Non-contact continuous blood pressure monitoring device, computer equipment and storage medium for nighttime sleep

Abstract

The present invention discloses a non-contact continuous blood pressure monitoring device, computer equipment and storage medium for use in a nighttime sleep state, including a signal acquisition module, which uses a millimeter wave radar to collect human carotid artery pulse wave signals; a signal processing module, which demodulates the echo signal collected by the radar, then performs body motion detection on the demodulated signal, screens out the body motion segment, and then removes the DC bias, filters and segments the signal; a pulse wave screening module, which refers to the standard human pulse wave waveform and performs waveform screening on the segmented pulse wave; a blood pressure measurement module, which uses a deep learning method to establish a mapping relationship between blood pressure and human carotid artery pulse wave. The present invention solves the problems of body motion interference, DC offset, breathing and noise interference, and the elimination of irregular waveforms by processing and analyzing the millimeter wave radar echo signal, and inputs the pulse wave signal into a deep network model to ensure the accuracy of blood pressure measurement.

Description

Technical Field

[0001] The present invention belongs to the technical field of vital sign detection, and in particular to a non-contact continuous blood pressure monitoring device, a computer device and a storage medium for use in a nighttime sleeping state. Background Art

[0002] Nowadays, hypertension has become a serious disease worldwide. This problem not only increases the burden on the cardiovascular system, but also can cause complications such as angina, heart attack, heart failure, sudden death and stroke. The World Health Organization reported that there are an estimated 1.28 billion adult hypertensive patients worldwide, and 46% of patients are still unaware of their condition. The number of patients with hypertension is increasing year by year, and the number of patients with cardiovascular and cerebrovascular diseases is also increasing year by year. Cardiovascular and cerebrovascular diseases have become an important problem that endangers people's health. Therefore, timely and accurate blood pressure monitoring is an important means to prevent hypertension and its complications.

[0003] Compared with daytime blood pressure monitoring, nighttime blood pressure is usually a "blind spot" in people's blood pressure management. Experts and scholars believe that the human body's daytime blood pressure level and fluctuations are more affected by external factors. Compared with daytime blood pressure, nighttime blood pressure is more closely related to the risk of all-cause mortality and cardiovascular and cerebrovascular disease mortality, and can better reveal the true level of individual blood pressure and the type of physiological fluctuations, and has a higher risk prediction value. In addition, nighttime blood pressure monitoring helps to detect nocturnal and hidden hypertension symptoms early and reduce the risk of complications.

[0004] At present, the traditional method of measuring blood pressure is usually an occasional measurement during the day, which cannot be used to measure blood pressure during sleep at night. In order to achieve continuous dynamic monitoring of blood pressure, foreign scholars have proposed a "cuffless" blood pressure monitoring technology. However, since this type of equipment needs to contact the wearer's body, wearing it for a long time at night will cause discomfort or skin irritation to the wearer, and the comfort level is low, which affects the effect of nighttime blood pressure measurement. In addition, this measurement method is not applicable to certain specific individuals (such as burn patients or patients with skin diseases). Based on the above situation, domestic and foreign scholars have adopted a non-contact method using video sensors, but this method has strict requirements on the tester's skin color and ambient light, and is difficult to apply when there is insufficient light at night.

[0005] Radar sensors can accurately sense pulse signals from multiple parts of the human body, and a large number of studies have shown that the characteristics of pulse pulsation are strongly correlated with blood pressure. Therefore, radar sensors are the preferred solution for developing a new type of non-contact nighttime blood pressure monitoring method that is non-contact, real-time, generalizable, and convenient for signal acquisition. Summary of the invention

[0006] The purpose of the present invention is to provide a non-contact continuous blood pressure monitoring device, a computer device and a storage medium for use in a sleep state at night in order to address the problems existing in the above-mentioned prior art.

[0007] The technical solution to achieve the purpose of the present invention is as follows: on the one hand, a non-contact continuous blood pressure monitoring device for use in a nighttime sleeping state is provided, the device comprising a signal acquisition module, a signal processing module, a pulse wave screening module and a blood pressure monitoring module;

[0008] The signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar;

[0009] The signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and to perform DC offset removal, filtering and carotid artery pulse wave segment segmentation on the radar signal after the body motion segment is removed;

[0010] The pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard;

[0011] The blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into a deep learning network model, and output corresponding systolic and diastolic pressure values; the deep learning model is an eighteen-layer residual neural network improved by adding an attention mechanism model.

[0012] Furthermore, the signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar, and specifically includes:

[0013] The subject lies flat or sideways on the bed, and the millimeter-wave radar is installed on a tripod with the antenna facing the carotid artery.

[0014] The millimeter-wave radar transmit synthesizer generates a linear frequency modulated continuous wave signal, which is reflected when it encounters the human body. Subsequently, the millimeter-wave radar orthogonal receiver captures the echo signal and performs orthogonal mixing with the transmit signal. The high-frequency part is filtered out by a low-pass filter to obtain an intermediate frequency signal, which is then sampled by an ADC.

[0015] After analog-to-digital conversion of the intermediate frequency signal, distance information is obtained through distance FFT, phase data within a fixed distance gate is selected and saved, and the carotid artery pulse wave signal can be obtained based on the phase data.

[0016] Furthermore, the signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and to perform DC offset removal, filtering and pulse wave segment segmentation on the radar signal after removing the body motion segment, specifically including:

[0017] Step 3-1, demodulating the phase data acquired by the signal acquisition module;

[0018] Step 3-2, body movement judgment: divide the demodulated signal into a window of fixed length, calculate the mean and standard deviation of the window, and filter out the data of the body movement segment by comparing the mean and standard deviation with the mean and standard deviation of the initially set stable state;

[0019] Step 3-3, removing DC bias: After completely removing the body motion segment signal, a non-convex optimization-based method is used to remove the DC bias of the IQ two-way signal respectively;

[0020] Step 3-4, filtering: using a bandpass filter with a passband of 0.5 to 10 Hz to eliminate noise and breathing signals in the radar signal after the DC bias is removed;

[0021] Step 3-5, carotid artery pulse wave segmentation: the lowest point between adjacent pulse waves is used as the basis for segmentation, and the continuous periodic pulse wave signal is segmented into carotid artery pulse wave segments.

[0022] Furthermore, in step 3-3, the DC offset of the IQ signals is removed respectively by using a method based on non-convex optimization. The specific process is as follows:

[0023] (1) Based on l p The objective function of DC bias estimation with norm minimization is written as:

[0024]

[0025] in,

[0026]

[0027] In the formula, 0<p<1, B I [n] and B Q [n] are the I and Q signals of the nth sampling point collected by the millimeter-wave radar, N is the number of sampling points, DC I and DC Q are the DC offsets of I and Q signals respectively, and A is the amplitude;

[0028] (2) Rewrite the above objective function as p The optimization problem of minimization is:

[0029]

[0030]

[0031] Different from the l1 norm minimization problem, since l p Norm minimization is non-convex, and the iterative reweighted l1 minimization (IRL1) algorithm is an effective solution for l pThe method of minimizing the norm, therefore, the present invention generalizes the IRL1 algorithm to solve formula (3). The key to the error approximation problem based on the IRL1 algorithm is to continuously iterate and solve the following formula:

[0032]

[0033] In the formula, d is an n-dimensional real vector, l represents the number of iterations, represents the recommended weight for the lth iteration; at the beginning of the algorithm, the initial weight is set to Where 1≤i≤n; in each subsequent iteration, the i-th weight of the l+1th iteration is updated as:

[0034]

[0035] In the formula, is the d value of the ith weight at the lth iteration, ξ is a positive real number close to 0;

[0036] (3) Iterate until the algorithm converges or l reaches the maximum number of iterations;

[0037] (4) The DC offset of the I and Q signals can be obtained by equation (4), and then the DC offset is removed.

[0038] Furthermore, the pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard, specifically including:

[0039] (1) Calculating the duration of a single segmented carotid artery pulse wave segment. If it exceeds or is less than a preset normal pulse wave duration, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is discarded;

[0040] (2) Calculate the duration of a single carotid artery pulse wave segment and capture the position of the maximum value within the single pulse cycle. If the time at the position of the maximum value exceeds half of the duration of the single carotid artery pulse wave, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is removed;

[0041] (3) Capture the maximum value, initial value and last value of a single carotid artery pulse wave segment, calculate the difference between the maximum value and the initial value, and the difference between the initial value and the last value. If the difference between the absolute values ​​of the two differences is less than a set threshold, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is removed. The specific calculation formula is:

[0042] A1<0.1×A2(7)

[0043] Wherein, A1 represents the absolute value of the difference between the initial value of a single carotid artery pulse wave segment and the last value of the carotid artery pulse wave segment, and A2 represents the absolute value of the difference between the maximum value and the initial value of a single carotid artery pulse wave segment.

[0044] Furthermore, the blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into the deep learning network model and output corresponding systolic and diastolic blood pressure values, specifically including:

[0045] Step 5-1, zero-filling the screened single carotid artery pulse wave cycle signal to 128 points;

[0046] Step 5-2, normalizing the signal after zero padding;

[0047] Step 5-3, input the normalized data into the deep learning network model, measure the blood pressure, and output the systolic and diastolic pressure values.

[0048] Furthermore, the deep learning network model is an 18-layer residual neural network improved by adding an attention mechanism model, specifically comprising: a preprocessing block connected in sequence, eight residual blocks and a fully connected layer;

[0049] The preprocessing block includes a first convolutional layer, a batch normalization layer, an activation function ReLU and a pooling layer connected in sequence;

[0050] The eight residual blocks include a first residual block, a second residual block, a third residual block, a fourth residual block, a fifth residual block, a sixth residual block, a seventh residual block and an eighth residual block connected in sequence; there are two types of residual blocks, the difference being whether there is a fourth convolutional layer for the input of the next residual block before being added to the output; the residual blocks with the fourth convolutional layer include the third residual block, the fifth residual block and the seventh residual block; each residual block includes a second convolutional layer, a batch normalization layer, an activation function ReLU, a third convolutional layer, a batch normalization layer, an activation function ReLU and a GC block connected in sequence;

[0051] A GC block is defined as:

[0052]

[0053]

[0054] δ(·)=W v2 ReLU(LN(W v1 (·))) (10)

[0055] Among them, i is the index of the position, j is the enumeration of all possible positions, and xi is the i-th input feature, x j is the jth input feature, x m is the mth input feature, z i is the i-th output feature, N p is the number of locations in the feature map, W k and W v , W v1 , W v2 Both represent linear transformation matrices; α j Represents the weight at the jth position of the signal after convolution; Represents the contextmodeling module, through the weight α j Aggregate features at all positions to obtain global context features; F(·,·) means using the fusion function to aggregate global context features to the features at each position; δ(·) means using feature transform to capture inter-channel dependencies, and LN(·) means the normalization layer.

[0056] On the other hand, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of each module of the above system when executing the computer program.

[0057] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the function of each module of the above system is realized.

[0058] Compared with the prior art, the present invention has the following significant advantages:

[0059] (1) The continuous blood pressure measurement method proposed in the present invention can achieve non-contact, non-breath-holding, dynamic continuous blood pressure measurement by measuring the human carotid artery pulse wave, so that the human body can monitor blood pressure in real time during sleep at night without any discomfort.

[0060] (2) The radar signal processing flow and pulse wave screening criteria proposed in the present invention are signal processing methods summarized based on a large number of measured signals, which can greatly improve the accuracy of the pulse wave acquisition database, ensure the effectiveness of the blood pressure model input, and greatly improve the accuracy of blood pressure measurement.

[0061] (3) The blood pressure measurement model proposed in the present invention can maintain the optimal state without reducing the accuracy of the training set too much if the network depth is deep enough. The improved model with the added attention mechanism can effectively integrate global information, focus limited attention on key information, quickly obtain the most effective information, and further improve the accuracy of blood pressure monitoring.

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The present invention is a schematic diagram of a non-contact continuous blood pressure monitoring device for use during nighttime sleep.

[0064] Figure 2 This is a test scene graph in an embodiment of the present invention.

[0065] Figure 3 : is a comparison diagram of radar signals before and after DC offset removal in an embodiment of the present invention, wherein Figure 3 (A) and (D) are the I channel signal with DC offset and the I channel signal without DC offset, respectively. Figure 3 (B) and (E) are the Q channel signal with DC offset and the Q channel signal without DC offset, respectively. Figure 3 (C) and (F) are respectively the demodulated signal with DC offset and the demodulated signal without DC offset.

[0066] Figure 4 This is a structural diagram of a deep learning network model in an embodiment of the present invention.

[0067] Figure 5 4 is a structural diagram of a global context (GC) block in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0070] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0071] In one embodiment, in combination Figure 1 , provides a non-contact continuous blood pressure monitoring device for use in a nighttime sleeping state, the device comprising a signal acquisition module, a signal processing module, a pulse wave screening module and a blood pressure monitoring module;

[0072] The signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar;

[0073] The signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and to perform DC offset removal, filtering and carotid artery pulse wave segment segmentation on the radar signal after the body motion segment is removed;

[0074] The pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard;

[0075] The blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into a deep learning network model, and output corresponding systolic and diastolic pressure values; the deep learning model is an eighteen-layer residual neural network improved by adding an attention mechanism model.

[0076] Furthermore, in one embodiment, the signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar, specifically comprising:

[0077] The subject lies flat or sideways on the bed, and the millimeter-wave radar is installed on a tripod with the antenna facing the carotid artery.

[0078] Preferably, a 24 GHz millimeter wave radar is used;

[0079] Here, preferably, the distance between the radar sensor and the measured part of the human body is about 30 centimeters. Figure 2 As shown, the human body can remain relaxed during the measurement.

[0080] The millimeter-wave radar transmit synthesizer generates a linear frequency modulated continuous wave signal, which is reflected when it encounters the human body. Subsequently, the millimeter-wave radar orthogonal receiver captures the echo signal and performs orthogonal mixing with the transmit signal. The high-frequency part is filtered out by a low-pass filter to obtain an intermediate frequency signal, which is then sampled by an ADC.

[0081] After analog-to-digital conversion of the intermediate frequency signal, distance information is obtained through distance FFT, phase data within a fixed distance gate is selected and saved, and the carotid artery pulse wave signal can be obtained based on the phase data.

[0082] Furthermore, in one embodiment, the signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and perform DC offset removal, filtering and pulse wave segment segmentation on the radar signal without the body motion segment, specifically including:

[0083] Step 3-1, demodulating the phase data acquired by the signal acquisition module;

[0084] Ideally, the baseband I / Q signal is:

[0085]

[0086]

[0087] Where θ is the phase shift caused by the propagation distance of the electromagnetic signal, is the residual phase noise, x(t) is the motion information of the target to be detected, and λ is the wavelength of the radar carrier. The inverse tangent operation of equation (12) can be used to extract the phase information, namely:

[0088]

[0089] When the distance between the radar and the person is fixed, φ can be considered as a constant. Therefore, the demodulated phase information is linearly related to the displacement x(t) caused by the carotid artery pulse movement, and the result of x(t) can be obtained.

[0090] Step 3-2, body movement judgment: divide the demodulated signal into a window of fixed length, calculate the mean and standard deviation of the window, and filter out the data of the body movement segment by comparing the mean and standard deviation with the mean and standard deviation of the initially set stable state;

[0091] Here, preferably, the demodulated signal is divided into a window of 5s length.

[0092] Step 3-3, removing DC bias: After completely removing the body motion segment signal, a non-convex optimization-based method is used to remove the DC bias of the IQ signals respectively; specifically, the method includes:

[0093] Due to the existence of DC bias, the I / Q signal under non-ideal conditions can be expressed as:

[0094]

[0095] (1) Based on l p The objective function of DC bias estimation with norm minimization is written as:

[0096]

[0097]

[0098] In the formula, 0<p<1, B I [n] and B Q [n] are the I and Q signals of the nth sampling point collected by the millimeter-wave radar, N is the number of sampling points, DC I and DC Q are the DC offsets of I and Q signals respectively, and A is the amplitude;

[0099] (2) The above formula can be rewritten as p The optimization problem of minimization is:

[0100]

[0101]

[0102] Different from the l1 norm minimization problem, since l p Norm minimization is non-convex, and the iterative reweighted l1 minimization (IRL1) algorithm is an effective solution for l p Therefore, this paper generalizes the IRL1 algorithm to solve equation (17). The key to the error approximation problem based on the IRL1 algorithm is to continuously iterate and solve the following formula:

[0103]

[0104] In the formula, d is an n-dimensional real vector, l represents the number of iterations, represents the recommended weight for the lth iteration; at the beginning of the algorithm, the initial weight is set to Where 1≤i≤n; in each subsequent iteration, the i-th weight of the l+1th iteration is updated as:

[0105]

[0106] In the formula, is the d value of the ith weight at the lth iteration, ξ is a positive real number close to 0, its main purpose is to prevent The denominator of this weight is 0.

[0107] (4) Repeat the iteration until the algorithm converges or reaches the maximum number of iterations.

[0108] (5) The DC offset of the I and Q signals can be obtained by equation (18), and then the DC offset is removed (I and Q signals minus the offset). Figure 3 A comparison of the signal before and after DC offset removal is shown.

[0109] Step 3-4, filtering: using a bandpass filter with a passband of 0.5 to 10 Hz to eliminate noise and breathing signals in the radar signal after the DC bias is removed;

[0110] Step 3-5, carotid artery pulse wave segmentation: the lowest point between adjacent pulse waves is used as the basis for segmentation, and the continuous periodic pulse wave signal is segmented into carotid artery pulse wave segments.

[0111] Furthermore, in one embodiment, the pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard, specifically including:

[0112] (1) Calculate the duration of a single segmented carotid artery pulse wave segment. If it exceeds or is less than the preset duration of a normal pulse wave cycle (between 0.5s and 1.5s), the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is discarded;

[0113] (2) Calculate the duration of a single carotid artery pulse wave segment and capture the position of the maximum value within the single pulse cycle. If the time at the position of the maximum value exceeds half of the duration of the single carotid artery pulse wave, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is removed;

[0114] (3) Capture the maximum value, initial value and last value of a single carotid artery pulse wave segment, calculate the difference between the maximum value and the initial value, and the difference between the initial value and the last value. If the difference between the absolute values ​​of the two differences is less than a set threshold, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is removed. The specific calculation formula is:

[0115] A1<0.1×A2(21)

[0116] Wherein, A1 represents the absolute value of the difference between the initial value of a single carotid artery pulse wave segment and the last value of the carotid artery pulse wave segment, and A2 represents the absolute value of the difference between the maximum value and the initial value of a single carotid artery pulse wave segment.

[0117] Here, the proposed radar signal processing process and pulse wave screening criteria are signal processing methods summarized based on a large number of measured signals. They can greatly improve the accuracy of the pulse wave acquisition database, ensure the effectiveness of the blood pressure model input, and greatly improve the accuracy of blood pressure measurement.

[0118] Furthermore, in one embodiment, the blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into the deep learning network model, and output corresponding systolic and diastolic pressure values, specifically including:

[0119] Step 5-1, zero-filling the screened single carotid artery pulse wave cycle signal to 128 points;

[0120] Step 5-2, normalizing the signal after zero padding;

[0121] Step 5-3, input the normalized data into the deep learning network model, measure the blood pressure, and output the systolic and diastolic pressure values.

[0122] Here, combined Figure 4 and Figure 5 The deep learning network model is an 18-layer residual neural network improved by adding an attention mechanism model, specifically comprising: a preprocessing block connected in sequence, eight residual blocks and a fully connected layer;

[0123] The preprocessing block includes a first convolutional layer, a batch normalization layer, an activation function ReLU and a pooling layer connected in sequence; the first convolutional layer has an input signal channel of 1, a convolution channel of 64, a convolution kernel size of 7, a convolution step of 2, and a number of layers of 0 added to each edge of the input of 3; the pooling window of the pooling layer is 2;

[0124] The eight residual blocks include a first residual block, a second residual block, a third residual block, a fourth residual block, a fifth residual block, a sixth residual block, a seventh residual block and an eighth residual block connected in sequence; there are two types of residual blocks, the difference being: whether there is a fourth convolutional layer for the input of the next residual block before being added to the output; the residual blocks with the fourth convolutional layer include the third residual block, the fifth residual block and the seventh residual block; each residual block includes a second convolutional layer, a batch normalization layer, an activation function ReLU, a third convolutional layer, a batch normalization layer, an activation function ReLU and a GC block connected in sequence; the convolution kernel size of the second convolutional layer is 3 and the convolution step size is 1; the convolution kernel size of the third convolutional layer is 3 and the convolution step size is 2; the convolution kernel size of the fourth convolutional layer is 1 and the convolution step size is 2;

[0125] A GC block is defined as:

[0126]

[0127]

[0128] δ(·)=W v2 ReLU(LN(W v1 (·))) (twenty four)

[0129] Among them, i is the index of the position, j is the enumeration of all possible positions, and x i is the i-th input feature, x j is the jth input feature, x m is the mth input feature, z i is the i-th output feature, N p is the number of locations in the feature map, W k and W v , W v1 , W v2 Both represent linear transformation matrices; α j Represents the weight at the jth position of the signal after convolution; Represents the contextmodeling module, through the weight α j Aggregate features at all positions to obtain global context features; F(·,·) means using the fusion function to aggregate global context features to the features at each position; δ(·) means using feature transform to capture inter-channel dependencies, and LN(·) means the normalization layer.

[0130] The last layer of the network is a fully connected layer, with an input feature number of 2048 and an output feature number of 1000. After the ReLU activation function, the output layer outputs 2 features, which are the systolic blood pressure (SBP) value and the diastolic blood pressure (DBP) value.

[0131] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of the modules of the device when executing the computer program:

[0132] The signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar;

[0133] The signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and to perform DC offset removal, filtering and carotid artery pulse wave segment segmentation on the radar signal after the body motion segment is removed;

[0134] The pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard;

[0135] The blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into a deep learning network model, and output corresponding systolic and diastolic pressure values; the deep learning model is an eighteen-layer residual neural network improved by adding an attention mechanism model.

[0136] For the specific definition of each module, please refer to the above definition of the non-contact continuous blood pressure monitoring device for use in the nighttime sleep state, which will not be repeated here.

[0137] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the processor executes the computer program, the functions of the modules of the device are implemented:

[0138] The signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar;

[0139] The signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and to perform DC offset removal, filtering and carotid artery pulse wave segment segmentation on the radar signal after the body motion segment is removed;

[0140] The pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard;

[0141] The blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into a deep learning network model, and output corresponding systolic and diastolic pressure values; the deep learning model is an eighteen-layer residual neural network improved by adding an attention mechanism model.

[0142] For the specific definition of each module, please refer to the above definition of the non-contact continuous blood pressure monitoring device for use in the nighttime sleep state, which will not be repeated here.

[0143] As a specific example, in one of the embodiments, the present invention is further verified and explained.

[0144] In this embodiment, the carotid artery pulse waveforms of 100 subjects were collected, and a cuff-type sphygmomanometer was used as the gold standard. A total of 20,000 pulse wave segments were input for blood pressure model training, and 2424 segments were used for testing. The mean error (ME), mean absolute error (MAE) and standard deviation (STD) between the test results and the reference blood pressure were used as evaluation indicators. Table 1 is a performance analysis of the blood pressure estimation model based on the ResNet-18+GCNet network of the present invention. It can be seen that the ME, MAE and STD of the 18-layer deep neural network model used are all less than 7mmHg. The mean errors (ME) of systolic pressure (SBP) and diastolic pressure (DBP) are 0.126mmHg and -0.017mmHg, respectively. The mean absolute errors (MAE) of SBP and DBP estimates are 3.51mmHg and 2.44mmHg, respectively, and the standard deviations (STD) of SBP and DBP estimates are 6.90mmHg and 4.61mmHg, respectively.

[0145] Table 1 Performance analysis of blood pressure estimation model based on ResNet-18+GCNet network

[0146]

[0147] It can be seen from the above table that the calculation results of the present invention are accurate, and it is effective and feasible for blood pressure monitoring during sleep at night, with reliable performance and higher accuracy.

[0148] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A non-contact continuous blood pressure monitoring device for use during sleep at night, characterized in that: The device includes a signal acquisition module, a signal processing module, a pulse wave screening module and a blood pressure monitoring module; The signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar; The signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and to perform DC offset removal, filtering and carotid artery pulse wave segment segmentation on the radar signal after the body motion segment is removed; The pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than a preset standard; The blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into the deep learning network model, and output the corresponding systolic and diastolic pressure values; the deep learning network model is an 18-layer residual neural network improved by adding an attention mechanism model; The pulse wave screening module is used to screen the segmented carotid artery pulse wave segments and remove the carotid artery pulse wave signal segments whose measurement results are worse than the preset standard, specifically including: (1) Calculating the duration of a single segmented carotid artery pulse wave segment. If it exceeds or is less than a preset normal pulse wave duration, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is discarded; (2) Calculate the duration of a single carotid artery pulse wave segment and capture the position of the maximum value within the single carotid artery pulse cycle. If the time at the maximum value exceeds half of the duration of the single carotid artery pulse wave, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is removed. (3) Capture the maximum value, initial value and last value of a single carotid artery pulse wave segment, calculate the difference between the maximum value and the initial value, and the difference between the initial value and the last value. If the difference between the absolute values ​​of the two differences is less than a set threshold, the carotid artery pulse wave segment is determined to be an irregular pulse wave signal and is removed. The specific calculation formula is: A1<0.1×A2 (1) Wherein, A1 represents the absolute value of the difference between the initial value of a single carotid artery pulse wave segment and the last value of the carotid artery pulse wave segment, and A2 represents the absolute value of the difference between the maximum value and the initial value of a single carotid artery pulse wave segment.

2. The non-contact continuous blood pressure monitoring device for nighttime sleep according to claim 1, characterized in that: The signal acquisition module is used to detect the human carotid artery pulse wave signal in a non-contact manner using a millimeter wave radar, and specifically includes: The subject lies flat or sideways on the bed, and the millimeter-wave radar is installed on a tripod with the antenna facing the carotid artery. The millimeter-wave radar transmit synthesizer generates a linear frequency modulated continuous wave signal, which is reflected when it encounters the human body. Subsequently, the millimeter-wave radar orthogonal receiver captures the echo signal and performs orthogonal mixing with the transmit signal. The high-frequency part is filtered out by a low-pass filter to obtain an intermediate frequency signal, which is then sampled by an ADC. After analog-to-digital conversion of the intermediate frequency signal, distance information is obtained through distance FFT, phase data within a fixed distance gate is selected and saved, and the carotid artery pulse wave signal can be obtained based on the phase data.

3. The non-contact continuous blood pressure monitoring device for use during sleep at night according to claim 2 is characterized in that: The signal processing module is used to perform signal processing and body motion segment detection on the radar echo signal, and remove the DC offset, filter and pulse wave segment segmentation on the radar signal without the body motion segment, specifically including: Step 3-1, demodulating the phase data acquired by the signal acquisition module; Step 3-2, body movement judgment: divide the demodulated signal into a window of fixed length, calculate the mean and standard deviation of the window, and filter out the data of the body movement segment by comparing the mean and standard deviation with the mean and standard deviation of the initially set stable state; Step 3-3, removing DC bias: After completely removing the body motion segment signal, a non-convex optimization-based method is used to remove the DC bias of the IQ two-way signal respectively; Step 3-4, filtering: using a bandpass filter with a passband of 0.5 to 10 Hz to eliminate noise and breathing signals in the radar signal after the DC bias is removed; Step 3-5, carotid artery pulse wave segmentation: the lowest point between adjacent pulse waves is used as the basis for segmentation, and the continuous periodic pulse wave signal is segmented into carotid artery pulse wave segments.

4. The non-contact continuous blood pressure monitoring device for use during sleep at night according to claim 3, characterized in that: Step 3-3 uses a non-convex optimization method to remove the DC offset of the IQ signals respectively. The specific process is as follows: (1) Based on l p The objective function of DC bias estimation with norm minimization is written as: in, In the formula, 0<p<1, B I [n] and B Q [n] are the I and Q signals of the nth sampling point collected by the millimeter-wave radar, N is the number of sampling points, DC I and DC Q are the DC offsets of I and Q signals respectively, and A is the amplitude; (2) Rewrite the above objective function as l p The optimization problem of minimization is: The IRL1 algorithm is used to solve formula (3). The key to the error approximation problem of the IRL1 algorithm is to continuously iterate and solve the following formula: In the formula, d is an n-dimensional real vector, l represents the number of iterations, represents the recommended weight of the lth iteration; at the beginning of the algorithm, the initial weight is set to W i (0) =1, where 1≤i≤n; in each subsequent iteration, the i-th weight of the l+1th iteration is updated as: In the formula, is the d value of the ith weight at the lth iteration, ξ is a positive real number close to 0; (3) Iterate until the algorithm converges or reaches the maximum number of iterations; (4) The DC offset of the I and Q signals can be obtained by equation (4), and then the DC offset is removed.

5. The non-contact continuous blood pressure monitoring device for nighttime sleep according to claim 1, characterized in that: The blood pressure monitoring module is used to input the screened carotid artery pulse wave signal segments as input data into the deep learning network model and output the corresponding systolic and diastolic blood pressure values, specifically including: Step 5-1, zero-filling the screened single carotid artery pulse wave cycle signal to 128 points; Step 5-2, normalizing the signal after zero padding; Step 5-3, input the normalized data into the deep learning network model, measure the blood pressure, and output the systolic and diastolic pressure values.

6. The non-contact continuous blood pressure monitoring device for nighttime sleep according to claim 1, characterized in that: The deep learning network model is an 18-layer residual neural network improved by adding an attention mechanism model, specifically comprising: a preprocessing block connected in sequence, eight residual blocks and a fully connected layer; The preprocessing block includes a first convolutional layer, a batch normalization layer, an activation function ReLU and a pooling layer connected in sequence; The eight residual blocks include a first residual block, a second residual block, a third residual block, a fourth residual block, a fifth residual block, a sixth residual block, a seventh residual block and an eighth residual block connected in sequence; there are two types of residual blocks, the difference being whether there is a fourth convolutional layer for the input of the next residual block before being added to the output; the residual blocks with the fourth convolutional layer include the third residual block, the fifth residual block and the seventh residual block; each residual block includes a second convolutional layer, a batch normalization layer, an activation function ReLU, a third convolutional layer, a batch normalization layer, an activation function ReLU and a GC block connected in sequence; A GC block is defined as: δ( )=W v2 ReLU(LN(W v1 (·))) (10) Among them, i is the index of the position, j is the enumeration of all possible positions, and x i is the i-th input feature, x j is the jth input feature, x m is the mth input feature, z i is the i-th output feature, N p is the number of locations in the feature map, W k and W v , W v1 , W v2 Both represent linear transformation matrices; α j Represents the weight at the jth position of the signal after convolution; Represents the contextmodeling module, through the weight α j Aggregate features at all positions to obtain global context features; F(·,·) means using the fusion function to aggregate global context features to the features at each position; δ(·) means using feature transform to capture inter-channel dependencies, and LN(·) means the normalization layer.

7. The non-contact continuous blood pressure monitoring device for use during sleep at night according to claim 6, characterized in that: The channel of the input signal of the first convolution layer is 1, the channel generated by convolution is 64, the convolution kernel size is 7, and the convolution step is 2; the pooling window of the pooling layer is 2; the convolution kernel size of the second convolution layer is 3, and the convolution step is 1; the convolution kernel size of the third convolution layer is 3, and the convolution step is 2; the convolution kernel size of the fourth convolution layer is 1, and the convolution step is 2.

8. A computer 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 computer program, the functions of the modules of the device according to any one of claims 1 to 7 are realized.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the processor executes the computer program, the functions of the modules of the device according to any one of claims 1 to 7 are realized.

Images