A non-contact blood pressure measurement method based on visible light vision

Abstract

The invention discloses a non-contact blood pressure measurement method based on visible light vision, which relates to the biomedical field. The present invention uses a multi-spectral camera to obtain pulse wave signals of multiple light bands, and uses the subtle differences between pulse waves of different bands and the correlation between multiple pulse waves to estimate blood pressure, thereby enhancing the independence of information carried in pulse wave signals corresponding to light of different wavelengths, and reducing information loss; and at the same time, using pulse wave signals corresponding to light of different wavelengths can suppress motion interference and environmental noise that occur during measurement, thereby improving measurement accuracy. The present invention only needs to collect one video signal to measure blood pressure. Compared with the method of simultaneously collecting PPG signals and ECG signals to measure blood pressure and simultaneously collecting PPG signals from multiple parts to measure blood pressure, this method significantly reduces the complexity of the blood pressure measurement operation.

Description

Technical Field

[0001] The invention relates to the field of biomedicine, and in particular to a non-contact blood pressure measurement system based on visible light vision. Background Art

[0002] Hypertension is the most common cardiovascular disease and the leading cause of premature death in the world. According to statistics from the World Health Organization, there are approximately 1.28 billion adults (30-79%) suffering from hypertension worldwide. Hypertension refers to high blood pressure in blood vessels. The diagnostic standard for hypertension in adults in China is ≥130 / 80 mmHg. Hypertension can cause symptoms such as headaches and blurred vision, and may also cause or aggravate other health problems such as heart disease and stroke. Hypertension is not easy to cure, so in addition to taking antihypertensive drugs, it is also necessary to monitor blood pressure changes in real time to understand blood pressure health status; in addition, some acute cardiovascular diseases will have a strong response in blood pressure, such as a sudden and significant increase in blood pressure. In summary, blood pressure monitoring is of great significance for the treatment of hypertension and the reduction of cardiovascular disease mortality, and it is necessary to achieve accurate and convenient blood pressure measurement.

[0003] Non-invasive cuff blood pressure measuring instruments obtain blood pressure information by applying pressure. The measurement results are accurate and widely used in various scenarios such as hospitals. However, the cuff will compress the patient's body tissues and cause pain during use, and the measuring instrument cannot achieve continuous monitoring. With the development of photoplethysmography (PPG), the pulse wave signal can be obtained non-invasively and painlessly using optical principles. The pulse wave is formed by the heart's pulsation along the arteries and blood flow, and contains information such as blood pressure, blood oxygen, and respiratory rate. It is feasible to estimate blood pressure directly based on the pulse wave signal, but the accuracy is generally low, and the difficulty of establishing a model is also high. In addition, this method also requires direct attachment to human skin to collect PPG signals, which limits many application scenarios. Based on PPG, non-contact photoplethysmography (remote PPG, rPPG) can achieve non-contact pulse wave acquisition. This non-contact pulse wave acquisition method has great advantages in special environments where close contact is not possible. Among them, rPPG signals are generally collected by cameras, and most cameras are multispectral. For example, the rPPG signal collected by a visible light red, green, and blue (RGB) camera contains signals of three channels, RGB, which are generated by three wavelengths of light. It is feasible to estimate blood pressure by taking only one of the signals of the channel, but the accuracy of estimating blood pressure is extremely low and cannot be applied in reality; fusing the signals of the three channels into one rPPG signal and then using the fused signal to estimate blood pressure can improve the problem of low accuracy of single channel signal estimation, but the fusion operation has certain complexity, and signal loss may occur during the fusion process, which also loses the accuracy of measurement to a certain extent. In addition, the method of estimating blood pressure based on pulse transit time (PTT) has also been widely studied. PTT is defined as the time for blood to flow from one point to another in the cardiovascular system when the heart fluctuates. It can reflect the blood flow velocity and has a high correlation with blood pressure. The common PTT is obtained by electrocardiogram signal and PPG or multi-PPG, which increases the complexity of signal acquisition to a certain extent. Summary of the invention

[0004] The technical problem to be solved by the present invention is to measure blood pressure by visible light, thereby realizing a non-contact, convenient and accurate blood pressure measurement system.

[0005] The technical solution adopted by the present invention is as follows: a non-contact blood pressure measurement method based on visible light vision, the method comprising:

[0006] Step 1: Use a multispectral camera to collect human video signals and select the best region of interest (ROI) for tracking, and use a continuous electronic blood pressure monitor to measure real-time continuous blood pressure; the multispectral camera resolution is at least 720p, and the frame rate range is set to 15-30fps according to the pulse wave frequency. The collection site is not limited, and the best region of interest is the skin area in the video;

[0007] Step 2: Convert the video signal into a time series signal of each channel;

[0008] Step 2.1: Divide the acquired image into three channels, namely, images of red spectral band, green spectral band and blue spectral band;

[0009] Step 2.2: After channel separation, the average pixel grayscale value in the best region of interest of each channel image is calculated respectively, and the calculated value is used as the amplitude value of the corresponding channel signal at that moment. The time series signal is the original pulse wave signal;

[0010] Step 3: Preprocessing the original pulse wave signal;

[0011] The original pulse wave signals of each channel are preprocessed in the same way to remove non-pulse wave components. The specific operations include denoising, drift removal, DC component removal, and abnormal point removal.

[0012] Step 4: Extract signal correlation features;

[0013] Extract the half-width pulse wave and the rising area of ​​the heart's systolic period of each channel's pulse wave, and then calculate the difference and ratio of the features extracted from each channel; locate the systolic peak, trough and maximum slope point of the pulse wave of each channel, and directly calculate the time difference and amplitude difference of the above feature points between each channel;

[0014] Step 5: Construct a blood pressure estimation model and establish a blood pressure measurement system;

[0015] Repeat the above four steps to extract different sample features, and build a blood pressure estimation model based on all sample features and real-time blood pressure measurement values. The input of the blood pressure estimation model is the signal correlation feature extracted in step 4, and the output is the blood pressure value; use the estimation model to establish a blood pressure measurement system.

[0016] Furthermore, the blood pressure estimation model constructed in step 5 is a multiple linear regression model, and a training set is formed using a part of the features in step 4 and the real-time blood pressure values ​​to obtain the regression parameters of the regression model.

[0017] The present invention has great advantages over the method described in the background art, as follows:

[0018] The present invention innovatively uses a multi-spectral camera to obtain pulse wave signals of multiple optical bands, and uses the subtle differences between pulse waves in different bands and the correlation between multiple pulse waves to estimate blood pressure, thereby enhancing the independence of information carried in pulse wave signals corresponding to light of different wavelengths and reducing information loss; and at the same time, using pulse wave signals corresponding to light of different wavelengths can suppress motion interference and environmental noise that occur during measurement, thereby improving measurement accuracy.

[0019] The present invention only needs to collect one video signal to measure blood pressure. Compared with the method of simultaneously collecting PPG signals and ECG signals to measure blood pressure and the method of simultaneously collecting PPG signals of multiple parts to measure blood pressure, this method significantly reduces the complexity of the blood pressure measurement operation.

[0020] The present invention does not limit the signal collection area, and any skin area exposed to the camera can be used for blood pressure measurement. Therefore, in actual application, there is no need to restrict the user's movement state. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the present invention.

[0022] Figure 2 It is a device for obtaining video signals and reference blood pressure.

[0023] Figure 3 It is a feature that can be extracted from the pulse wave signal of a single channel.

[0024] Figure 4 It is the time difference characteristic between the pulse wave signals of each two channels.

[0025] Figure 5 It is the amplitude difference characteristic between the pulse wave signals of each two channels. DETAILED DESCRIPTION

[0026] The present invention provides a non-contact blood pressure measurement system based on visible light vision. The core of the system is to estimate blood pressure by using the correlation characteristics between pulse wave signals corresponding to visible light of different wavelengths, so as to achieve non-invasive, non-contact and accurate blood pressure measurement. The specific implementation scheme is as follows, taking RGB three wavelengths of light as an example:

[0027] Step 1: Collect human video signals and select the best ROI for tracking, while measuring the reference continuous blood pressure value. In a bright environment, use an RGB camera to collect human video signals. The subjects work and study normally in front of the camera without being blocked. During the specific implementation, parameters such as video resolution, frame rate, and video acquisition time can be selected according to actual conditions. The best ROI includes large skin pixels and the proportion of skin pixels is not less than 95%. While collecting video signals, use an electronic blood pressure meter to measure real-time continuous blood pressure as a blood pressure reference value.

[0028] Step 2: Generate RGB three-channel original pulse wave signal. Separate the three channels of each frame of video image, and calculate the average pixel grayscale value in the ROI area of ​​the three channels respectively. The grayscale value is the amplitude value of the time series signal. The time series signal obtained is the original pulse wave signal. The calculation formula is as follows:

[0029]

[0030] Where n is the number of image frames, N is the total number of pixels, X(n) represents the average pixel grayscale value of the nth frame image, S n Represents the ROI corresponding to the nth frame image, and f(r,c) represents the pixel value at the pixel point (r,c).

[0031] Step 3: Preprocess the original pulse wave signal to remove non-pulse wave components. The same preprocessing method is used for the original three-channel pulse wave signals, where wavelet transform is used to remove baseline drift; DC components are removed by subtracting the signal average; points in the signal whose amplitude value exceeds three times the signal amplitude average are considered abnormal points and are directly deleted.

[0032] Step 4: Extract the correlation features between the pulse wave signals corresponding to each wavelength of visible light. Extract the contrast correlation features and direct correlation features between the three RGB channel signals respectively. This specific implementation scheme takes the video signals of the three RGB channels as an example. The blood absorbs the three wavelengths of light to different degrees, and the depths of the three wavelengths of light penetrating the skin are also different. The red light can penetrate deep into the skin. Therefore, the pulse wave signal of the red channel usually carries relevant information about the changes in arterial blood volume. Similarly, the pulse wave signal of the green channel carries information about the volume changes of small arteries and veins, while the pulse wave signal of the blue channel generally only contains information about the mechanical movement of the skin surface. The pulse wave signals of the three channels carry different information, that is, some features in the pulse wave signals of each channel will also be different. By comparing the features of the signals of each channel and extracting the contrast correlation features between the signals of each channel, information about blood pressure can be mined. The contrast features can be PWHA, systolic time (T sys ), systolic signal rising area (S sys ), the time from the end of contraction to the dicrotic notch, etc. Based on the propagation process of blood flowing out of the heart to the arteries and then flowing back to the heart through the veins during the cardiac cycle, it can be inferred that the pulse wave signal of the red channel first senses the change in blood volume, followed by green and blue light. Therefore, by extracting the directly related features between the signals of each channel, such as time difference and amplitude difference, blood pressure can also be estimated. The key to extracting the above features is the positioning of signal-related feature points. The specific feature extraction examples are as follows:

[0033] The comparison and correlation features are taken as examples of the ratio of PWHA between the two RG channel signals and the difference in the systolic rising area of ​​the two RG channel signals. The PPG amplitude can be calculated by locating the maximum and minimum values ​​in a single cycle of the pulse wave, and the half-amplitude point can be located by the PPG amplitude. Then the PWHA can be obtained by calculating the time difference between the half-amplitude points. Finally, the ratio of the PWHA of the two RG channel signals can be simply calculated to obtain the ratio of the PWHA between the two RG channel signals. The systolic rising area of ​​the two RG channel signals can be obtained by integration. First, the trough and peak of the signal in the cycle are located by peak detection, and then the integral between the two points is calculated to obtain the rising area. Finally, the difference in the systolic rising area of ​​the two RG channel signals can be obtained by simply calculating the difference.

[0034] The direct correlation feature takes the time difference between the maximum slope point of the R channel and the end-diastolic point of the G channel as an example. The maximum slope point of the R channel can be located by the peak of the first-order derivative of the pulse wave signal, and the end-diastolic point of the G channel can be located by the trough of the pulse wave signal. The time difference feature between the RG signals can be obtained by directly calculating the time difference between the two points.

[0035] Step 5: Build a blood pressure estimation model and establish a blood pressure measurement system for actual measurement. Repeat the above five steps to extract the features of different samples and reference blood pressure, and use these sample features and reference blood pressure to train a blood pressure estimation regression model. Finally, build a blood pressure measurement system, where the input of the system is the human body video signal, and the output is the measured real-time blood pressure and blood pressure health status.

Claims

1. A non-contact blood pressure measurement method based on visible light vision, the method comprising: Step 1: Use a multispectral camera to collect human video signals and select the best region of interest for tracking, and use a continuous electronic blood pressure meter to measure real-time continuous blood pressure; The resolution of the multispectral camera is at least 720p, the frame rate range is set to 15-30fps according to the pulse wave frequency, and the acquisition site is not limited, and the best area of ​​interest is the skin area in the video; Step 2: Convert the video signal into a time series signal of each channel; Step 2.1: Divide the acquired image into three channels, namely, images of red spectral band, green spectral band and blue spectral band; Step 2.2: After channel separation, the average pixel grayscale value in the best region of interest of each channel image is calculated respectively, and the calculated value is used as the amplitude value of the corresponding channel signal at that moment. The time series signal is the original pulse wave signal; Step 3: Preprocessing the original pulse wave signal; The original pulse wave signals of each channel are preprocessed in the same way to remove non-pulse wave components. The specific operations include denoising, drift removal, DC component removal and abnormal point removal. Step 4: Extract signal correlation features; Extract the half-width pulse wave and the rising area of ​​the heart's systolic period of each channel's pulse wave, and then calculate the difference and ratio of the features extracted from each channel; locate the systolic peak, trough and maximum slope point of the pulse wave of each channel, and directly calculate the time difference and amplitude difference of the above feature points between each channel; Step 5: Construct a blood pressure estimation model and establish a blood pressure measurement system; Repeat the above 4 steps to extract different sample features, and build a blood pressure estimation model based on all sample features and real-time blood pressure measurement values. The input of the blood pressure estimation model is the signal correlation feature extracted in step 4, and the output is the blood pressure value. Use the estimation model to establish a blood pressure measurement system. The constructed blood pressure estimation model is a multiple linear regression model, which uses a part of the features in step 4 and the real-time blood pressure values ​​to form a training set to obtain the regression parameters of the regression model.

Images