PPG blood pressure monitoring method and system

Abstract

The invention discloses a PPG blood pressure monitoring method and system, and the method comprises the steps: collecting a PPG signal of a user through a narrow-spectrum light source and a narrow-band response organic light detector, and carrying out the sliding window segmentation of the PPG signal of the user; preprocessing the expanded data set to obtain 20-dimensional features of each window; a plurality of machine learning algorithms are utilized, and 20-dimensional features are adopted to train the initial blood pressure prediction model; optimizing the 20-dimensional features to obtain an optimal feature subset combination of each model; retraining the initial blood pressure prediction model by using the optimal feature subset combination, and comparing the model performance with the training result obtained in the previous step to finally obtain a personalized blood pressure prediction model; according to the method, the narrow-spectrum red light source and the narrow-band organic photoelectric detector are cooperatively utilized, an innovative intelligent waveform processing algorithm is combined, ambient light interference is effectively reduced, PPG signal quality is optimized, and all-weather high-precision blood pressure monitoring is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of physiological parameter monitoring, in particular to a PPG blood pressure monitoring method and system. BACKGROUND

[0002] With the aging of the population and the increase of chronic diseases, blood pressure monitoring as an important part of health management has attracted more and more attention. Traditional blood pressure measurement methods mainly rely on cuff blood pressure meters, which are semi-invasive monitoring and can only provide single-point blood pressure data, cannot realize continuous monitoring, and the frequent inflation process will bring discomfort to the user. In recent years, non-invasive blood pressure monitoring technology based on photoplethysmography (PPG) has been favored due to its portability and continuous monitoring capability.

[0003] PPG technology is a method of detecting blood vessel volume changes through a photoelectric sensor, which can be used to monitor physiological parameters such as heart rate, blood oxygen saturation, and blood pressure. At present, there are many blood pressure monitoring devices based on PPG technology on the market, such as a blood pressure measurement system based on flexible organic light detector and deep learning algorithm disclosed in Chinese Patent No. CN116548935A, which realizes continuous monitoring of blood pressure values by wearing a flexible organic light detector and an ECG electrode on the wrist.

[0004] The existing PPG blood pressure monitoring technology still has the following problems:

[0005] Firstly, PPG signals are low-frequency signals, which are easily disturbed by environmental light noise, power frequency noise, and circuit disturbances of the device during acquisition. The light source used in the existing technology is mostly a wide-spectrum light source, which has a wide emission spectrum and is easily mixed with environmental light, resulting in a decrease in signal quality. At the same time, conventional photoelectric detectors respond to light of all wavebands, and cannot effectively filter out interfering light signals of non-characteristic wavebands.

[0006] Secondly, low-quality PPG signals will affect the training effect of the algorithm, resulting in a significant reduction in accuracy. The blood pressure prediction model in the existing technology is mostly based on traditional machine learning algorithms or more complex neural network architectures, which require high signal quality. In order to achieve the best prediction effect, a lot of time and effort is usually spent on fine-tuning the model architecture and parameters, and it is usually impossible to achieve a balance between local computing resources and model performance. The complex model with performance as the premise will encounter a lot of difficulties in the process of local deployment in actual application scenarios, and the prediction accuracy will be significantly reduced when the input signal is disturbed.

[0007] Finally, the prior art cannot achieve high-precision cuffless blood pressure monitoring. Due to the complex and nonlinear relationship between PPG signals and blood pressure, existing PPG-based blood pressure monitoring methods still have a lot of room for improvement in accuracy and stability, especially in continuous monitoring scenarios under different environmental light conditions.

[0008] Therefore, there is an urgent need to develop a technical solution that can obtain high-quality PPG signals and achieve accurate blood pressure monitoring under various environmental conditions. SUMMARY

[0009] In order to overcome the above-mentioned shortcomings and deficiencies of the prior art, the purpose of the present application is to provide a PPG blood pressure monitoring method and system.

[0010] The purpose of the present application is achieved by the following technical solutions:

[0011] A PPG blood pressure monitoring method, comprising:

[0012] A narrow-spectrum light source and a narrow-band response organic light detector are used to collect PPG signals of a user, forming a data set;

[0013] The user's PPG signals are divided into sliding windows to expand the data set;

[0014] The expanded data set is preprocessed, specifically using an intelligent waveform segmentation and feature extraction algorithm to extract 20-dimensional waveform features for each window, discard incomplete waveforms, and then extract the mean value of the complete waveform feature data in each window to obtain 20-dimensional features for each window;

[0015] A variety of machine learning algorithms are used to train the initial blood pressure prediction model using 20-dimensional features;

[0016] Based on the recursive feature elimination cross-validation algorithm, the 20-dimensional features are optimized to obtain the optimal feature subset combination for each model;

[0017] The initial blood pressure prediction model is retrained using the optimal feature subset combination, and the training results obtained in the previous step are compared in terms of model performance, and finally a personalized blood pressure prediction model is obtained;

[0018] The real-time collected PPG signals are extracted using the intelligent waveform segmentation and feature extraction algorithm, and a feature subset combination is constructed according to the requirements of the personalized blood pressure prediction model, and finally input into the personalized blood pressure prediction model to obtain the blood pressure prediction value.

[0019] Further, the narrow-spectrum light source is a human-safe power narrow-spectrum red light source with an emission peak of 627 nm and a half-peak width of 3.2 nm, prepared by Mn4+ doped fluorogallate single crystal red light material.

[0020] Further, the narrow-band response photodetector is prepared by introducing an optical depletion layer with a specific absorption spectrum, in combination with an organic photoactive layer, to obtain a narrow-band response organic photodetector with a response peak at 655 nm, wherein the process conditions of the optical depletion layer and the organic photoactive layer are as follows:

[0021] Optical depletion layer: P3HT:J71:IEICO-4F (1:1:1) dissolved in chloroform (CF), film thickness about 400-500 nm after spin coating, thermal annealing at 100°C for 10 min;

[0022] Organic photoactive layer: PM6:L8-BO (1:1.2) dissolved in chloroform (CF), film thickness about 150-200 nm after spin coating, thermal annealing at 100°C for 10 min.

[0023] Further, the PPG signal of the user is divided by a sliding window to expand the data set, specifically:

[0024] A 875-point sliding window is used to divide the 5000-point PPG signal collected each time;

[0025] Each window inherits the SBP / DBP labels of the original file;

[0026] Effective windows are automatically selected by waveform quality evaluation;

[0027] The data set is expanded by 5-10 times from the file level to the window level.

[0028] Further, the intelligent waveform segmentation and feature extraction algorithm includes:

[0029] Intelligent segmentation of PPG waveform based on trough detection;

[0030] Identify the peak of the tidal wave by first derivative analysis;

[0031] Automatically detect and discard incomplete waveforms;

[0032] Extract 20-dimensional waveform features;

[0033] Calculate the mean value of the complete waveform features to generate 20-dimensional window features.

[0034] Further, the PPG signal includes the main wave, the tidal wave, the descending middle isthmus, and the double beat wave.

[0035] Further, the 20-dimensional features of each window include peak features, slope features, ratio features, time features, and time ratio features.

[0036] Further, the plurality of machine learning algorithms includes SVR, GBR, RF, LR, and XGB.

[0037] Further, the 20-dimensional waveform features include:

[0038] Peak features: systolic peak, tidal peak, diastolic peak, bottom value;

[0039] Slope features: rising / falling slope between peaks;

[0040] Ratio features: relative relationship between peaks and bottom;

[0041] Time features: pulse width, diastolic time interval;

[0042] Time ratio features: relative relationship of time intervals.

[0043] A system for implementing the PPG blood pressure monitoring method, comprising:

[0044] A signal acquisition module adopts a fully integrated analog front-end module, connects a narrow-spectrum light source and a narrow-band response organic photodetector, and acquires a PPG signal of a user;

[0045] A PPG signal processing module is used for controlling and driving the narrow-spectrum light source and the narrow-band response organic photodetector to read signals and features in a time sequence, and transmitting the acquired PPG signal and blood oxygen data to a computer terminal host computer through a UART protocol;

[0046] The host computer inputs a personalized blood pressure prediction model after preprocessing and feature extraction of the PPG signal, and obtains a blood pressure value.

[0047] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0048] (1) The present application adopts a narrow-spectrum red light source and a narrow-spectrum red light detector to acquire a PPG signal. Compared with a commercial blood oxygen probe, the PPG signal acquired by the strategy has more significant details. After pulse waveform morphological feature extraction and analysis using an intelligent waveform segmentation and feature extraction algorithm, higher blood pressure monitoring accuracy can be achieved.

[0049] (2) The narrow-band organic photodetector described in the present application can adjust the response band and responsivity of the organic photodetector by flexible control of the thickness and material of the optical depletion layer. The device is prepared using a solution processing and evaporation process, and does not have the problem of lattice mismatch in processes such as lattice growth and vapor deposition. It has the feasibility and simplicity of operation, and is expected to be applied in the fields of flexible wearable devices, mobile sensing and medical health monitoring.

[0050] (3) The narrow-band response organic photodetector described in the application has strong response to only single range of light, can filter ambient light in an ambient light interference environment, can still collect high-quality PPG signals in a PPG signal collection process, and realizes all-weather blood pressure monitoring.

[0051] (4) The complete algorithm flow described in the application has clear logic and innovation, first solves the problem of poor model generalization ability caused by insufficient user personalized sample quantity in a traditional method through a small sample data set expansion technology, realizes 5-10 times expansion of a data set, provides sufficient training data basis for a subsequent algorithm, and ensures that an optimal personalized model conforming to a user can be built under limited data sets.

[0052] (5) The intelligent waveform segmentation and feature extraction algorithm described in the application realizes accurate extraction of key physiological features in a PPG signal through adaptive segmentation based on wave trough detection and intelligent identification of tidal wave features, significantly improves feature point positioning accuracy, lays a solid foundation for high-precision blood pressure prediction, perfectly adapts to high-quality PPG signals with more rich physiological health information collected by the system, and solves the problem of inaccurate and incomplete feature extraction in a traditional method.

[0053] (6) The feature subset selection and optimization algorithm described in the application can customize optimal feature combinations for different users based on RFECV multi-model feature selection, realizes personalized modeling, significantly improves the accuracy and stability of blood pressure prediction, eliminates feature redundancy and overfitting problems, and improves the generalization ability of the model.

[0054] (7) The personalized model construction technology described in the application realizes user customized high-precision blood pressure prediction through a feature subset retraining strategy, solves the problem of insufficient model generalization ability and low personalization degree in a traditional method, and can obtain an optimal personalized model conforming to a user under limited data sets.

[0055] (8) The all-weather monitoring technology described in the application can maintain stable monitoring performance under various environmental conditions through cooperative work of a narrow-spectrum light source and a narrow-band detector, combined with an innovative intelligent signal processing algorithm, meets the needs of clinical continuous monitoring, and realizes high-precision cuffless blood pressure monitoring.

[0056] (9) The all-weather PPG blood pressure monitoring system based on a narrow-spectrum light source and a narrow-band response organic photodetector described in the application has intelligent algorithm support for customized feature selection and personalized modeling, the system supports cyclic collection, the model algorithm can be updated in real time, personalized high-precision blood pressure parameter prediction can be realized, and the needs of clinical practical application are met.

[0057] (10) The all-weather PPG blood pressure monitoring system based on a narrow spectrum light source and a narrow band response organic light detector described in the application, through system optimization and program design, the whole system operation is simple. Through the cooperative work of computer-microcontroller-analog acquisition front-end chip, only the finger is put into the probe, the real-time monitoring and feedback of blood pressure can be realized, which has good user experience and clinical practicability. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 is the workflow schematic diagram of the application;

[0059] Figure 2 is the excitation spectrum diagram of the narrow spectrum light source used in the application;

[0060] Figure 3 is the emission spectrum diagram of the narrow spectrum light source used in the application;

[0061] Figure 4 is the external quantum efficiency data diagram of the narrow band response organic photodetector used in the application;

[0062] Figure 5 is the specific detectivity data diagram of the narrow band response organic photodetector used in the application;

[0063] Figure 6 is the high-quality pulse waveform reference diagram of the application;

[0064] Figure 7 is the PPG signal diagram based on the commercial blood oxygen probe collected in the application;

[0065] Figure 8 is the PPG signal diagram based on the narrow spectrum light source and the narrow band organic photodetector collected in the application;

[0066] Figure 9 is the small sample data set expansion effect diagram of the application;

[0067] Figure 10 is the intelligent waveform segmentation and feature extraction algorithm schematic diagram of the application;

[0068] Figure 11 is the model performance optimization diagram with feature subset selection of the application;

[0069] Figure 12 is the model performance comparison diagram before and after individual modeling of the application. DETAILED DESCRIPTION

[0070] The application will be further described in detail below in combination with embodiments, but the embodiments of the application are not limited thereto.

[0071] EMBODIMENT

[0072] AsFigure 1 As shown in the figure, a PPG blood pressure monitoring method comprises the following steps:

[0073] S1: preparing a narrow spectrum light source with human body safe power, concentrated emission spectrum and small half peak width.

[0074] Specifically, a narrow spectrum red light source with human body safe power, an emission peak of 627 nm and a half peak width of 3.2 nm is prepared by using a Mn4+ doped fluorogallate single crystal red light material. The preparation process of the narrow spectrum light source comprises the following steps: first, a fluorine single crystal red light material is prepared by using a saturated solution crystallization method, and the Mn4+ doped fluorogallate single crystal red light material is combined and packaged with a blue InGaN chip to obtain the final narrow spectrum red light source. The luminous intensity of the narrow spectrum light source depends entirely on the commercial blue light LED used for packaging, which is different from laser. The narrow spectrum light source is in the human body safe power range and can be used for long-term human health monitoring. At the same time, the emission peak of the light source is stable at 627 nm and supports long-term stable operation. Compared with existing commercial light sources, the emission spectrum has little thermal drift and small dispersion, and the half peak width is constant at 3.2 nm. The excitation spectrum diagram of the prepared narrow spectrum light source is as shown in the figure. Figure 2 As shown in the figure, the emission spectrum of the narrow spectrum light source is as shown in the figure. Figure 3

[0075] S2: preparing an organic photoelectric detector with a corresponding narrow band response.

[0076] Specifically, a narrow band response organic photoelectric detector with a response peak of 655 nm and a half peak width of 85 nm is prepared, which comprises, in sequence, an optical depletion layer, a transparent base substrate, a conductive anode, a hole transport layer, an organic photoactive layer, an electron transport layer and a conductive cathode.

[0077] The optical depletion layer is selected according to the target response waveband of the device, using a donor material, an acceptor material or blending the donor and acceptor materials. The final target response waveband corresponds to the weak absorption region of the optical depletion layer absorption spectrum. In this embodiment, P3HT:J71:IEICO-4F (1:1:1) is used as the optical depletion layer material, a film with a thickness of 450-500 nm is prepared, and the film is heat annealed at 100℃ for 10 min. The base substrate is glass. The conductive anode is ITO. The hole transport layer is PEDOT:PSS with a film thickness of 5-10 nm. The organic photoactive layer is blended from a donor material PBDB-T-2F (PM6) and an acceptor material L8-BO, a film with a thickness of 150-200 nm is prepared, and the film is heat annealed at 100℃ for 10 min. The electron transport layer is PFN-Br with a film thickness of 5-10 nm. The conductive cathode is aluminum with a film thickness of 80 nm. As shown in the figures Figure 4 and Figure 5 ​The detector has a narrow-band response spectrum with a peak of 655 nm and a half-peak width of 85 nm, low dark current, high specific detectivity, high sensitivity, and other characteristics, and is particularly suitable for precise detection of biomedical signals.

[0078] S3 obtains more physiological and healthy information and higher quality PPG signals by using a narrow-spectrum light source and a narrow-band response organic photodetector, and constitutes a data set. The PPG signal waveform features include a main wave, a tidal wave, a descending middle isthmus, and a heavy beat wave (also respectively referred to as a systolic peak, a tidal wave, a descending middle isthmus, and a diastolic peak). In this embodiment, a user places a finger between the narrow-spectrum light source and the narrow-band response organic photodetector, clicks Start Collection on a computer terminal after loading a blood pressure model on an upper computer interface, and can display real-time PPG waveforms of the user and corresponding blood pressure values.

[0079] Specifically, the narrow-spectrum red light source prepared in step one and the narrow-band response organic photodetector prepared in step two are respectively connected to a programmable integrated LED driver of an analog front-end chip AFE4490 and a low-noise detector receiving channel. After the 627 nm red light emitted by the narrow-spectrum light source penetrates the skin, it interacts with hemoglobin in the blood, part of the light is absorbed, and part of the light is scattered, and finally received by the narrow-band response organic photodetector. Because the concentration of hemoglobin in the blood changes periodically with the heartbeat, the received light signal also changes periodically, forming a high-quality PPG signal with obvious tidal wave characteristics that is obviously different from a commercial blood oxygen probe.

[0080] By synergistically using a narrow-spectrum red light source and a narrow-band organic photodetector with a customized waveband response, the interference of ambient light and motion artifacts can be effectively reduced. Because the emission spectrum of the narrow-spectrum light source is concentrated around 627 nm, the half-peak width is only 3.2 nm, compared with a wide-spectrum light source, the narrow-spectrum red light source can ensure more consistent tissue penetration and reduce wavelength-dependent scattering, thereby minimizing the signal broadening phenomenon caused by tissue heterogeneity response. The response peak of the organic photodetector is at 655 nm, and the half-peak width is only 85 nm, thereby forming a feature collection corresponding to the physical level of “characteristic emission + characteristic absorption”. The system can therefore have a natural inhibition effect on other wavelength components in ambient light. At the same time, the narrow-band response characteristic can also reduce the interference of scattered light on the skin surface and improve the signal-to-noise ratio.

[0081] By comparing the PPG signal waveform obtained by the commercial blood oxygen probe, as shown in Figure 7 and Figure 8 in combination with the high-quality pulse waveform reference Figure 6The results show that the method can significantly improve the PPG signal quality, reduce the non-characteristic wave band noise signal interference, and make the PPG signal show more abundant physiological health information. Specifically, the signal waveform is smoother, the original signal noise is smaller, and the details such as the tidal wave are more obvious. The morphology and amplitude of these characteristics are closely related to the vascular elasticity, peripheral resistance and heart-vessel coupling state.

[0082] S4 performs sliding window segmentation on the collected small sample PPG signal dataset, realizes 5-10 times expansion of the dataset from the file level to the window level, so as to solve the problem of poor model generalization ability caused by insufficient sample quantity in traditional PPG blood pressure monitoring.

[0083] Specifically, the small sample dataset expansion technology solves the key problem of insufficient user personalized sample quantity in traditional PPG blood pressure monitoring, such as Figure 9 As shown in the figure. The 5000-point PPG signal is segmented by using a 875-point sliding window, the window step is set to 250 points, and the dataset is expanded by 5-10 times. Each window inherits the SBP / DBP label of the original file to ensure data consistency. In cooperation with the intelligent waveform segmentation and feature extraction algorithm in S5, the waveform features of each window are extracted, so as to obtain 20-dimensional waveform features of each high-quality PPG signal in the window. After excluding abnormal and incomplete waveforms, the window waveform feature data is averaged to exclude the weight interference of accidental waveform errors. Finally, the expanded dataset is formed. Through this technology, even if there is only a small amount of user personalized data PPG file, enough training samples can be generated, the user's repeated collection times can be reduced, the PPG data collected at a single time is used to the greatest extent, and the problem of small user personalized dataset is effectively solved. This expansion method not only increases the sample quantity, but more importantly, maintains the physiological continuity and label consistency of the data, providing a reliable data basis for subsequent model training.

[0084] Further explanation: the waveform quality of each window is automatically evaluated by the intelligent algorithm, including signal integrity check: ensuring that the window contains complete pulse periods and avoiding the influence of truncated waveforms on feature extraction; noise level evaluation: calculating the signal-to-noise ratio, selecting high-quality signals, and ensuring the purity of training data; baseline stability analysis: detecting baseline drift to ensure signal quality and improve the accuracy of feature extraction.

[0085] S5, the intelligent waveform segmentation and feature extraction algorithm is adopted to extract 20-dimensional waveform features from each window, discard incomplete waveforms, and then extract the mean value of the complete waveform feature data in each window to obtain 20-dimensional features of each window, ensuring the consistency and reliability of the feature quality.

[0086] Specifically, the intelligent waveform segmentation and feature extraction algorithm realizes the accurate extraction of key physiological features in the PPG signal, such as Figure 10 As shown in the figure, the algorithm can automatically identify the key physiological features in the PPG signal and intelligently discard incomplete waveforms to ensure feature quality.

[0087] Further, the intelligent waveform segmentation and feature extraction algorithm has the following specific steps:

[0088] S5.1 Intelligent waveform segmentation based on trough detection;

[0089] The algorithm first performs trough detection on the preprocessed PPG signal, uses the zero-crossing point recognition method of the first derivative of the signal, and realizes accurate segmentation of the waveform. Unlike the traditional fixed-length segmentation method, this algorithm can adaptively adjust the segmentation points according to the physiological features of the PPG signal to ensure that each segmented segment contains a complete pulse cycle. The specific implementation method is as follows:

[0090] (1) Calculate the first derivative of the PPG signal and identify the zero-crossing point of the derivative;

[0091] (2) Find the local minimum value near the zero-crossing point to determine the trough position;

[0092] (3) Segment the waveform according to the trough position to ensure that each segment contains a complete pulse cycle;

[0093] (4) Verify the physiological reasonableness of the segmented segment and discard abnormal segments.

[0094] S5.2 Intelligent identification technology of tidal wave features;

[0095] The tidal wave is a key physiological feature that distinguishes high-quality PPG signals collected by the system from general PPG signals, as shown in Figure 7 and Figure 8 In combination with the high-quality pulse waveform reference Figure 6 , it is closely related to vascular elasticity and peripheral resistance, as shown in Figure 10 The algorithm intelligently identifies the tidal wave peak position by analyzing the negative slope section of the first derivative of the signal. The specific implementation method is as follows:

[0096] (1) Find the negative slope section of the first derivative after the systolic pressure peak;

[0097] (2) Sort by duration and select the three longest negative slope sections;

[0098] (3) Determine the tidal wave peak and diastolic pressure peak through position relationship;

[0099] (4) Automatically verify the physiological reasonableness of the feature points.

[0100] The extracted 20-dimensional waveform features, as shown in Table 1, include:

[0101] Peak features (4 dimensions): systolic peak, tidal peak, diastolic peak, bottom value, reflecting vascular elasticity and peripheral resistance;

[0102] Slope features (6 dimensions): rising / falling slopes between peaks, reflecting the rate of hemodynamic changes;

[0103] Ratio features (3 dimensions): relative relationships between peaks and bottoms, eliminating individual differences and enhancing feature stability;

[0104] Time features (4 dimensions): pulse width, diastolic time interval (DEIT), reflecting heart cycle and blood flow time;

[0105] Time ratio features (3 dimensions): relative relationships between time intervals, enhancing feature stability and improving model generalization ability.

[0106] Table 1

[0107]

[0108] S5.3 Intelligent waveform quality assessment and screening

[0109] The algorithm automatically detects the integrity of each segmented waveform and evaluates the waveform quality through the following standards:

[0110] (1) Feature point integrity: check whether all key feature points are successfully identified to ensure the integrity of feature extraction

[0111] (2) Physiological reasonableness: verify whether the feature values are within the physiological range to avoid the influence of abnormal values on model training

[0112] (3) Signal quality: evaluate signal-to-noise ratio and baseline stability to ensure the credibility of features

[0113] For incomplete or low-quality waveforms, the algorithm automatically discards them to ensure the reliability of the final feature vector and provide high-quality feature input for subsequent model training.

[0114] S6 Based on the expanded dataset, 20-dimensional features are used to train the initial blood pressure prediction model, which in this embodiment is a diastolic pressure model and a systolic pressure model.

[0115] Specifically, the initial model construction technique employs multiple machine learning algorithms to provide a foundation for subsequent feature subset selection. Multiple algorithms such as SVR, GBR, RF, LR, XGB, etc. are supported, and each algorithm is trained using the complete 20-dimensional waveform features to ensure the comprehensiveness and robustness of model selection. Model performance evaluation is performed on the validation set obtained by splitting the extended data set for each algorithm, providing a performance benchmark for subsequent feature subset selection.

[0116] S7 Based on the Recursive Feature Elimination Cross-Validation (RFECV) algorithm, feature subset selection and optimization are performed on the 20-dimensional features to customize the optimal feature subset combination for different users, eliminate redundant features, and improve model accuracy and generalization ability, as shown in Figure 11

[0117] Multi-model feature selection based on RFECV

[0118] The Recursive Feature Elimination Cross-Validation (RFECV) algorithm is used to independently select features for each machine learning algorithm, ensuring the comprehensiveness and accuracy of the selection. The RFECV algorithm eliminates the least important features recursively while using cross-validation to evaluate model performance, enabling the selection of the optimal feature subset.

[0119] Feature stability analysis

[0120] The stability threshold mechanism is used to select waveform features with an appearance frequency of more than 50% as stable features, avoiding overfitting. Feature stability analysis ensures the reliability of feature selection results and avoids unstable selection results due to data partition differences.

[0121] Target-oriented optimization

[0122] SBP and DBP select the optimal feature combination, considering the differences in feature requirements for different blood pressure indicators, achieving precise feature customization. SBP and DBP have different physiological mechanisms and influencing factors, and by optimizing the feature combination separately, they can better capture their respective physiological characteristics.

[0123] User adaptability analysis

[0124] Adjust the feature selection strategy according to the characteristics of the user's PPG signal to achieve personalized feature subset configuration and lay the foundation for personalized modeling. Different users' PPG signals have different feature patterns, and through user adaptability analysis, the most suitable feature combination can be selected for each user.

[0125] ​S8 will re-initialize the feature subset after optimization to train the blood pressure prediction model, and compare the performance with the model trained in S6, select the model with superior performance as the high-precision personalized blood pressure prediction model, realize accurate modeling of user physiological characteristics, and realize user customized high-precision blood pressure prediction, such as Figure 12

[0126] Feature subset retraining

[0127] The optimized feature subset is re-input into the model training to ensure the best matching of the model and the features. By retraining the model using the optimized feature subset, the influence of redundant features can be eliminated, and the precision and generalization ability of the model can be improved.

[0128] Feature adaptation optimization

[0129] Adjust the model structure according to the user feature subset to ensure the best matching of the model and the features. Through feature adaptation optimization, the user-specific physiological characteristics can be fully utilized to improve the prediction accuracy of the model.

[0130] S9 will extract features from the real-time collected PPG signal through intelligent waveform segmentation and feature extraction algorithm, construct a personalized model feature combination according to the feature subset requirements of the high-precision personalized blood pressure prediction model finally trained, and finally input the high-precision personalized blood pressure prediction model, thereby obtaining accurate blood pressure prediction value.

[0131] Specifically, by inputting the collected user high-quality PPG data into the trained model, more accurate systolic blood pressure (SBP) and diastolic blood pressure (DBP) can be obtained. Compared with traditional methods, the method of the present application has a significant improvement in blood pressure prediction accuracy.

[0132] Embodiment 2

[0133] A kind of all-weather PPG blood pressure monitoring system based on narrow spectrum light source and narrow band response organic light detector, as shown in Figure 1 It includes:

[0134] The signal acquisition module acquires user high-quality PPG signals by synchronously controlling narrow spectrum light source and narrow band response organic photoelectric detector according to time sequence through fully integrated analog front-end chip;

[0135] The PPG signal processing module communicates with the fully integrated analog front-end chip based on SPI communication protocol through microcontroller, obtains the collected user high-quality PPG signal, and then transmits the PPG signal to the computer terminal host computer through UART communication protocol;

[0136] ​The host computer, the computer end host computer back end, obtains the high-quality PPG signal data of the user through the UART communication protocol, analyzes according to the protocol, inputs the high-quality PPG signal data of the user into the intelligent waveform segmentation and feature extraction algorithm, obtains 20-dimensional high-quality waveform features, selects feature combinations according to the requirements of the personalized blood pressure prediction model, finally inputs the selected feature combinations into the model, obtains the high-precision blood pressure value, and finally displays the real-time PPG waveform collected by the user and the predicted blood pressure value on the front-end user interface.

[0137] The signal acquisition module uses the AFE4490 full-integrated analog front-end chip of TI company, mainly including a programmable integrated LED driver, a 22-bit integrated analog-to-digital converter (ADC) and a low-noise detector receiving channel, and an LED fault diagnosis circuit, which can perfectly realize the programmable driving current intensity regulation of the narrow spectrum light source, the high-precision analog-to-digital conversion of the narrow-band response organic photodetector output signal, and the TIA transimpedance amplification and hardware low-pass filtering of the PPG signal. Through flexible pulse sequencing and timing control, high-quality PPG signals of users can be collected.

[0138] Specifically, the narrow spectrum light source in the signal acquisition module is prepared by using Mn4+ doped fluorogallate single crystal red light material, with an emission peak of 627 nm and a half peak width of 3.2 nm, and the power is controlled within the safe range for human body. The light source is connected to the programmable integrated LED driver of the AFE4490 chip, and the driving current is provided by the programmable current source inside the chip. The AFE4490 chip supports high-precision LED driving current accurate regulation in the range of 0-200mA, and has a dynamic range of 110dB (low noise can be realized at low LED current), which can flexibly adjust the light source intensity according to the skin characteristics of different users and environmental conditions, and ensure the best signal quality.

[0139] The narrow-band response organic photodetector has a response peak at 655 nm and a half-peak width of 85 nm. The low-noise detector receiving channel connected to the AFE4490 chip. This channel contains a transimpedance amplifier (TIA) that converts the weak current signal generated by the photodetector into a voltage signal and provides 7 programmable gain settings (feedback resistor RF range: 10kΩ, 25kΩ, 50kΩ, 100kΩ, 250kΩ, 500kΩ, 1MΩ). After amplification by the TIA, the signal is amplified again by a programmable gain amplifier (PGA) with a gain range of 0dB / 3.5dB / 6dB / 9.5dB / 12dB (corresponding to linear gain 1× / 1.5× / 2× / 3× / 4×). Then, the signal is preliminarily filtered by a low-pass filter with a fixed 500Hz cutoff frequency to remove high-frequency noise. Finally, the processed signal is sent to a 22-bit high-precision ADC for digital conversion, with a sampling rate determined by the pulse repetition frequency (PRF), up to 4×PRF (PRF range: 62.5SPS to 5k SPS, corresponding to a maximum sampling rate of 20kSPS), ensuring the capture of subtle changes in the PPG signal.

[0140] The AFE4490 chip also provides flexible timing control functions that can accurately control the lighting time and sampling time of the LED, enabling time-division multiplexing operation, reducing power consumption and improving signal quality. The built-in LED fault diagnosis circuit in the chip can monitor the working state of the LED in real time, ensuring reliable operation of the system.

[0141] The PPG signal processing module uses a high-performance microcontroller with a main frequency of 20MHz, built-in 2KB RAM and 32KB Flash memory, sufficient to handle complex signal processing tasks. The microcontroller communicates with the AFE4490 chip through the SPI communication protocol, writing relevant register configuration instructions during the system initialization stage to control the driving of the narrow-spectrum light source and the narrow-band organic photodetector to read signals in time sequence. SPI communication uses full-duplex mode with a clock frequency of 2MHz, ensuring efficient data transmission.

[0142] After receiving the original PPG data from the AFE4490 chip, the microcontroller performs preliminary data processing, including data format conversion and simple filtering. The processed PPG signal data is transmitted to the computer end host computer through the UART communication protocol. UART communication uses a baud rate of 115200, ensuring stable and reliable data transmission.

[0143] The host computer is divided into front end and back end two parts. The front end adopts modern UI design, has an intuitive user interface, can display the collected user PPG waveform in real time, and the real-time updated blood pressure value. Various control buttons are also provided on the interface, allowing users to start / stop data collection, freeze waveform window, save historical data, etc.

[0144] The back end reads and parses the PPG signal data transmitted by the microcontroller through the UART protocol, and performs data preprocessing on the PPG data, including software band-pass filtering, baseline correction, etc. The software band-pass filter uses a Butterworth filter, with a cutoff frequency set to 0.8-10Hz, effectively removing high-frequency noise; the baseline correction uses a moving average method with a window size of 5 seconds, effectively eliminating baseline drift. The preprocessed PPG signal data is input into the intelligent waveform segmentation and feature extraction algorithm to obtain 20-dimensional high-quality waveform features, then the features are selected according to the requirements of the personalized blood pressure prediction model, and finally the selected feature combination is input into the model to obtain a high-precision blood pressure prediction value.

[0145] The core advantage of the system is the cooperative work of narrow-spectrum light source and narrow-band response organic photodetector, which effectively reduces environmental light interference and motion artifacts, making the collected PPG signal have high quality and not be affected by environmental light interference. At the same time, make full use of the collected high-quality PPG signal of the user, through small sample data set expansion, intelligent waveform feature extraction, recursive feature elimination cross-validation algorithm optimization feature subset, etc. to build a high-precision personalized blood pressure prediction model for the user, the predicted blood pressure value is more accurate, and finally realize high-precision, all-weather PPG blood pressure health monitoring. Compared with traditional blood pressure monitoring methods, the system does not need a cuff, is more convenient to operate, has better user experience, and can realize continuous monitoring, providing more comprehensive blood pressure health management for users.

[0146] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited by the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application are equivalent replacement methods, which are all included in the protection scope of the present application.

Claims

1. A PPG blood pressure monitoring method, characterized by, The application relates to a personalized blood pressure prediction method and device. The PPG signal of a user is collected by using a narrow-spectrum light source and a narrow-band response organic light detector to form a data set; The PPG signal of the user is cut by a sliding window to expand the data set; The expanded data set is preprocessed, specifically, an intelligent waveform segmentation and feature extraction algorithm is used to extract 20-dimensional waveform features of each window, discard incomplete waveforms, and then extract the mean value of the complete waveform feature data in each window to obtain 20-dimensional features of each window; A plurality of machine learning algorithms are used to train an initial blood pressure prediction model by using the 20-dimensional features; The 20-dimensional features are optimized by using a recursive feature elimination cross-validation algorithm to obtain an optimal feature subset combination of each model; The initial blood pressure prediction model is retrained by using the optimal feature subset combination, and the training results obtained in the last step are compared in performance to finally obtain a personalized blood pressure prediction model; The PPG signal collected in real time is extracted by the intelligent waveform segmentation and feature extraction algorithm, and a feature subset combination is constructed according to the requirements of the personalized blood pressure prediction model, and finally input into the personalized blood pressure prediction model to obtain a blood pressure prediction value.

2. The PPG blood pressure monitoring method of claim 1, wherein, The narrow-spectrum light source is a human body safety power narrow-spectrum red light source with an emission peak value of 627 nm and a half peak width of 3.2 nm, which is prepared by using a Mn4+ doped fluorogallate single crystal red light material.

3. The PPG blood pressure monitoring method of claim 1, wherein, The narrow-band response light detector is prepared by introducing a specific absorption spectrum optical depletion layer and an organic photoactive layer to obtain a narrow-band response organic photodetector with a response peak value of 655 nm, wherein the process conditions of the optical depletion layer and the organic photoactive layer are as follows: Optical depletion layer: P3HT: J71: IEICO-4F is dissolved in chloroform, a film with a thickness of about 400-500 nm is formed after spin coating, and thermal annealing is carried out at 100 DEG C for 10 minutes; Organic photoactive layer: PM6: L8-BO is dissolved in chloroform, a film with a thickness of about 150-200 nm is formed after spin coating, and thermal annealing is carried out at 100 DEG C for 10 minutes.

4. The PPG blood pressure monitoring method of claim 1, wherein, The PPG signal of the user is cut by a sliding window to expand the data set, specifically as follows: 875-point sliding windows are used to cut the 5000-point PPG signal collected each time; Each window inherits the SBP / DBP label of the original file; Effective windows are automatically screened through waveform quality evaluation; The data set is expanded by 5-10 times from the file level to the window level.

5. The PPG blood pressure monitoring method of claim 1, wherein, The intelligent waveform segmentation and feature extraction algorithm comprises the following steps: PPG waveform intelligent segmentation based on trough detection; Identify the peak value of the tidal wave through first-order derivative analysis; Automatically detect and discard incomplete waveforms; Extract 20-dimensional waveform features; Calculate the mean value of the complete waveform features to generate 20-dimensional features of the window.

6. The PPG blood pressure monitoring method of claim 1, wherein, The PPG signal comprises a main wave, a tidal wave, a descending middle isthmus and a double beat wave.

7. The PPG blood pressure monitoring method of claim 6, wherein, The 20-dimensional features of each window comprise peak features, slope features, ratio features, time features and time ratio features.

8. The PPG blood pressure monitoring method of claim 1, wherein, The plurality of machine learning algorithms comprise SVR, GBR, RF, LR and XGB.

9. The PPG blood pressure monitoring method of claim 7, wherein, The 20-dimensional waveform features comprise: Peak features: systolic pressure peak value, tidal wave peak value, diastolic pressure peak value, bottom value; Slope features: rising / descending slope between peaks; Ratio features: relative relationship between peak and bottom; Time features: pulse width, diastolic time interval; Time ratio features: relative relationship between time intervals.

10. A system implementing the PPG blood pressure monitoring method of any one of claims 1-9, characterized in that, Comprise: A signal acquisition module, which adopts a fully integrated analog front-end module, connects a narrow-spectrum light source and a narrow-band response organic photodetector, and acquires PPG signals of a user; A PPG signal processing module, which is used for controlling and driving the narrow-spectrum light source and the narrow-band response organic photodetector to read signals and features in a time sequence, and transmitting the acquired PPG signals and blood oxygen data to a computer terminal host computer through a UART protocol; A host computer, which inputs a personalized blood pressure prediction model after preprocessing and feature extraction of the PPG signals, and obtains a blood pressure value.

Images