Non-invasive blood pressure detection systems, methods, electronic devices, and storage media

Abstract

The application provides a non-invasive blood pressure detection system, method, electronic equipment and storage medium, the system comprises a photoelectric signal acquisition module, an analog signal processing module, a digital signal processing module and a blood pressure calculation module, wherein: the photoelectric signal acquisition module is used for collecting the photoelectric volume pulse wave of specific tissues of the human body; the analog signal processing module is used for amplifying and filtering out analog high-frequency noise of the photoelectric volume pulse wave to obtain a first signal; the digital signal processing module is used for converting the first signal into a digital signal to obtain a second signal, and is also used for signal optimization and signal cleaning of the second signal to obtain a third signal; and the blood pressure calculation module is used for obtaining a blood pressure value based on the third signal and using a 2DBP_Net dimensional transformation convolution network, so that the precision of blood pressure detection can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomedical measurement, and in particular to a non-invasive blood pressure detection system and method, an electronic device and a storage medium. BACKGROUND

[0002] The related art method for measuring blood pressure using a photoplethysmogram includes a method based on pulse transit time and a method based on machine learning. The method based on pulse transit time is more intuitive in theory and calculates the blood vessel pressure according to a hemodynamic model. However, this method does not take into account the changes in the state of the human blood vessels. The research on the method based on machine learning mainly focuses on the representation and extraction of waveform features, and there is still a lack of a unified and convincing guideline for the feature extraction system. The related art method for measuring blood pressure using a photoplethysmogram has low accuracy. SUMMARY

[0003] The present application provides a non-invasive blood pressure detection system and method, an electronic device and a storage medium, which can improve the accuracy of blood pressure detection.

[0004] The present application provides a non-invasive blood pressure detection system, which includes a photoplethysmogram signal acquisition module, an analog signal processing module, a digital signal processing module and a blood pressure calculation module.

[0005] The photoplethysmogram signal acquisition module is used to acquire the photoplethysmogram of a specific tissue of the human body.

[0006] The analog signal processing module is used to amplify the effective information of the photoplethysmogram and filter out analog high-frequency noise to obtain a first signal.

[0007] The digital signal processing module is used to convert the first signal into a digital signal to obtain a second signal, and is also used to optimize and clean the second signal to obtain a third signal.

[0008] The blood pressure calculation module is used to obtain a blood pressure value using a 2DBP_Net dimensional transformation convolution network based on the third signal.

[0009] The non-invasive blood pressure detection system provided by the present application includes an encoding module, a two-dimensional convolution module and a linear connection module, and the two-dimensional convolution module includes a frequency domain transformation method and a two-dimensional convolution network.

[0010] The encoding module is used to add amplitude information and position information in a fourth signal to obtain a fifth signal, and the fourth signal includes the third signal, a first derivative signal of the third signal and a second derivative signal of the third signal.

[0011] The frequency domain transformation method is used for transforming the fifth signal from time domain to frequency domain, determining a plurality of main constituent frequencies of the fifth signal based on frequency domain amplitudes, and converting the plurality of main constituent frequencies into a slicing period, slicing the fifth signal according to the slicing period and stacking the fifth signal in sequence into a two-dimensional signal.

[0012] The two-dimensional convolution network is used for extracting intra-period dependence and extra-period dependence of the two-dimensional signal, one-dimensionally splicing and weightedly averaging the two-dimensional signal according to frequency domain amplitudes, and obtaining a one-dimensional signal.

[0013] The linear connection module is used for obtaining a blood pressure value based on the one-dimensional signal.

[0014] According to the non-invasive blood pressure detection system provided by the application, the digital signal processing module is specifically used for:

[0015] The second signal is standardized, Butterworth filtered, wavelet transformed and slidingly windowed.

[0016] According to the non-invasive blood pressure detection system provided by the application, the digital signal processing module is specifically used for:

[0017] The second signal is signal cleaned by using an isolated forest anomaly detection method.

[0018] The application further provides a non-invasive blood pressure detection method, comprising:

[0019] An optical volume pulse wave of a specific tissue of a human body is collected.

[0020] The optical volume pulse wave is amplified for effective information and filtered for analog high-frequency noise, and a first signal is obtained.

[0021] The first signal is converted into a digital signal, and a second signal is obtained.

[0022] The second signal is signal optimized and signal cleaned, and a third signal is obtained.

[0023] A blood pressure value is calculated by using a 2DBP_Net dimensional transformation convolution network based on the third signal.

[0024] According to the non-invasive blood pressure detection method provided by the application, the blood pressure value is calculated by using the 2DBP_Net dimensional transformation convolution network, comprising:

[0025] Amplitude information and position information are added to a fourth signal to obtain a fifth signal, the fourth signal comprising the third signal, a first derivative signal of the third signal and a second derivative signal of the third signal.

[0026] transform the fifth signal from a time domain to a frequency domain, and determine a plurality of main constituent frequencies of the fifth signal based on a frequency domain amplitude;

[0027] convert the plurality of main constituent frequencies into a slicing period, slice the fifth signal according to the slicing period, and stack the fifth signal in sequence into a two-dimensional signal;

[0028] extract intra-period dependence and extra-period dependence of the two-dimensional signal, and perform one-dimensional splicing and weighted average on the two-dimensional signal according to a frequency domain amplitude, to obtain a one-dimensional signal;

[0029] obtain a blood pressure value based on the one-dimensional signal.

[0030] According to the non-invasive blood pressure detection method provided by the application, the signal optimization of the second signal comprises:

[0031] standardizing, Butterworth filtering, wavelet transforming and sliding windowing the second signal.

[0032] According to the non-invasive blood pressure detection method provided by the application, the signal cleaning of the second signal comprises:

[0033] using an Isolation Forest anomaly detection method to clean the second signal.

[0034] The application further provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the non-invasive blood pressure detection method according to any one of the above when executing the program.

[0035] The application further provides a non-transitory computer readable storage medium, which stores a computer program, wherein the computer program is executable on a processor to implement the non-invasive blood pressure detection method according to any one of the above.

[0036] The application further provides a computer program product, which includes a computer program, wherein the computer program is executable on a processor to implement the non-invasive blood pressure detection method according to any one of the above.

[0037] The non-invasive blood pressure detection system, method, electronic device and storage medium provided by the application use a photoelectric signal acquisition module to acquire a photoelectric volume pulse wave of a specific tissue of a human body, and perform analog signal processing and digital signal processing on the photoelectric volume pulse wave, and finally perform blood pressure detection through a 2DBP_Net dimensional transformation convolution network of a blood pressure calculation module, so that the accuracy of blood pressure measurement of an algorithm can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to make the technical solutions in the present application or prior art clearer, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0039] Figure 1 is one of the structural schematic diagrams of the non-invasive blood pressure detection system provided by the present application;

[0040] Figure 2 is the second structural schematic diagram of the non-invasive blood pressure detection system provided by the present application;

[0041] Figure 3 is the structural schematic diagram of the 2DBP_Net dimension conversion convolution network provided by the present application;

[0042] Figure 4 is the flow schematic diagram of the non-invasive blood pressure detection method provided by the present application;

[0043] Figure 5 is the structural schematic diagram of the electronic device provided by the present application. DETAILED DESCRIPTION

[0044] In order to make the technical solutions in the present application or prior art clearer, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0045] Firstly, the following contents will be introduced:

[0046] The blood pressure detection methods widely used in the related art include invasive catheter measurement method, Korotkoff auscultation method, oscillography method and constant volume method. These methods are limited by the use scene, and at the same time, they will bring discomfort to the user. Therefore, it is urgent to develop a continuous blood pressure estimation technology which is expected to be used in portable devices.

[0047] The rapid development of modern medicine, optoelectronics, deep neural networks and intelligent devices has paved the way for photoplethysmography technology, opening up its application prospects for continuous monitoring in various daily life scenarios. The photoelectric sensing technology is a non-invasive biological measurement method that detects the pulse wave fluctuation in the skin blood vessels through a photoelectric sensor. The photoelectric sensing technology monitors the blood flow dynamic changes of human tissues by means of the light absorption characteristics, and is commonly used to measure physiological parameters such as heart rate, blood oxygen saturation and blood pressure, and has the advantages of simple operation, low cost and non-invasive, and is therefore widely used in health management, medical diagnosis and other fields.

[0048] The related art uses a photoplethysmography method for measuring blood pressure, which includes a pulse transit time-based method and a machine learning-based method. The pulse transit time-based method is more intuitive in theory and calculates the blood vessel pressure based on the blood flow dynamics model. However, this method does not take into account the changes in the state of the human blood vessels. The research on the machine learning-based method mainly focuses on the representation and extraction of waveform features, and there is still a lack of a unified and convincing guideline for the feature extraction system. The related art photoplethysmography method for measuring blood pressure has low accuracy in blood pressure detection.

[0049] To overcome the above-mentioned defects, the present application provides a non-invasive blood pressure detection system, method, electronic device and storage medium, which can improve the accuracy of blood pressure detection.

[0050] The following will be combined Figures 1-3 The non-invasive blood pressure detection system provided by the present application is described.

[0051] Figure 1 is one of the structure schematic diagrams of the non-invasive blood pressure detection system 100 provided by the present application, as Figure 1 shown, the system includes a photoelectric signal acquisition module 110, an analog signal processing module 120, a digital signal processing module 130 and a blood pressure calculation module 140, wherein:

[0052] The photoelectric signal acquisition module 110 is used to acquire the photoplethysmography of the specific tissue of the human body;

[0053] Optionally, the specific tissue of the human body can be the skin tissue of various parts of the human body, such as the wrist, or the finger, or the earlobe, etc., and the present application does not limit this.

[0054] Optionally, in actual application, the non-invasive blood pressure detection system provided by the present application can be placed on the specific tissue of the human body, and then the photoelectric signal acquisition module can acquire the photoplethysmography of the specific tissue of the human body.

[0055] Optionally, the advantage of the photoelectric plethysmogram technology lies in its non-invasiveness and real-time nature. Compared with traditional blood pressure measurement methods, the photoelectric plethysmogram technology does not require the use of inflatable cuffs, avoiding discomfort and blood flow obstruction, while having higher measurement frequency and more sensitive detection capability, which makes the technology widely used in clinical and scientific research fields. Through analysis and processing of the signal, the physiological state of the human body can be monitored and evaluated, providing important data support for personalized medicine, intelligent health and disease prevention.

[0056] Optionally, the photoelectric signal acquisition module can include a light source and a photosensitive array. The light source is an exit unit composed of one or more light-emitting elements with light intensity exceeding a preset light intensity, used to irradiate human skin tissue. The photosensitive array is a photosensitive unit composed of one or more photosensitive elements, used to receive light absorbed and reflected by the skin tissue and complete the conversion of optical signal to electrical signal.

[0057] Optionally, the light source can be three wavelengths of 525nm, 680nm and 930nm light-emitting diodes with light intensity of 1200nit. Visible light or different wavelength combinations can also be used.

[0058] Optionally, the photosensitive array can be two photoelectric receivers that sequentially acquire photoelectric plethysmogram signals of different wavelengths. CMOS sensors with different pixel sizes can also be used.

[0059] Optionally, a certain time, usually 15s, is required between the emission of light from the light source and the reception of light by the photosensitive array. Therefore, the photoelectric signal acquisition module should acquire the photoelectric plethysmogram for a time greater than this time to ensure that the blood pressure value is obtained.

[0060] Optionally, after the photoelectric signal acquisition module obtains the photoelectric plethysmogram, it can send the photoelectric plethysmogram to the analog signal processing module for processing.

[0061] The analog signal processing module 120 is used to amplify the effective information of the photoelectric plethysmogram and filter out analog high-frequency noise to obtain a first signal.

[0062] Optionally, the analog signal processing module can receive and process the photoelectric plethysmogram sent by the photoelectric signal acquisition module to amplify the effective information of the photoelectric plethysmogram and filter out analog high-frequency noise to obtain a first signal.

[0063] Optionally, the analog signal processing module can include a signal amplifier and an analog low-pass filter. The signal amplifier is used to amplify the effective information of the photoelectric plethysmogram, and the analog low-pass filter is used to filter out analog high-frequency noise from the photoelectric plethysmogram.

[0064] Optionally, after obtaining the first signal, the analog signal processing module can send the first signal to the digital signal processing module for signal conversion, signal optimization and signal cleaning.

[0065] The digital signal processing module 130 is configured to convert the first signal into a digital signal to obtain a second signal, and to perform signal optimization and signal cleaning on the second signal to obtain a third signal.

[0066] Optionally, the digital signal processing module can receive the first signal sent by the analog signal processing module, and then perform signal conversion, i.e., convert the first signal into a digital signal.

[0067] Optionally, the digital signal processing module can include an analog-to-digital converter for converting the first signal into a digital signal to obtain a second signal.

[0068] Optionally, after obtaining the second signal, the digital signal processing module can perform signal processing, i.e., signal optimization and signal cleaning, on the second signal to obtain a processed third signal.

[0069] Optionally, the digital signal processing module can include a digital signal preprocessing algorithm for signal optimization and a signal quality evaluation algorithm for signal cleaning.

[0070] Optionally, after obtaining the third signal, the digital signal processing module can send the third signal to the blood pressure calculation module for obtaining a blood pressure value.

[0071] The blood pressure calculation module 140 is configured to use a 2DBP_Net dimensional transformation convolution network to obtain a blood pressure value based on the third signal.

[0072] Optionally, the blood pressure calculation module can receive the third signal sent by the digital signal processing module, and use a 2DBP_Net dimensional transformation convolution network to obtain a blood pressure value based on the third signal.

[0073] Optionally, the blood pressure calculation module can include a processor for overall planning and issuing instructions to various modules, and a 2DBP_Net dimensional transformation convolution network for obtaining a blood pressure value.

[0074] Optionally, the 2DBP_Net dimensional transformation convolution network discards the traditional idea of two models predicting two target values, adopts a method of synchronously outputting systolic and diastolic pressure prediction values, and realizes parameter sharing by combining the correlation of the two task targets.

[0075] Optionally, the non-invasive blood pressure detection system can further comprise a storage module and a display module, the storage module being connected with the digital signal processing module and the blood pressure calculation module, for storing the photoplethysmogram and the blood pressure value in a certain period, and the display module being connected with the blood pressure calculation module, for displaying the blood pressure measurement result.

[0076] Optionally, the storage module can save the photoplethysmogram and the blood pressure measurement value of a user for one month, and update in real time.

[0077] Optionally, the display module can use an LED screen, and can display the personal information of the user, the blood pressure measurement value and the blood pressure history.

[0078] Figure 2 Figure 2 is a structural schematic diagram of the non-invasive blood pressure detection system provided by the application. Figure 2 As shown in the figure, in an embodiment of the application, the non-invasive blood pressure detection system comprises 1-a photoplethysmogram signal acquisition module, 2-an analog signal processing module, 3-a digital signal processing module, 31-a digital-to-analog converter, 32-a digital signal preprocessing algorithm, 33-a signal quality evaluation algorithm, 4-a blood pressure calculation module, 41-a processor, 42-a 2DBP_Net dimensional transformation convolution network, 5-a storage module and 6-a display module.

[0079] Optionally, in actual measurement, the user can place the non-invasive blood pressure detection system on the skin tissue, and the blood pressure detection system waits for the user to issue a measurement instruction; after the user issues the measurement instruction, the processor controls the photoplethysmogram signal acquisition module to acquire the photoplethysmogram signal, and simultaneously detects the adhesion of the blood pressure detection system and the skin tissue; the processor controls the analog signal processing module and the digital signal processing module to perform signal optimization, signal evaluation and signal windowing on the waveform signal, and saves the waveform signal to the storage module; the blood pressure calculation module calculates the blood pressure by calling the 2DBP_Net dimensional transformation convolution network, and the display module displays the measurement result and saves the blood pressure value to the storage module.

[0080] The non-invasive blood pressure detection system provided by the application uses the photoplethysmogram signal acquisition module to acquire the photoplethysmogram of a specific tissue of a human body, and performs analog signal processing and digital signal processing thereon, and finally performs blood pressure detection through the 2DBP_Net dimensional transformation convolution network of the blood pressure calculation module, so that the accuracy of the algorithm blood pressure measurement can be improved.

[0081] Optionally, the 2DBP_Net dimensional transformation convolution network comprises an encoding module, a two-dimensional convolution module and a linear connection module, the two-dimensional convolution module comprises a frequency domain transformation method and a two-dimensional convolution network, wherein:

[0082] The encoding module is configured to add amplitude information and position information in a fourth signal to obtain a fifth signal, wherein the fourth signal comprises the third signal, a first derivative signal of the third signal, and a second derivative signal of the third signal.

[0083] The frequency domain transformation method is configured to transform the fifth signal from a time domain to a frequency domain, determine a plurality of main constituent frequencies of the fifth signal based on a frequency domain amplitude, and convert the plurality of main constituent frequencies into a slicing period, slice the fifth signal according to the slicing period, and stack the sliced fifth signal in sequence into a two-dimensional signal.

[0084] The two-dimensional convolution network is configured to extract intra-period dependence and extra-period dependence of the two-dimensional signal, and perform one-dimensional splicing and weighted averaging on the two-dimensional signal according to a frequency domain amplitude to obtain a one-dimensional signal.

[0085] The linear connection module is configured to obtain a blood pressure value based on the one-dimensional signal.

[0086] Optionally, after the blood pressure calculation module obtains the third signal, the first derivative signal and the second derivative signal of the third signal can be calculated as algorithm inputs, which are used to guide the model to extract blood pressure-related feature information such as blood vessel elasticity and blood viscosity.

[0087] Specifically, the first derivative signal and the second derivative signal of the third signal can be calculated by the following formulas:

[0088] VPG(n)=PPG(n+1)-PPG(n)

[0089] APG(n)=PPG(n+1)-2PPG(n)+PPG(n-1)

[0090] wherein VPG(n) is the first derivative signal of the third signal, APG(n) is the second derivative signal of the third signal, and PPG(n) is the third signal at time n.

[0091] Figure 3 is a structural schematic diagram of a 2DBP_Net dimensional transformation convolution network provided by the present application, as Figure 3 shown, the 2DBP_Net dimensional transformation convolution network comprises an encoding module, a two-dimensional convolution module, and a linear connection module, wherein the two-dimensional convolution module comprises a frequency domain transformation method and a two-dimensional convolution network.

[0092] Optionally, the encoding module comprises amplitude encoding and position encoding. After the first derivative signal and the second derivative signal of the third signal are obtained, the third signal, the first derivative signal of the third signal, and the second derivative signal of the third signal can be collectively taken as a fourth signal, which is input into the encoding module to perform amplitude encoding and position encoding on the fourth signal, add amplitude information and position information in the fourth signal, and obtain a fifth signal.

[0093] Optionally, the amplitude encoding can extract the amplitude feature in the form of one-dimensional convolution.

[0094] Optionally, the position encoding can be cosine encoding for different time points, and the amplitude encoding and the position encoding can convert the one-dimensional time sequence signal into one-dimensional shallow features.

[0095] Optionally, the two-dimensional convolution module is composed of three dimensional transformation convolution blocks, and the convolution blocks are connected in residual connection. Except for the first dimensional transformation convolution block, the input of each convolution block is the output of the previous convolution block. Inside the dimensional transformation convolution block, a frequency domain transformation method and a two-dimensional convolution network are included. The frequency domain transformation method first performs Fast Fourier Transform (FFT) on the input signal, and takes out three main frequency components by comparing the amplitude in the frequency domain. The one-dimensional signal is divided into three two-dimensional signals according to the top-3 frequencies, and the divided two-dimensional period segments are stacked, each column containing all the value points in a time period, and each row containing the value points of the same stage in different time periods. The two-dimensional convolution network is used to extract the intra-period and inter-period dependencies of the two-dimensional signal, and perform one-dimensional splicing and weighted average on the two-dimensional signal according to the frequency domain amplitude, to obtain a one-dimensional signal. According to the different amplitude sizes corresponding to the k frequencies, the three one-dimensional feature combinations are adaptively fused.

[0096] Optionally, the backbone network of the dimensional transformation convolution module of the two-dimensional convolution module can adopt Resnet, VGG, etc., for two-dimensional feature extraction.

[0097] The non-invasive blood pressure detection system provided by the application uses the 2DBP_Net method combining the dimensional transformation strategy and the two-dimensional convolution neural network, greatly improves the blood pressure prediction ability of the model, and reduces the parameter quantity and floating point calculation quantity of the model by using the small network structure, thereby reducing the performance requirements of the model on the non-invasive blood pressure detection system. The use of the two-dimensional convolution neural network can improve the blood pressure feature extraction capability of the algorithm and improve the accuracy of the blood pressure measurement of the algorithm.

[0098] Optionally, the digital signal processing module is specifically used for:

[0099] The second signal is standardized, Butterworth filtered, wavelet transformed and slidingly windowed.

[0100] Optionally, in order to process the second signal, the second signal can be standardized, Butterworth filtered, wavelet transformed and slidingly windowed.

[0101] Optionally, the digital signal pre-processing algorithm in the digital signal processing module can use a Butterworth filter for low-pass filtering to filter out digital high-frequency noise above 7Hz.

[0102] Optionally, the digital signal preprocessing algorithm can use DB8 wavelet transform to remove baseline drift caused by interference such as motion and breathing.

[0103] Optionally, the digital signal preprocessing algorithm can use mean smoothing filtering to remove signal spikes, and divide the signal window into 15-second intervals.

[0104] The non-invasive blood pressure detection system provided by this invention performs standardization, Butterworth filtering, wavelet transform, and sliding windowing on the second signal to improve the accuracy of the final blood pressure value prediction.

[0105] Optionally, the digital signal processing module is specifically used for:

[0106] The Lonely Forest anomaly detection method was used to clean the second signal.

[0107] Optionally, the Lonely Forest anomaly detection method retains signal segments longer than the duration of the first detection signal for each user and divides them into windows for blood pressure prediction. These data, as a given dataset, can be represented as X = {x1, x2, ..., x...} n},in x i Let x be the feature vector of the i-th window. i = (f1,......,f8), where f1 to f8 represent different dimensions of characteristics such as kurtosis, skewness, mean, variance, maximum value, minimum value, zero-crossing rate, and entropy.

[0108] Optionally, the duration of the first detection signal can be 5 minutes, 10 minutes, or 15 minutes, and the present invention does not limit it.

[0109] Alternatively, the steps for establishing an evaluation model are as follows:

[0110] 1. Randomly select n samples from the training data as a subset, put them into the root node of an isolated tree, randomly specify a feature dimension, and generate a split point p within the data range of the current node;

[0111] 2. A hyperplane is formed by selecting this split point, which divides the current node data into two subspaces. Points with dimensions smaller than the split point p are placed in the left branch of the current node, and points with dimensions greater than or equal to p are placed in the right branch of the current node.

[0112] 3. Continue recursively performing steps 1 and 2 until it is impossible to continue splitting;

[0113] 4. After obtaining t isolated trees, data evaluation can be performed. For each window x... iPerform traversal calculations on t trees and combine the results for each tree:

[0114]

[0115] Where h(x) i ) is x i At the height of each tree, c(ψ) is the average path length for a given number of samples ψ, used to evaluate the sample x. i The path length is standardized.

[0116] Optionally, if window x i If the score is close to 1, the data is considered to be of poor quality and contains anomalies, and it will be removed.

[0117] Optionally, the anomaly detection score of a qualified signal sample should be less than a first signal quality threshold, which can be set according to actual conditions. In one embodiment of the present invention, it is set to 0.5.

[0118] The non-invasive blood pressure detection system provided by this invention uses the Lonely Forest anomaly detection method to clean the second signal, thereby improving the accuracy of the final blood pressure value prediction.

[0119] The non-invasive blood pressure detection method provided by the present invention will be described below. The non-invasive blood pressure detection method described below can be referred to in correspondence with the non-invasive blood pressure detection system described above.

[0120] Figure 4 This is a schematic flowchart of the non-invasive blood pressure detection method provided by the present invention, as shown below. Figure 4 As shown, the method includes the following steps:

[0121] Step 400: Collect photoplethysmography (PPG) waves from specific tissues in the human body;

[0122] Optionally, the specific human tissue can be skin tissue from various parts of the human body, such as the wrist, fingers, or earlobe, etc., and the present invention does not limit it.

[0123] Optionally, in practical applications, the non-invasive blood pressure detection system provided by this invention can be placed on specific tissues of the human body, and then the photoelectric signal acquisition module can acquire the photoplethysmography (PPG) wave of the specific tissues of the human body.

[0124] Optionally, the advantage of the photoelectric plethysmogram technology lies in its non-invasiveness and real-time nature. Compared with traditional blood pressure measurement methods, the photoelectric plethysmogram technology does not require the use of inflatable cuffs, avoiding discomfort and blood flow obstruction, while having a higher measurement frequency and more sensitive detection capability, which makes the technology widely used in clinical and scientific research fields. Through the analysis and processing of the signal, the physiological state of the human body can be monitored and evaluated, providing important data support for personalized medicine, intelligent health and disease prevention.

[0125] Optionally, the photoelectric signal acquisition module can include a light source and a photosensitive array. The light source is an exit unit composed of one or more light-emitting elements with light intensity exceeding a preset light intensity, used to irradiate human skin tissue. The photosensitive array is a photosensitive unit composed of one or more photosensitive elements, used to receive light absorbed and reflected by the skin tissue and complete the conversion of optical signal to electrical signal.

[0126] Optionally, the light source can be three wavelengths of 525nm, 680nm and 930nm light-emitting diodes with a light intensity of 1200nit. Visible light or different wavelength combinations can also be used.

[0127] Optionally, the photosensitive array can be two photoelectric receivers that sequentially acquire photoelectric plethysmogram signals of different wavelengths. CMOS sensors of different pixel sizes can also be used.

[0128] Optionally, a certain time, generally 15s, is required between the emission of light from the light source and the reception of light by the photosensitive array. Therefore, the time for the photoelectric signal acquisition module to acquire the photoelectric plethysmogram should be greater than this time to ensure that the blood pressure value is obtained.

[0129] Optionally, after the photoelectric signal acquisition module obtains the photoelectric plethysmogram, the photoelectric plethysmogram can be sent to the analog signal processing module for processing.

[0130] Step 410: amplifying the effective information of the photoelectric plethysmogram and filtering out analog high-frequency noise to obtain a first signal.

[0131] Optionally, the analog signal processing module of the non-invasive blood pressure detection system provided by the present application can receive and process the photoelectric plethysmogram sent by the photoelectric signal acquisition module to amplify the effective information of the photoelectric plethysmogram and filter out analog high-frequency noise to obtain a first signal.

[0132] Optionally, the analog signal processing module can include a signal amplifier and an analog low-pass filter. The signal amplifier is used to amplify the effective information of the photoelectric plethysmogram, and the analog low-pass filter is used to filter out analog high-frequency noise from the photoelectric plethysmogram.

[0133] Optionally, after obtaining the first signal, the analog signal processing module can send the first signal to the digital signal processing module for signal conversion, signal optimization and signal cleaning.

[0134] Step 420, converting the first signal into a digital signal to obtain a second signal;

[0135] Optionally, the digital signal processing module can receive the first signal sent by the analog signal processing module, and then perform signal conversion, i.e., converting the first signal into a digital signal.

[0136] Optionally, the digital signal processing module can include an analog-to-digital converter for converting the first signal into a digital signal to obtain a second signal.

[0137] Step 430, signal optimization and signal cleaning are performed on the second signal to obtain a third signal;

[0138] Optionally, after obtaining the second signal, the digital signal processing module can perform signal processing, i.e., signal optimization and signal cleaning, on the second signal to obtain a processed third signal.

[0139] Optionally, the digital signal processing module can include a digital signal preprocessing algorithm for signal optimization of the second signal and a signal quality evaluation algorithm for signal cleaning of the second signal.

[0140] Optionally, after obtaining the third signal, the digital signal processing module can send the third signal to the blood pressure calculation module for obtaining a blood pressure value.

[0141] Step 440, based on the third signal, a 2DBP_Net dimensional transformation convolution network is used to calculate a blood pressure value.

[0142] Optionally, the blood pressure calculation module can include a processor for overall planning and issuing instructions to each module, and a 2DBP_Net dimensional transformation convolution network for obtaining a blood pressure value.

[0143] Optionally, the 2DBP_Net dimensional transformation convolution network discards the traditional idea of two models predicting two target values respectively, adopts a method of synchronously outputting systolic and diastolic pressure prediction values, and realizes parameter sharing by combining the correlation of two task targets.

[0144] Optionally, during actual measurement, the user can place the non-invasive blood pressure detection system on the skin tissue, and the blood pressure detection system waits for the user to issue a measurement instruction; after the user issues the measurement instruction, the processor controls the photoelectric signal acquisition module to acquire the photoplethysmographic pulse wave signal, and simultaneously detects the fit degree of the blood pressure detection system and the skin tissue; the processor controls the analog signal processing module and the digital signal processing module to perform signal optimization, signal evaluation and signal windowing on the waveform signal, and saves the waveform signal to the storage module; the blood pressure calculation module calculates the blood pressure by calling the 2DBP_Net dimensional transformation convolution network, and the display module displays the measurement result and saves the blood pressure value to the storage module.

[0145] The non-invasive blood pressure detection method provided by the application uses a photoelectric signal acquisition module to acquire the photoplethysmographic pulse wave of a specific tissue of a human body, and performs analog signal processing and digital signal processing thereon, and finally performs blood pressure detection through the 2DBP_Net dimensional transformation convolution network of the blood pressure calculation module, which can improve the accuracy of the algorithm blood pressure measurement.

[0146] Optionally, the blood pressure value calculated by using the 2DBP_Net dimensional transformation convolution network comprises:

[0147] In the fourth signal, the amplitude information and the position information are added to obtain a fifth signal, and the fourth signal comprises the third signal, a first derivative signal of the third signal and a second derivative signal of the third signal;

[0148] The fifth signal is transformed from the time domain to the frequency domain, and a plurality of main constituent frequencies of the fifth signal are determined based on the frequency domain amplitude;

[0149] The plurality of main constituent frequencies are converted into segmentation periods, the fifth signal is segmented according to the segmentation periods and stacked in order into a two-dimensional signal;

[0150] The intra-period dependence and the extra-period dependence of the two-dimensional signal are extracted, and the two-dimensional signal is one-dimensionally spliced and weightedly averaged according to the frequency domain amplitude to obtain a one-dimensional signal;

[0151] Based on the one-dimensional signal, a blood pressure value is obtained.

[0152] Optionally, after the blood pressure calculation module obtains the third signal, the first derivative signal and the second derivative signal of the third signal can be calculated first as algorithm inputs, which are used to guide the model to extract blood pressure related characteristic information such as blood vessel elasticity and blood viscosity.

[0153] Specifically, the first derivative signal and the second derivative signal of the third signal can be calculated by the following formula:

[0154] VPG(n)=PPG(n+1)-PPG(n)

[0155] APG(n) = PPG(n+1) - 2PPG(n) + PPG(n-1)

[0156] wherein VPG(n) is a first derivative signal of the third signal, APG(n) is a second derivative signal of the third signal, and PPG(n) is the third signal at time n.

[0157] Figure 3 is a structural schematic diagram of the 2DBP_Net dimension transformation convolutional network provided by the application, as Figure 3 shown, the 2DBP_Net dimension transformation convolutional network comprises an encoding module, a two-dimensional convolution module, and a linear connection module, the two-dimensional convolution module comprises a frequency domain transformation method and a two-dimensional convolutional network.

[0158] Optionally, the encoding module comprises amplitude encoding and position encoding, after obtaining the first derivative signal and the second derivative signal of the third signal, the third signal, the first derivative signal of the third signal, and the second derivative signal of the third signal can be collectively taken as a fourth signal, input into the encoding module to perform amplitude encoding and position encoding on the fourth signal, add amplitude information and position information in the fourth signal, and obtain a fifth signal.

[0159] Optionally, the amplitude encoding can extract amplitude features in a one-dimensional convolution manner.

[0160] Optionally, the position encoding can perform cosine encoding on different time points, and the amplitude encoding and the position encoding can convert a one-dimensional time sequence signal into a one-dimensional shallow feature.

[0161] Optionally, the two-dimensional convolution module is composed of 3 dimension transformation convolutional blocks, and residual connection is performed between the convolutional blocks. Except for the first dimension transformation convolutional block, the input of each convolutional block is the output of the previous convolutional block. Inside the dimension transformation convolutional block, a frequency domain transformation method and a two-dimensional convolutional network are included. The frequency domain transformation method first performs Fast Fourier Transform (FFT) on the input signal, and takes out 3 main constituent frequencies by comparing the amplitude size in the frequency domain. The one-dimensional signal is divided into 3 two-dimensional signals according to the top-3 frequencies, the divided two-dimensional periodic segments are stacked, each column contains all the value points in a time period, and each row contains the value points of the same stage in different time periods. The two-dimensional convolutional network is used to extract the intra-period dependence and inter-period dependence of the two-dimensional signal, and perform one-dimensional splicing and weighted average on the two-dimensional signal according to the frequency domain amplitude, to obtain a one-dimensional signal. According to the different amplitude sizes corresponding to the k frequencies, adaptive fusion is performed on the 3 one-dimensional feature combinations.

[0162] Optionally, the backbone network of the dimension transformation convolutional module of the two-dimensional convolution module can adopt Resnet, VGG, etc., for two-dimensional feature extraction.

[0163] The non-invasive blood pressure detection method provided by this invention uses the 2DBP_Net method, which combines a dimensionality transformation strategy and a two-dimensional convolutional neural network, to greatly improve the model's blood pressure prediction capability. The miniaturized network structure reduces the number of model parameters and floating-point calculations, thereby reducing the performance requirements of the model on the non-invasive blood pressure detection system. The use of a two-dimensional convolutional neural network can improve the algorithm's blood pressure feature extraction capability and improve the accuracy of the algorithm's blood pressure measurement.

[0164] Optionally, the signal optimization of the second signal includes:

[0165] The second signal is then standardized, Butterworth filtered, wavelet transformed, and subjected to sliding windowing.

[0166] Alternatively, in order to process the second signal, it can be standardized, Butterworth filtered, wavelet transformed, and windowed.

[0167] Optionally, a Butterworth filter can be used for low-pass filtering via a digital signal preprocessing algorithm in the digital signal processing module to filter out digital high-frequency noise above 7Hz.

[0168] Optionally, DB8 wavelet transform can be used to remove baseline drift caused by interference such as motion and breathing.

[0169] Optionally, mean smoothing filtering can be used to remove signal glitches, with the signal window divided into 15-second intervals.

[0170] The non-invasive blood pressure detection method provided by this invention performs standardization, Butterworth filtering, wavelet transform, and sliding windowing on the second signal to improve the accuracy of the final blood pressure value prediction.

[0171] Optionally, the signal cleaning of the second signal includes:

[0172] The Lonely Forest anomaly detection method was used to clean the second signal.

[0173] Optionally, the Lonely Forest anomaly detection method retains signal segments longer than the duration of the first detection signal for each user and divides them into windows for blood pressure prediction. These data, as a given dataset, can be represented as X = {x1, x2, ..., x...} n},in x i Let x be the feature vector of the i-th window. i = (f1, ..., f8), where f1 to f8 represent different dimensions of characteristics such as kurtosis, skewness, mean, variance, maximum value, minimum value, zero-crossing rate, and entropy.

[0174] Optionally, the first detection signal duration can be 5 minutes, or 10 minutes, or 15 minutes, and the present application does not limit this.

[0175] Optionally, the step of establishing an evaluation model is as follows:

[0176] 1. Randomly select n samples from the training data as a sub-sample set, put them into the root node of an isolated tree, randomly assign a feature dimension, and generate a split point p in the data range of the current node;

[0177] 2. The selection of this split point forms a hyperplane, which divides the current node data into two subspaces, and the points less than the split point p in the current selected feature dimension are put into the left branch of the current node, and the points greater than or equal to p are put into the right branch of the current node;

[0178] 3. Continue to recursively perform steps 1 and 2 until it cannot continue to split;

[0179] 4. After obtaining t isolated trees, data evaluation can be performed, for each window x i , t tree traversal calculations are performed, and the results of each tree are integrated:

[0180]

[0181] Where h(x i ) is the height of x i in each tree, and c(ψ) is the average path length for a given sample number ψ, which is used to normalize the path length of the sample x i .

[0182] Optionally, if the window x i score is close to 1, it is considered that the data quality is poor and there is an anomaly, and it is removed.

[0183] Optionally, the qualified signal sample anomaly detection score should be less than the first signal quality threshold, which can be set according to the actual situation, and in an embodiment of the present application, it is set to 0.5.

[0184] The non-invasive blood pressure detection system provided by the present application uses the isolated forest anomaly detection method to perform signal cleaning on the second signal, so as to improve the final blood pressure value prediction accuracy.

[0185] Figure 5 An example of an electronic device entity structure schematic diagram is shown as Figure 5As shown, the electronic device can include a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 complete mutual communication through the communications bus 540. The processor 510 can invoke a logical instruction in the memory 530 to execute a non-invasive blood pressure detection method, which includes: collecting a photoelectric plethysmogram of a specific tissue of a human body; performing effective information amplification and filtering out analog high-frequency noise on the photoelectric plethysmogram to obtain a first signal; converting the first signal into a digital signal to obtain a second signal; performing signal optimization and signal cleaning on the second signal to obtain a third signal; and calculating a blood pressure value based on the third signal using a 2DBP_Net dimensional transformation convolution network.

[0186] In addition, the logical instruction in the memory 530 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0187] On the other hand, the present application also provides a computer program product, which includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the non-invasive blood pressure detection method provided by the above-mentioned methods, which includes: collecting a photoelectric plethysmogram of a specific tissue of a human body; performing effective information amplification and filtering out analog high-frequency noise on the photoelectric plethysmogram to obtain a first signal; converting the first signal into a digital signal to obtain a second signal; performing signal optimization and signal cleaning on the second signal to obtain a third signal; and calculating a blood pressure value based on the third signal using a 2DBP_Net dimensional transformation convolution network.

[0188] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a non-invasive blood pressure detection method provided by any of the above methods, the method comprising: collecting a photoelectric plethysmogram of a specific tissue of a human body; performing effective information amplification and filtering of analog high-frequency noise on the photoelectric plethysmogram to obtain a first signal; converting the first signal into a digital signal to obtain a second signal; performing signal optimization and signal cleaning on the second signal to obtain a third signal; and calculating a blood pressure value based on the third signal using a 2DBP_Net dimensional transformation convolution network.

[0189] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0190] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in terms of contribution to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0191] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A non-invasive blood pressure monitoring system, characterized in that, It includes a photoelectric signal acquisition module, an analog signal processing module, a digital signal processing module, and a blood pressure calculation module, among which: The photoelectric signal acquisition module is used to acquire photoplethysmography (PPG) waves of specific human tissues. The analog signal processing module is used to effectively amplify the information of the photoplethysmography pulse wave and filter out analog high-frequency noise to obtain a first signal; The digital signal processing module is used to convert the first signal into a digital signal to obtain a second signal, and is also used to optimize and clean the second signal to obtain a third signal. The blood pressure calculation module is used to obtain the blood pressure value based on the third signal using a 2DBP_Net dimensionality-transformed convolutional network. The 2DBP_Net dimensionality-transformed convolutional network includes an encoding module, a two-dimensional convolution module, and a linear connection module. The two-dimensional convolution module includes a frequency domain transformation method and a two-dimensional convolutional network, wherein: The encoding module is used to add amplitude information and position information to the fourth signal to obtain the fifth signal. The fourth signal includes the third signal, the first derivative signal of the third signal, and the second derivative signal of the third signal. The frequency domain transformation method is used to transform the fifth signal from the time domain to the frequency domain, and to determine multiple main constituent frequencies of the fifth signal based on the frequency domain amplitude. It is also used to convert the multiple main constituent frequencies into a segmentation period, segment the fifth signal according to the segmentation period, and stack them into a two-dimensional signal in sequence. The two-dimensional convolutional network is used to extract the intra-period dependency and out-period dependency of the two-dimensional signal, and to perform one-dimensional splicing and weighted averaging of the two-dimensional signal according to the frequency domain amplitude to obtain a one-dimensional signal. The linear connection module is used to obtain blood pressure values ​​based on the one-dimensional signal.

2. The non-invasive blood pressure monitoring system according to claim 1, characterized in that, The digital signal processing module is specifically used for: The second signal is then standardized, Butterworth filtered, wavelet transformed, and subjected to sliding windowing.

3. The non-invasive blood pressure monitoring system according to claim 1, characterized in that, The digital signal processing module is specifically used for: The Lonely Forest anomaly detection method was used to clean the second signal.

4. A non-invasive blood pressure detection method, characterized in that, include: Collect photoplethysmography (PPG) waves from specific tissues in the human body; The photoplethysmography pulse wave is effectively amplified and analog high-frequency noise is filtered out to obtain a first signal; The first signal is converted into a digital signal to obtain the second signal; The second signal is optimized and cleaned to obtain the third signal; Based on the third signal, the blood pressure value is calculated using a 2DBP_Net dimension-transformed convolutional network. The blood pressure value calculated using the 2DBP_Net dimension-transformed convolutional network includes: Amplitude and position information are added to the fourth signal to obtain the fifth signal, wherein the fourth signal includes the third signal, the first derivative signal of the third signal, and the second derivative signal of the third signal; The fifth signal is transformed from the time domain to the frequency domain, and several main constituent frequencies of the fifth signal are determined based on the frequency domain amplitude. The multiple main constituent frequencies are converted into segmentation periods, and the fifth signal is segmented according to the segmentation periods and stacked in sequence into a two-dimensional signal; Extract the intra-period and out-of-period dependencies of the two-dimensional signal, and perform one-dimensional splicing and weighted averaging on the two-dimensional signal according to the frequency domain amplitude to obtain a one-dimensional signal; Based on the one-dimensional signal, the blood pressure value is obtained.

5. The non-invasive blood pressure detection method according to claim 4, characterized in that, The signal optimization of the second signal includes: The second signal is then standardized, Butterworth filtered, wavelet transformed, and subjected to sliding windowing.

6. The non-invasive blood pressure detection method according to claim 4, characterized in that, The signal cleaning of the second signal includes: The Lonely Forest anomaly detection method was used to clean the second signal.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the non-invasive blood pressure detection method as described in any one of claims 4 to 6.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the non-invasive blood pressure detection method as described in any one of claims 4 to 6.

Images