PPG blood pressure estimation method and system combining deep learning and clinical priori knowledge

Abstract

The invention discloses a PPG blood pressure estimation method and system combining deep learning and clinical priori knowledge. The PPG blood pressure estimation method combining deep learning and clinical priori knowledge comprises the following steps: firstly, acquiring a pulse wave signal through a signal acquisition module; preprocessing the pulse wave signal, and respectively outputting a first branch pulse wave signal and a second branch pulse wave signal; performing dimension conversion on the first branch pulse wave signal, extracting multi-dimensional features through a deep learning network, and extracting clinical priori knowledge features in the second branch pulse wave signal; and finally, constructing a blood pressure estimation model of a corresponding time period according to the multi-dimensional features and the clinical priori knowledge features, and predicting and outputting a blood pressure value of the corresponding time period through the blood pressure estimation model, so that the correlation between the extracted PPG features and BP and the blood pressure measurement precision are improved by combining deep learning and double-branch feature extraction of clinical priori knowledge. The effectiveness and interpretability of the blood pressure estimation method are ensured, and the credibility of the estimated blood pressure value is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of blood pressure monitoring, and in particular to a PPG blood pressure estimation method and system combining deep learning and clinical prior knowledge. Background Art

[0002] Traditional blood pressure monitoring methods include non-invasive and invasive. Non-invasive blood pressure measurement requires a cuff and inflation, which results in a longer measurement time and makes it difficult to achieve long-term continuous monitoring of blood pressure. Although invasive blood pressure measurement can monitor blood pressure continuously for a long time, it is invasive and will increase the risk of infection or even make it more serious.

[0003] The existing blood pressure monitoring method based on photoplethysmography can avoid the inflation and deflation process of the cuff and realize cuff-free blood pressure measurement. It uses photoplethysmography to measure blood pressure, that is, first extracting multiple features from the PPG (Photoplethysmogram) signal, and then fitting the regression line using linear regression and other methods, and using the extracted features to estimate blood pressure. It can also use a neural network and deep learning-based method to use PPG as the input of the network, and use continuous arterial blood pressure as a label for supervised learning, output continuous BP (Blood pressure) values, and construct an end-to-end method. However, the accuracy of the blood pressure measurement method based only on PPG features depends on the validity of the extracted features, and there are problems of low accuracy and poor stability. Moreover, the use of deep learning to build an end-to-end measurement model is overly dependent on the amount of data. In addition, the data standard blood pressure value used in the method is a continuous arterial blood pressure waveform, which requires invasive measurement and poses an infection risk. Summary of the invention

[0004] The purpose of the present invention is to design a PPG blood pressure estimation method that combines deep learning and clinical prior knowledge in order to solve the above problems.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] The PPG blood pressure estimation method combining deep learning and clinical prior knowledge includes the following steps:

[0007] Acquire pulse wave signal;

[0008] Preprocessing the pulse wave signal, and outputting a first branch pulse wave signal and a second branch pulse wave signal respectively;

[0009] Performing dimension conversion on the first branch pulse wave signal and extracting multi-dimensional features through a deep learning network;

[0010] Extracting clinical prior knowledge features from the second branch pulse wave signal;

[0011] A blood pressure estimation model for a corresponding time period is constructed based on the multidimensional features and the clinical prior knowledge features, and the blood pressure value for the corresponding time period is predicted and outputted through the blood pressure estimation model.

[0012] The present invention further proposes a PPG blood pressure estimation system combining deep learning and clinical prior knowledge, the PPG blood pressure estimation system combining deep learning and clinical prior knowledge is used to implement the above-mentioned PPG blood pressure estimation method combining deep learning and clinical prior knowledge, the PPG blood pressure estimation system combining deep learning and clinical prior knowledge comprises:

[0013] A signal acquisition module, used for acquiring a pulse wave signal;

[0014] A preprocessing module, used for preprocessing the pulse wave signal;

[0015] Dual-branch feature extraction module, used to extract multi-dimensional features and clinical prior knowledge features;

[0016] The blood pressure estimation model automatic construction module is used to automatically construct the blood pressure estimation model of the corresponding time period according to the multidimensional features and clinical prior knowledge features, and output the blood pressure value of the corresponding time period.

[0017] The beneficial effects of the present invention are:

[0018] The PPG blood pressure estimation method combining deep learning and clinical prior knowledge first obtains the pulse wave signal through the signal acquisition module; then preprocesses the pulse wave signal to output the first branch pulse wave signal and the second branch pulse wave signal respectively; then, the first branch pulse wave signal is dimensionally converted and multidimensional features are extracted through a deep learning network to extract clinical prior knowledge features in the second branch pulse wave signal; finally, a blood pressure estimation model for the corresponding time period is constructed based on the multidimensional features and the clinical prior knowledge features, and the blood pressure value for the corresponding time period is predicted and output through the blood pressure estimation model, thereby improving the correlation between the extracted PPG features and BP and the blood pressure measurement accuracy through the dual-branch feature extraction combining deep learning and clinical prior knowledge, ensuring the effectiveness and interpretability of the blood pressure estimation method, and improving the credibility of the estimated blood pressure value. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of the process of the PPG blood pressure estimation method combining deep learning and clinical prior knowledge of the present invention;

[0020] Figure 2 The overall block diagram of the PPG blood pressure estimation method combining deep learning and clinical prior knowledge of the present invention;

[0021] Figure 3The convolution signal waveform diagram of the PPG blood pressure estimation method combining deep learning and clinical prior knowledge of the present invention;

[0022] Figure 4 A diagram of the systolic peak interval of the PPG blood pressure estimation method combining deep learning and clinical prior knowledge of the present invention;

[0023] Figure 5 A schematic diagram of PPG segment segmentation of a PPG blood pressure estimation method combining deep learning and clinical prior knowledge according to the present invention;

[0024] FIG6 is a schematic diagram of dimensional transformation of the PPG blood pressure estimation method combining deep learning and clinical prior knowledge according to the present invention;

[0025] FIG7 is a schematic diagram of waveform feature extraction based on clinical prior knowledge of the PPG blood pressure estimation method of the present invention that combines deep learning with clinical prior knowledge. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0027] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0029] In the description of the present invention, it should be understood that the terms "upper", "lower", "inside", "outside", "left", "right", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are directions or positional relationships in which the product of the invention is usually placed when in use, or are directions or positional relationships commonly understood by those skilled in the art. These directions or positional relationships are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0030] Furthermore, the terms “first”, “second”, etc. are merely used for distinguishing descriptions and should not be understood as indicating or implying relative importance.

[0031] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms such as "setting" and "connection" should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0032] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0033] like Figure 1 As shown in Figure 7, the PPG blood pressure estimation method combining deep learning and clinical prior knowledge includes the following steps:

[0034] Acquire pulse wave signal;

[0035] Preprocessing the pulse wave signal, and outputting a first branch pulse wave signal and a second branch pulse wave signal respectively;

[0036] Performing dimension conversion on the first branch pulse wave signal and extracting multi-dimensional features through a deep learning network;

[0037] Extracting clinical prior knowledge features from the second branch pulse wave signal;

[0038] A blood pressure estimation model for a corresponding time period is constructed based on the multidimensional features and the clinical prior knowledge features, and the blood pressure value for the corresponding time period is predicted and outputted through the blood pressure estimation model.

[0039] In one embodiment, the step of converting the first branch pulse wave signal into a dimension and extracting multi-dimensional features through a deep learning network is specifically as follows:

[0040] Converting the first branch pulse wave signal into a two-dimensional space to form a visibility map;

[0041] Based on the VGG19 model, extract local features and texture information in the visibility map;

[0042] The convolution module and the MobileViT v2 module are combined to extract the global topological structure in the visibility graph.

[0043] In one embodiment, the clinical prior knowledge features include time domain features, morphological features and statistical features.

[0044] In one embodiment, the step of constructing a blood pressure estimation model for a corresponding time period according to the multidimensional features and the clinical prior knowledge features is specifically as follows:

[0045] Performing feature fusion processing on the multidimensional features and the clinical prior knowledge features, and outputting a feature vector set for each time period;

[0046] A blood pressure estimation model for the corresponding time period is automatically constructed according to the feature vector set for each time period.

[0047] In one embodiment, the step of preprocessing the pulse wave signal is specifically:

[0048] Firstly, eliminating the noise signal and discontinuous / saturated signal in the pulse wave signal, and filtering out the abnormal segments in the pulse wave signal;

[0049] The pulse wave signal is then finely segmented.

[0050] In this embodiment, a single-channel pulse wave sensor is first used to collect pulse wave signals at the fingertips. The single measurement time is short and data collection is simpler. The preprocessing module is then used to eliminate noise signals and discontinuous / saturated signals in the pulse wave signal, and to filter out abnormal segments in the pulse wave signal to improve the quality of the PPG signal. The pulse wave signal is then segmented.

[0051] Specifically, the noise signal and discontinuous / saturated signal in the pulse wave signal are eliminated as follows: for the PPG data in the morning, afternoon and evening time periods, a 5th-order Butterworth bandpass filter with a passband of 0.5 Hz-5 Hz is first used to eliminate baseline drift and high-frequency noise; then discontinuous signals and signals with saturated amplitude are removed: the PPG signal is divided into 10-second segments, and in each 10-second PPG segment, if any three adjacent sampling points have equal values, the segment contains discontinuous or saturated PPG data, and this 10-second PPG segment is deleted.

[0052] In this embodiment, the abnormal segments in the pulse wave signal are screened out as follows: first, the amplitude of the PPG signal in all segments is truncated to remove the amplitude value less than zero, leaving the PPG signal with positive amplitude for easy processing; then, two different convolution kernels are defined:

[0053]

[0054]

[0055] Among them, f sis the sampling frequency of PPG, w1 and w2 are two manually defined time variables, which are used to highlight each cardiac cycle (beat) and PPG contraction peak, respectively. The values ​​of w1 and w2 are roughly equal to the duration of each cardiac cycle and each contraction peak, respectively, which are set to 0.667s and 0.111s; then, these two convolution kernels are applied to the filtered PPG segment to produce two convolved signals, namely, the convolution signal PPG beat and PPG peak :

[0056]

[0057]

[0058] Among them, PPG seg represents the filtered PPG segment, Figure 3 In the convolution signal PPG beat and PPG peak The different colors show that the convolution produces higher amplitudes within each cardiac cycle and at the peak of systole, respectively.

[0059] To further locate the peak of each cardiac contraction, a threshold curve was calculated using the following formula, denoted as TH. peak :

[0060]

[0061] Among them, α is related to PPG beat The associated offset is defined as:

[0062]

[0063] Where z is PPG seg In this embodiment, the coefficient β is manually set to 0.03.

[0064] Figure 4 PPG is also displayed seg 、PPG peak and TH peak , TH peak The sawtooth waveform indicates the position of each cardiac cycle in the PPG, and the PPG peak Each peak of corresponds to each contraction peak position; therefore, we look for PPG peak Greater than TH peak The intervals of are regarded as the intervals containing the PPG contraction peaks. The number of contraction peaks in each PPG segment is obtained by counting the intervals in each segment.

[0065] Normally, the normal heart rate of an adult ranges from 60 to 100 beats per minute, that is, under normal circumstances, a 10-second PPG segment has at least 10 cardiac cycles, resulting in 10 systolic peaks. Since the systolic peaks at the beginning and end of each segment may not be completely included in the segment, we set the threshold of the number of systolic peaks to 8, and only retain PPG segments with more than 8 systolic peaks to avoid including poor quality segments in the PPG signal.

[0066] Through the above operations, 3112 10-second PPG segments were finally obtained in the morning, afternoon and evening time periods; the means and standard deviations of SBP (Systolic pressure) and DBP (Diastolic pressure) corresponding to the three time periods were 108.05±14.83mmHg (SBP) and 70.02±10.36mmHg (DBP), 108.01±14.25mmHg (SBP) and 71±9.70mmHg (DBP), and 112.17±9.71mmHg (SBP) and 73.90±12.57mmHg (DBP), respectively.

[0067] After the above operation, the PPG signal has been segmented into 10-second segments, ensuring the existence of a normal waveform. In order to obtain more detailed information in a smaller time interval, the PPG signal is further segmented in this embodiment; specifically, a systolic peak is added before and after each cardiac cycle to capture the PPG signal of each cycle and its changing trend. Figure 5 To further segment the 10-second PPG segment, a ROI (Region of Interest) was defined for every three PPG contraction peaks. Each PPG ROI included a complete cardiac cycle, as well as the previous descending branch and the next ascending branch of the previous cycle and the next cycle, forming a PPG window.

[0068] Since one-dimensional analysis alone is difficult to reveal the hidden patterns and dynamic characteristics in the PPG signal, the PPG signal needs to be converted into a two-dimensional space for analysis, and the PPG window is converted into a corresponding two-dimensional image according to the VG (Visibility Graph); wherein VG is an undirected graph generated based on the natural visibility between sample amplitudes in the time series, and natural visibility refers to the connectivity between each node in the one-dimensional sequence.

[0069] Specifically, the amplitudes of all extracted PPG windows are first remapped to the range of 0 to 1; in order to determine the natural visibility between points in the sequence, the amplitude peaks of the PPG are regarded as nodes with different heights, as shown in Figure 6 (a), and the visibility between two nodes is determined by the line segment connecting them. If the line segment connecting two nodes intersects with the amplitudes of other nodes, the two nodes do not have natural visibility; conversely, if the line segment connecting two nodes does not intersect with the amplitude of any other sample, the two points have natural visibility. In Figure 6 (a), the two nodes at both ends of the red line segment have natural visibility because the line does not pass through other amplitudes; and there is no natural visibility between the two nodes at both ends of the blue line segment because their line passes through other amplitudes; the line segment connecting any node with the remaining nodes is regarded as an undirected edge, and each node and all undirected edges form an undirected graph.

[0070] Then, a binary image is generated based on the adjacency matrix of this undirected graph, whose horizontal and vertical coordinates represent the serial numbers of the nodes in the undirected graph; the value of each pixel in the binary image depends on the natural visibility between the two nodes with the horizontal and vertical coordinates as the serial numbers. If there is natural visibility between the two nodes, the pixel value is set to 1, otherwise it is set to 0. Therefore, all PPG sequences are converted into a symmetrical image, namely VG, through the adjacency matrix of their undirected graphs, as shown in Figure 6, where the red and blue edges in (a) correspond to the white and black pixels in (b), respectively.

[0071] In order to improve the correlation between the extracted PPG features and BP and the accuracy of blood pressure measurement, this embodiment uses a dual-branch feature extraction module that combines deep learning and clinical prior knowledge to extract features, that is, extracting the multidimensional features of one branch PPG signal and simultaneously extracting the clinical prior knowledge features of the other branch PPG signal to ensure the effectiveness and interpretability of the blood pressure estimation method and improve the credibility of the blood pressure value estimated by the method.

[0072] This embodiment uses the transfer learning model in deep learning for feature extraction, that is, the two complementary network architectures of VGG19 model and MobileViT v2 module are used to extract local connection patterns, global topological structures and multi-scale information from VG.

[0073] It can be understood that in order to meet the model input requirements of three-channel RGB images, each VG needs to be replicated three times to create a three-channel image, and then the image size is resized to 224×224 using bilinear interpolation; specifically, the first network is based on the modified VGG19 model, which extracts rich local features and texture information from the VG image to obtain a 4096-dimensional feature vector, so as to understand the local changes of the PPG signal related to DBP; the second network combines the convolution module with the attention mechanism of MobileViTv2 to capture the global topological structure in the VG image, and obtains an 896-dimensional feature vector, so as to understand the overall dynamic characteristics of the PPG signal related to SBP. The two complementary feature vectors extracted by the two networks will be used as the regression input of the DBP and SBP estimation models respectively.

[0074] In order to address the limitations of poor interpretability of feature extraction by deep learning networks and lack of application of clinical prior knowledge, this embodiment extracts a series of features with physiological significance and clinical interpretability from the PPG waveform, that is, 72 feature parameters are extracted from each PPG window, and the feature parameters include time domain features, morphological features and statistical feature classes.

[0075] Specifically, the time domain features can directly reflect the basic characteristics of the PPG waveform, such as amplitude, duration and slope, which are all related to the physiological mechanism of blood pressure. As shown in Figure 7(a), 12 time domain features are extracted from the PPG waveform according to the amplitude, duration and specific slope position. The time domain features extracted from the PPG waveform include PE (peak height), VD (Valley depth), PD (Period duration), AS (Ascending slope), DS (Descending slope), WH (Waveform height), WD (Waveform depth), AD (Ascending duration), DD (Descending duration), AA (Ascending area), DA (Descending area) and K value. Among them, the K value is calculated by the maximum, average and minimum values ​​of the PPG signal through the formula:

[0076]

[0077] Among them, PPG period Represents a complete PPG waveform cycle.

[0078] The morphological features can describe the geometric shape of the waveform in more detail and capture subtle changes that may reflect the dynamic characteristics of the vascular system. The morphological features include 8 characteristic parameters in 6 groups, that is, the PPG waveform is described by 48 characteristic parameters; the PPG waveform is divided into two parts vertically from the PEAK point. For the left part, the WH is divided into 10 parts on the decomposition line, and the horizontal distance from each WH segment point to the left waveform is calculated to obtain SW[x]; according to the distance from the WD segment point to the right waveform, the DW[x] of the right part is calculated, and the corresponding sum and ratio of SW[x] and DW[x] can obtain the sum SSD of SW[x] and DW[x] and the ratio RSD of SW[x] and DW[x].

[0079] Statistical features can reflect the overall trend and variability of PPG signals, which are related to the long-term blood pressure regulation mechanism. This embodiment uses the central tendency, discreteness and shape principles to extract six statistical features to describe the PPG signal as a probability distribution. The statistical features include mean ME, variance VAR, MF (Marginal factor), FF (Form factor), skewness SK and kurtosis KU. The calculation formulas of the six statistical features are as follows:

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] in, is the sampling point in the PPG cycle, and n represents the length of the signal cycle; the statistical features not only supplement the local information, but also capture the long-term changes related to the regulation of the autonomic nervous system and vascular compliance.

[0087] In order to solve the risk of model overfitting, the nonlinear interaction between features makes the use of traditional linear regression models complicated, and the dedicated models for SBP and DBP estimation in different time periods increase the complexity, this embodiment adopts AutoML (Automated Machine Learning) as the modeling strategy. AutoML handles complex data sets and regression tasks with minimal human intervention and is easy to deploy. It optimizes the machine learning pipeline by exploring the combination of preprocessing steps, models and hyperparameters. In this embodiment, TPOT (Tree-based Pipeline Optimization Tool) is preferably used as the AutoML tool to simplify the model selection and parameter tuning process.

[0088] It can be understood that the multidimensional features and clinical prior knowledge features are firstly concatenated to obtain a feature vector set for blood pressure estimation in each time period; then, TPOT generates a random population according to the feature vector set of each time period, and each individual represents a random machine learning scheme. Specifically, this embodiment uses five-fold cross-validation to evaluate the adaptability of each scheme, and the adaptability uses MSE (Mean-squared error) as an evaluation indicator. The adaptability is that TPOT simulates natural selection to select schemes with better performance for reproduction, and generates new schemes through crossover and mutation. Crossover is to combine parts of two schemes into a new scheme, and mutation is to randomly change part of the scheme. TPOT optimizes the pipeline through selection, crossover and mutation iteration, and outputs the pipeline with the best performance; finally, the optimized scheme will train a dedicated model for SBP and DBP estimation for each time period respectively, and each dedicated model can output the predicted values ​​of SBP and DBP for the corresponding time period, so as to achieve the problem of solving the diurnal variation of BP by modeling different time periods separately, thereby improving the stability and practicality of BP measurement.

[0089] The PPG blood pressure estimation method combining deep learning and clinical prior knowledge of the present invention first obtains the pulse wave signal through the signal acquisition module; then preprocesses the pulse wave signal, and outputs the first branch pulse wave signal and the second branch pulse wave signal respectively; then, the first branch pulse wave signal is dimensionally converted and multidimensional features are extracted through the deep learning network, and the clinical prior knowledge features in the second branch pulse wave signal are extracted; finally, a blood pressure estimation model for the corresponding time period is constructed according to the multidimensional features and the clinical prior knowledge features, and the blood pressure value for the corresponding time period is predicted and output through the blood pressure estimation model, thereby improving the correlation between the extracted PPG features and BP and the blood pressure measurement accuracy through the dual-branch feature extraction combining deep learning and clinical prior knowledge, ensuring the effectiveness and interpretability of the blood pressure estimation method, and improving the credibility of the estimated blood pressure value.

[0090] The present invention also proposes a PPG blood pressure estimation system that combines deep learning and clinical prior knowledge. The PPG blood pressure estimation system that combines deep learning and clinical prior knowledge is used to implement the above-mentioned PPG blood pressure estimation method that combines deep learning and clinical prior knowledge. The specific method of the PPG blood pressure estimation method that combines deep learning and clinical prior knowledge refers to the above-mentioned embodiment. Since the PPG blood pressure estimation system that combines deep learning and clinical prior knowledge adopts all the technical solutions of all the above-mentioned embodiments, it at least has all the beneficial effects brought by the technical solutions of the above-mentioned embodiments, which will not be elaborated here one by one.

[0091] In one embodiment, the PPG blood pressure estimation system combining deep learning and clinical prior knowledge includes:

[0092] A signal acquisition module, used for acquiring a pulse wave signal;

[0093] A preprocessing module, used for preprocessing the pulse wave signal;

[0094] Dual-branch feature extraction module, used to extract multi-dimensional features and clinical prior knowledge features;

[0095] The blood pressure estimation model automatic construction module is used to automatically construct the blood pressure estimation model of the corresponding time period according to the multidimensional features and clinical prior knowledge features, and output the blood pressure value of the corresponding time period.

[0096] In one embodiment, the signal acquisition module is a single-channel pulse wave sensor.

[0097] In this embodiment, the PPG blood pressure estimation system that combines deep learning and clinical prior knowledge collects pulse signals through a signal acquisition module, and realizes blood pressure estimation based on pulse wave through a preprocessing module, a dual-branch feature extraction module and a blood pressure estimation model automatic construction module.

[0098] Since the blood pressure measurement method based on PTT (Pulse transmit time) explores the relationship between PWV (Pulse wave velocity) and blood pressure, various physical models are constructed to represent blood pressure through PWV, so as to achieve cuffless blood pressure measurement. Among them, PWV is generally quantified by PTT, and the calculation of PTT requires the acquisition of two or more physiological signals, usually PPG and another physiological signal, such as ECG (Electrocardiogram). PTT is calculated by the time difference between the peaks of two synchronous signals. Multi-wavelength PPG can also be used to replace multiple physiological signals, but multi-wavelength PPG needs to be complexly deconstructed and reshaped. These methods need to collect multiple physiological signals from different parts or different parts at the same time or need to perform complex signal processing. There are problems such as complex signal acquisition process, low signal quality and complex signal processing process. Therefore, the signal acquisition module in this embodiment is preferably a single-channel pulse wave sensor, that is, a single-channel PPG signal is collected at the fingertips through a single pulse heart rate sensor, and a single measurement does not exceed 5 minutes to ensure the simplicity of data collection. The collected single-channel PPG does not need to be processed in a complicated way, and only needs to go through conventional denoising and screening methods to ensure the quality of the data used, thereby ensuring the speed of the signal preprocessing process, shortening the data processing time, and avoiding the simultaneous collection of multiple physiological signals to reduce the number of sensors in the signal acquisition module.

[0099] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A PPG blood pressure estimation method combining deep learning and clinical prior knowledge, characterized in that: The PPG blood pressure estimation method combining deep learning and clinical prior knowledge comprises the following steps: Acquire pulse wave signal; Preprocessing the pulse wave signal, and outputting a first branch pulse wave signal and a second branch pulse wave signal respectively; Performing dimension conversion on the first branch pulse wave signal and extracting multi-dimensional features through a deep learning network; Extracting clinical prior knowledge features from the second branch pulse wave signal; A blood pressure estimation model for a corresponding time period is constructed based on the multidimensional features and the clinical prior knowledge features, and the blood pressure value for the corresponding time period is predicted and outputted through the blood pressure estimation model.

2. The PPG blood pressure estimation method combining deep learning and clinical prior knowledge according to claim 1, characterized in that: The step of converting the dimension of the first branch pulse wave signal and extracting multi-dimensional features through a deep learning network is specifically as follows: Converting the first branch pulse wave signal into a two-dimensional space to form a visibility map; Based on the VGG19 model, extract local features and texture information in the visibility map; The convolution module and the MobileViT v2 module are combined to extract the global topological structure in the visibility graph.

3. The PPG blood pressure estimation method combining deep learning and clinical prior knowledge according to claim 1, characterized in that: The clinical prior knowledge features include time domain features, morphological features and statistical features.

4. The PPG blood pressure estimation method combining deep learning and clinical prior knowledge according to claim 1, characterized in that: The step of constructing a blood pressure estimation model for a corresponding time period according to the multidimensional features and the clinical prior knowledge features is specifically as follows: Performing feature fusion processing on the multidimensional features and the clinical prior knowledge features, and outputting a feature vector set for each time period; A blood pressure estimation model for the corresponding time period is automatically constructed according to the feature vector set for each time period.

5. The PPG blood pressure estimation method combining deep learning and clinical prior knowledge according to claim 1, characterized in that: The step of preprocessing the pulse wave signal is specifically as follows: Firstly, eliminating the noise signal and discontinuous / saturated signal in the pulse wave signal, and filtering out the abnormal segments in the pulse wave signal; The pulse wave signal is then finely segmented.

6. A PPG blood pressure estimation system combining deep learning and clinical prior knowledge, characterized in that: The PPG blood pressure estimation system combining deep learning and clinical prior knowledge is used to implement the PPG blood pressure estimation method combining deep learning and clinical prior knowledge as claimed in any one of claims 1 to 5, and the PPG blood pressure estimation system combining deep learning and clinical prior knowledge includes: A signal acquisition module, used for acquiring a pulse wave signal; A preprocessing module, used for preprocessing the pulse wave signal; Dual-branch feature extraction module, used to extract multi-dimensional features and clinical prior knowledge features; The blood pressure estimation model automatic construction module is used to automatically construct the blood pressure estimation model of the corresponding time period according to the multidimensional features and clinical prior knowledge features, and output the blood pressure value of the corresponding time period.

7. The PPG blood pressure estimation system combining deep learning and clinical prior knowledge according to claim 6, characterized in that: The signal acquisition module is a single-channel pulse wave sensor.

Images