CN115770027A - Blood pressure detection method and device based on artificial intelligence - Google Patents
Blood pressure detection method and device based on artificial intelligence Download PDFInfo
- Publication number
- CN115770027A CN115770027A CN202211594244.8A CN202211594244A CN115770027A CN 115770027 A CN115770027 A CN 115770027A CN 202211594244 A CN202211594244 A CN 202211594244A CN 115770027 A CN115770027 A CN 115770027A
- Authority
- CN
- China
- Prior art keywords
- blood pressure
- artificial intelligence
- dimensional
- layer
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
Abstract
The invention belongs to the field of medical instruments, relates to a blood pressure detection technology, and particularly relates to a blood pressure detection method and a blood pressure detection device based on artificial intelligence, wherein the method comprises the following steps of obtaining an original photoplethysmographic signal; performing band-pass filtering smoothing on the pulse wave signals; performing wavelet transformation on the smoothed signal, and converting the one-dimensional PPG signal segment into a two-dimensional time-frequency chromatogram; and inputting the two-dimensional time-frequency chromatogram into a blood pressure prediction model for prediction to obtain the blood pressure. The device comprises the following modules: the device comprises a photoplethysmography module, an operation processing module for operating a blood pressure prediction model, a storage module for storing acquired waveform data and storing the blood pressure prediction model, and a waveform and result display module.
Description
Technical Field
The invention belongs to the field of medical instruments, relates to a blood pressure detection technology, and particularly relates to a blood pressure detection method and device based on artificial intelligence.
Background
Blood Pressure (BP) is the pressure of blood flowing in a blood vessel acting on the blood vessel wall per unit area, and is the driving force for driving blood to flow in the blood vessel. In different vessels are called arterial blood pressure, capillary blood pressure and venous blood pressure, respectively, and the blood pressure is commonly referred to as arterial blood pressure of the systemic circulation. The blood pressure can judge the heart function and the peripheral vascular resistance, and is also an important content for diagnosing diseases, observing the change of illness states and judging the treatment effect.
The current non-invasive blood pressure measuring method mainly comprises a Korotkoff sound auscultation method and an oscillography method, wherein the Korotkoff sound auscultation method is a method for measuring blood pressure by Korotkoff sound, the cuff is firstly inflated by an inflation ball in the working process, and when the pressure in the cuff exceeds the systolic pressure of an artery, the artery is closed and the blood flow is not communicated. Then the needle valve is opened to enable the pressure in the cuff to be slowly deflated at the speed of 2-3 mmHg/s, when the systolic pressure is higher than the pressure in the cuff, part of arteries are opened, blood is sprayed to form vortex or turbulent flow, and the blood is vibrated and transmitted to the body surface to be Korotkoff sounds. The method is a qualified blood pressure measuring and measuring method which is only approved by international medicine, and the auscultatory method is a scientific and classical blood pressure measuring and measuring instrument which can not be replaced so far. Most of the electronic sphygmomanometers sold in the market adopt an oscillometric method, the blood pressure is predicted by utilizing the oscillation wave generated in the cuff air bag when the pressure is reduced, and when the amplitude of the oscillation wave is maximum, the pressure of the air sleeve is the average pressure of the artery. The systolic pressure of the artery corresponds to the first inflection point of the amplitude envelope and the diastolic pressure corresponds to the second inflection point of the envelope. The instrument is simple and convenient to operate and is suitable for families.
However, the auscultation method is relatively complex in operation, and although the existing full-automatic auscultation method sphygmomanometer is available, most of the auscultation methods still need manual inflation, deflation and reading, and a cuff is used for pressurizing the brachial artery, so that a measurer feels uncomfortable, and the product is large in size and not easy to carry; the oscillometric electronic sphygmomanometer is complex in shock wave source, different in human individuals, and has a generally applicable conversion relation of calculating systolic pressure and diastolic pressure from shock wave amplitude, certain errors exist, so that the oscillometric electronic sphygmomanometer is not suitable for medical treatment, and the oscillometric electronic sphygmomanometer is complicated in operation because a cuff is required to be used for pressurization.
Obtaining blood pressure parameter information based on volume pulse wave is a focus and focus of research in recent years. Based on an LED light source and a detector, measuring the attenuated light reflected and absorbed by blood vessels and tissues of a human body, tracing the change of the volume of the blood vessels in a cardiac cycle, and when the heart contracts, the peripheral blood volume is the largest, and the maximum light absorption amount and the minimum detected light intensity are the smallest; in diastole, on the contrary, the detected light intensity is the maximum, so the light intensity received by the light receiver is in pulsatile change, therefore, the volume pulse blood flow contains a plurality of important physiological information of cardiovascular systems such as blood flow, and the like, and is an important information source for researching human body circulatory systems, including information of blood pressure, heart rate and the like. The patent with the application number of 201410461910X discloses a noninvasive continuous blood pressure beat-by-beat measuring device, which utilizes the pulse wave conduction time and establishes a blood pressure detection model of a measured object based on standard blood pressure measurement data, thereby realizing noninvasive continuous blood pressure monitoring; the patent with the application number 2014101634254 is a blood pressure measuring device based on pulse wave characteristic parameters, but at present, the unified standards of pulse wave velocity and the correlation between pulse wave conduction time and arterial pulse pressure are still lacking, and the factors influencing the pulse wave velocity are more, so that the influence of factors such as vascular elasticity and the like on the blood pressure wave velocity needs to be considered, and a reliable and accurate algorithm needs to be developed and a more appropriate regression model needs to be established.
Disclosure of Invention
The invention aims to provide a blood pressure detection method based on artificial intelligence, which is high in measurement precision and reliable in result, and aims to solve the technical problems that the existing blood pressure measurement scheme is complicated and is not suitable for continuous multiple measurements, and the existing blood pressure measurement device is large in size and inconvenient to carry.
The technical scheme for achieving the purpose comprises the following contents.
A blood pressure detection method based on artificial intelligence comprises the following steps,
acquiring an original photoplethysmography signal;
performing band-pass filtering smoothing processing on the pulse wave signals;
performing wavelet transformation on the smoothed signals, and converting the one-dimensional PPG signal segment into a two-dimensional time-frequency spectrum;
and inputting the two-dimensional time-frequency chromatogram into a blood pressure prediction model for prediction to obtain the blood pressure.
Further, the blood pressure prediction model comprises a CNN-LSTM network model which is formed by mixing and building a precursor two-dimensional CNN network and an LSTM network, the subsequent LSTM network integrates and summarizes sequential characteristic quantities of deep matrixes output by the former two-dimensional CNN, and related time sequence information is obtained through a circulating link structure in the LSTM network, so that the classification of the model is assisted.
Further, the 1 st to 13 th layers of the CNN-LSTM network model are constructed by mutually crossing and mixing convolution layers and maximum pooling layers, the 14 th to 15 th layers are respectively an LSTM layer and a full-connection layer, a LeakyRelu activation function layer is arranged behind each two-dimensional convolution layer and each full-connection layer, and a BN layer is arranged in front of each maximum pooling layer. At the joint of the maximum pooling layer and the LSTM layer, the shape with the size of (none, 32, 128) is changed into (256, 192) by modifying the data dimension, before the full connection layer, the multidimensional data of the upper layer is subjected to one-dimensional operation by using a Flatten structure, the full connection layer performs probability calculation on the one-dimensional data of the upper layer, and finally the regression condition under the maximum probability is output.
The blood pressure detection method of the invention uses the strong deep learning characteristic extraction capability to extract the characteristics of the PPG signal, avoids the problem that the correlation between the pulse wave velocity and the pulse wave conduction time and the arterial pulse pressure is lack of unified standard, gives full play to the deep learning performance, converts the one-dimensional PPG signal segment into the two-dimensional time-frequency chromatogram by continuous wavelet transformation, can not only avoid the loss of information, but also fully display the time-frequency information in the original signal, and finally realizes the single-path photoplethysmography waveform blood pressure measurement, thereby improving the reliability of the detection result.
Based on the blood pressure detection method, the invention also provides a blood pressure detection device based on artificial intelligence, which comprises the following modules:
a photoelectric volume pulse wave acquisition module;
the operation processing module is used for performing operation processing on the acquired waveform data and operating the blood pressure prediction model;
the storage module is used for storing the acquired waveform data and storing the blood pressure prediction model;
the device comprises a waveform and result display module for displaying waveform data and measurement results, a photoelectric volume pulse wave acquisition module connected with an operation processing module, and an operation processing module connected with a storage module and the waveform and result display module.
Further, the photoplethysmography acquisition module comprises a dual-wavelength sensor and a signal conditioning circuit, and the signal conditioning circuit conditions signals obtained by the dual-wavelength sensor and transmits the signals to the operation processing module.
Further, the operation processing module adopts an embedded Linux board.
Further, the embedded Linux board is orange pyro 2.
The device has smaller volume, can be put into a pocket to carry, is convenient to measure, is suitable for family and community medical monitoring, and provides important reference physiological indexes for preventing early hypertension and hypotension; meanwhile, the blood pressure detection device is convenient to operate and quick in detection, the blood pressure can be predicted only by pressing a finger in a designated area, the psychological burden of a patient during diagnosis and detection is reduced, and the accuracy of detection is improved.
The invention applies the advanced artificial intelligence deep learning technology to the embedded system, so that the advanced theory can be moved out of a laboratory and into the life of people, the research becomes application, and the method becomes significant.
Drawings
FIG. 1 is a flow chart of the blood pressure detecting method based on artificial intelligence of the present invention;
FIG. 2 is a detailed architecture diagram of a 2D-CNN-LSTM network model;
FIG. 3 is an overall frame diagram of an artificial intelligence based blood pressure monitor according to an embodiment;
FIG. 4 is a comparison graph of an image before and after filtering;
FIG. 5 is a two-dimensional image output after continuous wavelet transform when the blood pressure monitor of the embodiment is in use;
FIG. 6 is a graph showing a comparison of the results of blood pressure measurements performed by the blood pressure measuring device of the example and a commercially available Ohlong sphygmomanometer.
Detailed Description
The blood pressure prediction method disclosed by the invention can be used for predicting blood pressure and measuring heart rate, and the invention is described in detail with reference to the embodiment.
Referring to fig. 1 to 5, the present embodiment provides a blood pressure detecting apparatus based on artificial intelligence, including a pulse wave collecting module 2, an operation processing module 6, an external large-capacity storage module 7, and a waveform and result display module 5, wherein the pulse wave collecting module 2 is connected to the operation processing module 6 to complete the collection of pulse wave signals, and the operation processing module 6 includes the pulse wave signal collection, band-pass filtering, continuous wavelet transformation and blood pressure prediction which are sequentially executed. The output end of the operation processing module 6 is respectively connected with the external large-capacity storage module 7 and the waveform and result display module 5, and the functions of storing and displaying the calculation result are respectively realized. The orange pyro 2 used by the operation processing module 6 is a small Linux development board, the chip adopts an ARM framework full log H616 chip, and the output end of the chip is respectively connected with an external large-capacity storage module and a waveform and result display module. The method is mainly used for collecting and processing conditioned signals and providing hardware support for the deep learning algorithm model.
The volume pulse wave acquisition module 2 uses MAX30102 sensor of Meixin company, which is a dual-wavelength sensor for obtaining original weak signals of photoelectric volume pulse waves at finger end of the living body object, and an internal integrated signal conditioning circuit, the output of which is connected with the operation processing module.
The external mass storage module 7 is used for temporarily storing data in the operation processing and permanently storing operation results, and a 32G TF card is used.
During detection, the method comprises the following steps:
(1) And pressing a power button to wait for the system to start.
(2) After the system is started, a start detection button is pressed, a finger 1 is placed on a shell 3, infrared light emitted by a pulse wave acquisition module 2 is reflected by a human finger 1 and then received by an infrared light receiving tube of the pulse wave acquisition module 2, and because blood in blood vessels of the human finger is regularly changed along with heartbeat, the change is captured by the infrared light receiving tube and converted into an electric signal, the electric signal is converted into a digital signal through a built-in AD after ambient light is eliminated, and the digital signal is stored in an internal FIFO.
(3) The infrared light data stored in the sensor FIFO is transmitted to the arithmetic processing module 6 through IIC communication.
(4) The operation processing module 6 stores the data read by the IIC into the large-capacity storage module 7, predicts the data by the deep learning model to obtain a result, and displays the result on the display module 5. The blood pressure prediction model is realized based on a 2D-CNN-LSTM network, and is mainly formed by hybrid construction of a precursor two-dimensional CNN network and a rear LSTM network, and the detailed architecture of the model is shown in figure 2. The subsequent LSTM network integrates and induces the deep matrix output by the earlier two-dimensional CNN to obtain sequential characteristic quantity, and obtains relevant time sequence information through a special cyclic chain structure in the LSTM network, thereby assisting the classification of the models. The 1 st to 13 th layers of the CNN-LSTM network model are mainly constructed by mutually crossing and mixing convolution layers and maximum pooling layers, the 14 th to 15 th layers are respectively an LSTM layer and a full-connection layer, wherein a LeakyRelu activation function layer is arranged behind each two-dimensional convolution layer and the full-connection layer, and the front of each maximum pooling layer comprises a BN layer. At the joint of the maximum pooling layer and the LSTM layer, the shape with the size of (none, 32, 128) is changed into (256, 192) by modifying the data dimension, before the full connection layer, the multidimensional data of the upper layer is subjected to one-dimensional operation by using a Flatten structure, the full connection layer carries out probability calculation on the one-dimensional data of the upper layer, and finally the regression condition under the maximum probability is output. The detailed architecture of the 2D-CNN-LSTM network model is shown in FIG. 2.
(5) The operation processing module 6 is a core processor of the whole system, is responsible for the control including the acquisition, storage and display of data, and more importantly, carries out deep learning algorithm model prediction on the acquired data, and the operation processing process mainly comprises the following steps:
reading the acquired pulse wave data and storing the pulse wave data in a large-capacity memory 7;
performing band-pass filtering on the pulse wave data read above, and removing high-frequency interference and low-frequency interference to make a waveform image smoother, wherein the images before and after filtering are shown in fig. 4;
after filtering, converting the one-dimensional PPG signal segment into a two-dimensional time-frequency chromatogram map through wavelet transformation, and storing the two-dimensional time-frequency chromatogram map into a large-capacity storage module 7, wherein a wavelet transformation image is shown in figure 5;
the operation processing module 6 inputs the stored wavelet-transformed image into the 2D-CNN-LSTM deep learning model to predict the blood pressure;
the data obtained above is displayed through the waveform and result display module 5, and is stored locally and in the cloud.
The built-in deep learning blood pressure prediction model of the blood pressure detection instrument of the embodiment uses the powerful feature extraction ability of deep learning to carry out feature extraction to the PPG signal, has avoided the problem that pulse wave velocity and pulse wave conduction time and artery pulse pressure correlation lack unified standard, full play deep learning performance to through continuous wavelet transform, convert one-dimensional PPG signal segment into two-dimensional time-frequency chromatogram, not only can avoid losing of information, can also fully demonstrate the time-frequency information in the original signal. Finally, the blood pressure is measured by the waveform of the single-path photoplethysmography pulse wave, continuous measurement can be realized, data can be stored, and the method is convenient to operate and use. The invention is light in weight, is based on embedded equipment, has small volume, low power consumption and portability, is provided with a lithium battery, does not need to be externally connected for power supply all the time, and also has the following advantages: the operation is easy, one-key measurement can be realized after the power switch is turned on, and the measurement can be repeated for multiple times, so that the operation is simple and convenient; the waveform, the heart rate and the blood pressure are displayed through the OLED screen, and the measurement result can be visually seen by a user; data storage, wherein the measurement result can be stored in an external large-capacity memory and can be synchronously stored in a cloud terminal, so that data reference is provided for the health management of a user; the collection mode is convenient, and blood pressure and heart rate information can be measured by only placing one finger at a specified position. The blood pressure detection instrument of the embodiment is used for carrying out blood pressure detection comparison with a commercially available Oldham sphygmomanometer, the Oldham sphygmomanometer is a cuff type electronic sphygmomanometer and is used for measuring by adopting an oscillometric principle, a cuff needs to be sleeved on an arm during measurement, two times of measurement need to be separated for two minutes, the steps are complicated, discomfort is brought by inflating the cuff and continuous measurement is not suitable, the detection result is shown in figure 6, the abscissa is the blood pressure value measured by the blood pressure detection instrument, the ordinate represents the blood pressure value measured by the Oldham sphygmomanometer, the better the correlation is, the closer the measurement result is, the better the correlation is, and the blood pressure detection instrument meets the AAMI standard as shown in figure 6.
Claims (7)
1. A blood pressure detection method based on artificial intelligence is characterized by comprising the following steps,
acquiring an original photoplethysmography signal;
performing band-pass filtering smoothing on the pulse wave signals;
performing wavelet transformation on the smoothed signals, and converting the one-dimensional PPG signal segment into a two-dimensional time-frequency spectrum;
and inputting the two-dimensional time-frequency chromatogram into a blood pressure prediction model for prediction to obtain the blood pressure.
2. The artificial intelligence based blood pressure detection method according to claim 1, wherein the blood pressure prediction model comprises a CNN-LSTM network model constructed by mixing a predecessor two-dimensional CNN network and a successor LSTM network, the successor LSTM network integrates and generalizes a deep matrix output by the predecessor two-dimensional CNN to obtain sequential characteristic quantities, and obtains relevant time sequence information through a cyclic chain structure in the LSTM network, thereby assisting the model in classification.
3. The artificial intelligence based blood pressure detection method according to claim 2, wherein the 1 st to 13 th layers of the CNN-LSTM network model are constructed by cross-blending convolution layers and max pooling layers, the 14 th to 15 th layers are LSTM layers and full connection layers respectively, wherein a LeakyRelu activation function layer is arranged behind each two-dimensional convolution layer and full connection layer, and a BN layer is arranged in front of each max pooling layer, the shape of (none, 32, 128) is changed to (256, 192) by modifying the data dimension at the connection of max pooling layer and LSTM layer, before the full connection layer, the multidimensional data of the upper layer is unidimensionally modified by using a fiatten structure, the full connection layer performs probability calculation on the one-dimensional data of the upper layer, and finally outputs the regression of the situation under the maximum probability.
4. A blood pressure detection device based on artificial intelligence is characterized by comprising the following modules:
a photoelectric volume pulse wave acquisition module;
the operation processing module is used for performing operation processing on the acquired waveform data and operating a blood pressure prediction model;
the storage module is used for storing the acquired waveform data and storing the blood pressure prediction model;
the device comprises a waveform and result display module for displaying waveform data and measurement results, a photoelectric volume pulse wave acquisition module connected with an operation processing module, and an operation processing module connected with a storage module and the waveform and result display module.
5. The artificial intelligence based blood pressure detecting device according to claim 4, wherein the photoplethysmography module comprises a dual wavelength sensor and a signal conditioning circuit, and the signal conditioning circuit conditions signals obtained by the dual wavelength sensor and transmits the conditioned signals to the operation processing module.
6. The artificial intelligence based blood pressure detecting device according to claim 4, wherein the operation processing module adopts an embedded Linux board card.
7. The artificial intelligence based blood pressure detecting device according to claim 6, wherein said embedded Linux board is orange zero2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211594244.8A CN115770027A (en) | 2022-12-13 | 2022-12-13 | Blood pressure detection method and device based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211594244.8A CN115770027A (en) | 2022-12-13 | 2022-12-13 | Blood pressure detection method and device based on artificial intelligence |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115770027A true CN115770027A (en) | 2023-03-10 |
Family
ID=85392097
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211594244.8A Pending CN115770027A (en) | 2022-12-13 | 2022-12-13 | Blood pressure detection method and device based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115770027A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116246786A (en) * | 2023-05-06 | 2023-06-09 | 太原理工大学 | Health data monitoring system and method based on LORA (local area network) ad hoc network communication |
| CN116982950A (en) * | 2023-06-26 | 2023-11-03 | 深圳先进技术研究院 | Cuff-free blood pressure monitoring system |
-
2022
- 2022-12-13 CN CN202211594244.8A patent/CN115770027A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116246786A (en) * | 2023-05-06 | 2023-06-09 | 太原理工大学 | Health data monitoring system and method based on LORA (local area network) ad hoc network communication |
| CN116982950A (en) * | 2023-06-26 | 2023-11-03 | 深圳先进技术研究院 | Cuff-free blood pressure monitoring system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ding et al. | Continuous blood pressure measurement from invasive to unobtrusive: Celebration of 200th birth anniversary of Carl Ludwig | |
| CN107233087A (en) | A kind of Woundless blood pressure measuring device based on photoplethysmographic feature | |
| CN115770027A (en) | Blood pressure detection method and device based on artificial intelligence | |
| KR20210005644A (en) | Method for estimating blood pressure and arterial stiffness based on light volumetric variability recording (PPG) signal | |
| CN108186000B (en) | Real-time blood pressure monitoring system and method based on ballistocardiogram signal and photoelectric signal | |
| CN109512409B (en) | Cardiovascular health assessment device and use method thereof | |
| RU2011119486A (en) | SYSTEM AND DEVICE FOR NON-INVASIVE ARTERIAL PRESSURE MEASUREMENT | |
| JP2006158974A (en) | Integral type physiologic signal evaluation apparatus | |
| US10694959B2 (en) | Image based blood pressure monitoring method | |
| CN106137161A (en) | A kind of heart rate and blood pressure measurement apparatus and method | |
| CN103536281A (en) | Vascular endothelial function detecting device based on fingertip temperature change | |
| CN1762300A (en) | Human body physiological parameter monitor device based on high frequency light capacity trace signal | |
| Taha et al. | A review on non-invasive hypertension monitoring system by using photoplethysmography method | |
| WO2018006501A1 (en) | Integrated circuit structure for continuous detection of human blood pressure | |
| Qiu et al. | A wireless wearable sensor patch for the real-time estimation of continuous beat-to-beat blood pressure | |
| US20070239039A1 (en) | Method and apparatus for measuring blood pressures by using blood oxygen concentration and electrocardiography | |
| CN109770887B (en) | Cuff type noninvasive hemodynamic artificial intelligence cloud monitoring system and method | |
| US20230293026A1 (en) | Photoplethysmography-based Blood Pressure Monitoring Device | |
| CN210095711U (en) | Noninvasive continuous blood pressure measuring equipment | |
| CN209360675U (en) | Wristwatch type blood pressure monitoring device based on pulse wave velocity method | |
| TWM473189U (en) | Simple measuring device for arteriosclerosis | |
| CN105310678B (en) | It is a kind of heart is calculated based on S pulse analysis method often to fight the detection method of blood volume | |
| CN215424574U (en) | Korotkoff sound electronic sphygmomanometer | |
| US11653847B2 (en) | Method and apparatus for hypertension classification | |
| CN211155750U (en) | Vital sign detection device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |