPG signal.

3. The cuffless blood pressure monitoring system according to claim 1, characterized in that: The blood pressure prediction model is a CNN-LSTM prediction model based on the attention mechanism.

4. The cuffless blood pressure monitoring system according to claim 3, characterized in that: The blood pressure prediction model includes a first convolution module, a second convolution module, a third convolution module and two LSTM modules; The first convolution module and the second convolution module each include a convolution layer and a SENet unit; the third convolution module includes a convolution layer, and a normalization layer and an activation function are set after each convolution layer.

5. The cuffless blood pressure monitoring system according to claim 4, characterized in that: The SENet unit includes a compression unit and an excitation unit, and the compression unit and the excitation unit are respectively expressed as follows: （1） （2） Among them, Squeeze and Excitation represent compression unit and excitation unit respectively; 。 6. The cuffless blood pressure monitoring system according to claim 4, characterized in that: A maximum pooling layer is provided between the third convolution module and the two LSTM modules; The LSTM module is used to eliminate the influence of nonlinear physiological characteristic changes on blood pressure prediction; The LSTM module includes an LSTM unit and a Dropout unit; The blood pressure prediction model finally uses a fully connected layer to return the predicted value.

7. The cuffless blood pressure monitoring system according to claim 1, characterized in that: The symmetrical dual-channel PPG acquisition device is located at the radial artery outside the wrist joint to collect radial artery blood pressure.

8. The cuffless blood pressure monitoring system according to claim 1, characterized in that: Also includes: A preprocessing module provided between the symmetrical dual-channel PPG acquisition device and the feature extraction module; The preprocessing module is used to preprocess the dual-channel pulse signal collected by the symmetrical dual-channel PPG acquisition device.

9. The cuffless blood pressure monitoring system according to claim 8, characterized in that: The preprocessing module includes a Butterworth filter and a sliding average filter; the preprocessing module uses the Butterworth filter to eliminate the interference of the electromyographic signal in the dual-channel PPG signal, and uses the sliding average filter to eliminate the influence of the baseline drift in the dual-channel PPG signal; The amplitude-frequency response is brought close to the ideal state by adjusting the filter order; The amplitude and frequency of the Butterworth filter of order are expressed by formula (3): （3） The function transfer form is obtained by formula (3), as shown in formula (4): （4） Calculate the coefficients b0 to b in formula (4) according to the required cutoff frequency and filter order. N-1 , completing the design of the Butterworth filter.

Images