Non-invasive continuous blood pressure monitoring method and system based on patch type ultrasound

Abstract

The invention relates to the field of non-invasive blood pressure measurement, and discloses a non-invasive continuous blood pressure monitoring method and system based on patch type ultrasound, and the method comprises the steps: obtaining an ultrasonic video of a blood vessel in real time; synchronously acquiring a real blood pressure value; performing normalization processing on the true value of the blood pressure; constructing a blood pressure prediction model, and training the blood pressure prediction model by taking the normalized real blood pressure value as a training label and taking the ultrasonic video as the input of the blood pressure prediction model; performing initial calibration on the blood pressure of the user to be measured; and predicting the blood pressure in real time by adopting the trained blood pressure prediction model. The AI model trained by the invasive blood pressure data and the calibration algorithm aiming at individual differences are combined, so that the high consistency of the predicted value and the real blood pressure is ensured.

Description

Technical Field

[0001] The present invention relates to the field of non-invasive blood pressure measurement, and in particular to a non-invasive continuous blood pressure monitoring method and system based on patch-type ultrasound. Background Art

[0002] Blood pressure is a key indicator for measuring cardiovascular health. Traditional cuff-type sphygmomanometers have discontinuity and errors in daily measurements, while invasive blood pressure measurement, although accurate, is only applicable to specific scenarios and has a relatively high risk.

[0003] In recent years, with the development of ultrasound imaging technology and artificial intelligence algorithms, non-invasive blood pressure measurement has received extensive attention, but existing methods still have deficiencies in terms of accuracy. Summary of the Invention

[0004] The main object of the present invention is to solve the technical problem that existing methods still have deficiencies in terms of accuracy. A non-invasive continuous blood pressure monitoring method based on patch-type ultrasound includes the following steps:

[0005] Obtain the ultrasound video of blood vessels in real time; synchronously obtain the true blood pressure value; perform normalization processing on the true blood pressure value; construct a blood pressure prediction model, use the normalized true blood pressure value as the training label, use the ultrasound video as the input of the blood pressure prediction model, and train the blood pressure prediction model; perform initial calibration on the blood pressure of the user to be measured; use the trained blood pressure prediction model to perform real-time prediction on the blood pressure.

[0006] The formula for performing normalization processing on the true blood pressure value is:

[0007]

[0008] where y is the actual measured value, y max is the maximum value in a single measurement process, y min is the minimum value in a single measurement process, and y b is the processed data.

[0009] Constructing a blood pressure prediction model, using the normalized true blood pressure value as the training label, using the ultrasound video as the input of the blood pressure prediction model, and training the blood pressure prediction model includes:

[0010] Adopt a method based on 3D-UNet to extract the hidden features in the ultrasound video; design a prediction head to map the image features to continuous blood pressure values;

[0011] The input of the blood pressure prediction model is the consecutive frames of the ultrasound video, forming a 5D tensor of B×N×C×H×W. Here, N is the number of frames, set to 30, the frame rate of the ultrasound image is set to 30, corresponding to collecting 1s of ultrasound video, H and W are the height and width of the image, the original image is reset to 256*256, both H and W are 256, B is the Batch size, the amount of data fed to the network in each batch, and C is the number of image channels;

[0012] The output of the blood pressure prediction model is a blood pressure sequence of B×M×1; where, M is the sequence length of the consecutive blood pressure outputs; M is 1000;

[0013] Normalize the blood pressure labels;

[0014] Perform random cropping, rotation, and adding noise transformation on each frame of the ultrasound video, and add Gaussian noise with a mean of 0 and σ = 0.2 to the blood pressure signal;

[0015] Build a 3D-Unet structure, construct an encoder and a decoder; among them, the encoder is 4 layers of 3D convolution + BatchNorm + ReLU + 3D MaxPooling, repeated four times, and the number of channels doubles layer by layer; the output shapes of the four layers are respectively: [B,N,64,H / 2,W / 2], [B,N,128,H / 4,W / 4], [B,N,256,H / 8,W / 8], [B,N,512,H / 16,W / 16]; the size of the 3D convolution kernel used is [3,3,3], and the size of the pooling layer is [1,2,2]; the decoder is 3D transposed convolution + upsampling + skip connection, repeated 4 times, and the number of channels is halved layer by layer. Finally, the decoder outputs: [B,N,64,H,W];

[0016] Sample the network output vector to [B,N,5,5] through multi-layer CNN convolution;

[0017] Use an MLP head to transform the network output dimension of the previous step to [B,M,1];

[0018] The loss function is Mean Squared Error (MSE), which calculates the mean square error between the predicted blood pressure sequence and the true value. The formula is:

[0019]

[0020] where y i is the blood pressure label value, is the blood pressure value predicted by the network; M is the sequence length of the blood pressure sequence predicted by the network.

[0021] The training settings are as follows: Use the AdamW optimizer with weight decay, set the initial learning rate to 1e-4, and adopt the CosineAnnealingLR scheduling; momentum: β1 = 0.9, β2 = 0.999, weight decay: 1e-5 (to prevent overfitting); set early stopping.

[0022] The initialization calibration of the blood pressure of the user to be measured includes:

[0023] Calibrate according to the diastolic blood pressure and systolic blood pressure of the patient; when used for the first time, obtain the reference blood pressure of the user using a cuff blood pressure monitor or other standard device; then add the reference blood pressure value to the result obtained by the network model to obtain the blood pressure value predicted based on the ultrasound video.

[0024] The second aspect of the present invention provides a non-invasive continuous blood pressure monitoring system based on patch-type ultrasound, including:

[0025] An ultrasound device for real-time acquisition of ultrasound videos of blood vessels;

[0026] A monitor for synchronously acquiring the true blood pressure value;

[0027] A normalization processing unit for normalizing the true blood pressure value;

[0028] A model construction and training unit for constructing a blood pressure prediction model, using the normalized true blood pressure value as the training label, using the ultrasound video as the input of the blood pressure prediction model, and training the blood pressure prediction model;

[0029] A calibration unit for initializing and calibrating the blood pressure of the user to be measured;

[0030] A prediction unit for real-time prediction of blood pressure using the trained blood pressure prediction model.

[0031] The third aspect of the present invention provides an electronic device, including: a memory and at least one processor, instructions are stored in the memory, and the memory and the at least one processor are interconnected by a line; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned non-invasive continuous blood pressure monitoring method based on patch-type ultrasound.

[0032] The fourth aspect of the present invention provides a computer-readable storage medium, instructions are stored in the computer-readable storage medium, and when it runs on a computer, it causes the computer to execute the above-mentioned non-invasive continuous blood pressure monitoring method based on patch-type ultrasound.

[0033] The present invention has the following beneficial effects:

[0034] The present invention combines an AI model trained with invasive blood pressure data and a calibration algorithm for individual differences to ensure a high degree of consistency between the predicted value and the true blood pressure.

[0035] The present invention uses a patch-type sensor to collect data, avoiding the invasiveness and discomfort of traditional measurement methods.

[0036] The sensor of the present invention is designed in a miniaturized manner, suitable for daily monitoring and mobile health management scenarios.

[0037] The present invention dynamically monitors the changes in the brachial artery, radial artery, and popliteal artery, providing a continuous and real-time blood pressure trend graph. Description of the Drawings

[0038] Figure 1 is a flowchart of a non-invasive continuous blood pressure monitoring method based on patch-type ultrasound;

[0039] Figure 2 is the arterial ultrasound image collected;

[0040] Figure 3 is the continuous invasive blood pressure data collected by the monitor;

[0041] Figure 4 The prediction result graph of the non-invasive continuous blood pressure monitoring method based on patch-type ultrasound of the present invention. Detailed Embodiments

[0042] In the specification and claims of the present invention and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not have to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0043] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to Figures 1-3 , the embodiments of the non-invasive continuous blood pressure monitoring method based on patch-type ultrasound in the embodiments of the present invention include:

[0044] Obtain the ultrasonic video of blood vessels in real time; synchronously obtain the true blood pressure value; perform normalization processing on the true blood pressure value; construct a blood pressure prediction model, use the normalized true blood pressure value as the training label, use the ultrasonic video as the input of the blood pressure prediction model, and train the blood pressure prediction model; perform initial calibration on the blood pressure of the user to be measured; use the trained blood pressure prediction model to perform real-time prediction on the blood pressure.

[0045] Specifically as follows:

[0046] 1. Hardware part:

[0047] The ultrasonic patch sensor is composed of a high-frequency ultrasonic transducer and a flexible material, which adapts to the anatomical characteristics of the carotid artery, brachial artery, radial artery and popliteal artery. The patch-type ultrasonic probe sensor is a rigid linear array probe, and the surface is wrapped with a flexible material, which fits the skin and is convenient for long-term wearing.

[0048] The data transmission module uses an Android tablet device, which can locally save the real-time image or send it to the cloud.

[0049] Attach ultrasonic sensors at positions such as the carotid artery, brachial artery, radial artery, and popliteal artery to obtain clear arterial ultrasonic B-Mode images.

[0050] 2. Data acquisition and synchronization scheme:

[0051] Collect real-time image sequences of arterial cross-sectional changes through a patch-type ultrasonic device. And use the method of arterial catheter to synchronously obtain the real-time blood pressure waveform of the brachial artery.

[0052] In the actual acquisition process, it is necessary to consider the time synchronization between the ultrasonic video and the invasive continuous blood pressure, and a synchronization measurement scheme is designed. First, the ultrasonic video data acquisition is triggered by a pulse signal. When the external trigger input is high level, the ultrasonic device will store the collected arterial B-Mode images in a specified folder in real time. When the external trigger input is low level, the ultrasonic device stops storing images. The monitor converts the continuous blood pressure data into an analog signal output, and a USB data acquisition card is used for ADC acquisition. At the same time, the USB data acquisition card can control the high and low levels of the IO output, which can be used as the external trigger input of the ultrasonic device. Therefore, as long as the analog signal lines led out from the ultrasonic device and the monitor are connected to the same USB data acquisition card, the ADC acquisition and the output of high level can be started simultaneously by controlling the USB data acquisition card, and the synchronization of the collected ultrasonic video and the continuous blood pressure data in time can be achieved.

[0053] Due to the inconsistent sampling rates of the ultrasonic video and the invasive blood pressure, it is necessary to perform upsampling processing on the invasive blood pressure data. Generally, the frame rate of the ultrasonic video is between 10 and 30 Hz, and the frame rate of the invasive blood pressure data can reach 1000 Hz. The appropriate upsampling interval can be selected according to the actual situation.

[0054] 3. Data preprocessing

[0055] Use the obtained invasive blood pressure data as the true blood pressure value for normalization processing.

[0056]

[0057] where y is the actual measured value, y max is the maximum value during a single measurement, y min is the minimum value during a single measurement, y b is the processed data.

[0058] To improve the generalization of the model, the ultrasonic video data is processed by adding random noise, cropping, scaling, rotation, etc. to achieve data augmentation.

[0059] 4. AI algorithm and neural network model:

[0060] Use the processed blood pressure data as the training label, and use the video data obtained by the patch-type ultrasonic device as the input of the neural network to train the neural network model.

[0061] 4.1. Adopt a method based on 3D-UNet to extract the hidden features in the ultrasonic video, and then design a network layer based on convolution and MLP to map the image features into continuous blood pressure values.

[0062] 4.2. The network input is the continuous frames of the ultrasonic video, forming a 5D tensor (B×N×C×H×W, where N is the number of frames, set to 30, the frame rate of the ultrasonic image is set to 30, so it corresponds to collecting 1s of ultrasonic video, H and W are the height and width of the image, resize the original video to 256*256, then both H and W are 256, B is the Batch size, the amount of data fed to the network in each batch), and C is the number of channels of the image.

[0063] 4.3. The output of the network is a blood pressure sequence (B×M×1). M is the sampling frequency of the continuous blood pressure output. Since the sampling frame rate of the continuous blood pressure signal is much greater than the ultrasonic frame rate, time super-resolution processing is designed. Here M is 1000.

[0064] 4.4 Data preprocessing: To normalize the blood pressure label.

[0065]

[0066] 4.5. Data augmentation: Randomly crop, rotate, add noise, etc. to each frame of the ultrasonic video, and add Gaussian noise with a mean of 0 and σ = 0.2 to the blood pressure signal.

[0067] 4.6. Use PyTorch to build a 3D-Unet structure, constructing an encoder and a decoder. The encoder consists of 4 layers of 3D convolution + BatchNorm + ReLU + 3D MaxPooling, repeated four times, with the number of channels doubling layer by layer. The output shapes of the four layers are respectively: (B, 64, N, H / 2, W / 2), (B, 128, N, H / 4, W / 4), (B, 256, N, H / 8, W / 8), (B, 512, N, H / 16, W / 16). The size of the 3D convolution kernel used is (3, 3, 3), and the size of the pooling layer is (1, 2, 2). The decoder is 3D transposed convolution + upsampling + skip connection, repeated 4 times, with the number of channels halving layer by layer. Finally, the decoder outputs: (B, 64, N, H, W).

[0068] 4.7. Transform the network output matrix into (B, N, 5, 5) through multi-layer CNN convolution, pooling, and non-linear transformation;

[0069] 4.8. Use an MLP head to transform the network output dimension in the previous step into (B, M, 1);

[0070] 4.9. Loss function: Mean Squared Error (MSE), calculate the mean squared error between the predicted blood pressure sequence and the true value.

[0071]

[0072] where y i is the blood pressure label value, is the blood pressure value predicted by the network. M is the length of the blood pressure sequence predicted by the network.

[0073] 4.10. Training settings: Use the AdamW (with weight decay) optimizer, set the initial learning rate: 1e-4, and adopt the cosine annealing scheduler (CosineAnnealingLR). Momentum: β1 = 0.9, β2 = 0.999, weight decay: 1e-5 (to prevent overfitting). Set early stopping.

[0074] 4. Calibration algorithm design:

[0075] To improve the generalization of the model, continuous blood pressure is normalized during training. However, during actual testing, calibration needs to be performed according to the diastolic and systolic blood pressures of the patient. When using it for the first time, use a cuff blood pressure monitor or other standard device to obtain the user's baseline blood pressure. Then add the baseline blood pressure value to the result obtained by the network model to get the blood pressure value predicted based on the ultrasound image.

[0076] 5. Blood pressure prediction and output:

[0077] Based on the trained neural network model, when deployed to a cloud server or a local device, it can output in real time the blood pressure values measured from the carotid artery, brachial artery, radial artery and popliteal artery.

[0078] 6. Experimental verification:

[0079] Data of the carotid artery, brachial artery, radial artery and popliteal artery are collected from subjects of different ages, genders and health conditions. During the collection process, the subjects use the supine position to make the height of the invasive arterial catheter pressure sensor level with the heart; the invasive blood pressure measurement results are used as the true values to verify the accuracy of the prediction model. The final result shows that the blood pressure error is < 4 mmHg, meeting the daily use requirements. The prediction result diagram of the present invention is shown in Figure 4 .

[0080] The non-invasive continuous blood pressure monitoring method based on patch-type ultrasound in the embodiments of the present invention is described above. Next, the non-invasive continuous blood pressure monitoring device based on patch-type ultrasound in the embodiments of the present invention is described, as shown in Figure 2 and Figure 3 :

[0081] A patch-type ultrasound device for real-time acquisition of ultrasound videos of blood vessels;

[0082] A monitor for synchronously acquiring the true blood pressure values;

[0083] A normalization processing unit for normalizing the true blood pressure values;

[0084] A model construction and training unit for constructing a blood pressure prediction model, using the normalized true blood pressure values as training labels, using the ultrasound videos as inputs of the blood pressure prediction model, and training the blood pressure prediction model;

[0085] A calibration unit for initializing and calibrating the blood pressure of the user to be measured;

[0086] A prediction unit for real-time prediction of blood pressure using the trained blood pressure prediction model.

[0087] An embodiment of the present invention also provides an electronic device, which may vary greatly due to different configurations or performances. It may include one or more processors (central processing units, CPUs) (for example, one or more processors) and a memory, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage media may be transient storage or persistent storage. The programs stored in the storage media may include one or more modules, and each module may include a series of instruction operations on the electronic device. Further, the processor may be configured to communicate with the storage media and execute a series of instruction operations in the storage media on the electronic device.

[0088] The electronic device may further include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that the structure of the electronic device in this embodiment does not constitute a limitation on the electronic device, and it may include more or fewer components, or combine certain components, or have different component arrangements.

[0089] An embodiment of the present invention provides a structure of an electronic device, which may vary greatly due to different configurations or performances. It may include one or more processors (central processing units, CPUs) (for example, one or more processors) and a memory, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage media may be transient storage or persistent storage. The programs stored in the storage media may include one or more modules, and each module may include a series of instruction operations on the electronic device. Further, the processor may be configured to communicate with the storage media and execute a series of instruction operations in the storage media on the electronic device.

[0090] The electronic device may further include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that the structure of the electronic device does not constitute a limitation on the electronic device, and it may include more or fewer components than the foregoing, or combine certain components, or have different component arrangements.

[0091] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are run on a computer, the computer is caused to execute the steps of the foregoing method.

[0092] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, device, or unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0093] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0094] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A non-invasive continuous blood pressure monitoring method based on patch-type ultrasound, characterized in that It includes the following steps: Obtain the ultrasonic video of blood vessels in real time; Synchronously obtain the true blood pressure value; Perform normalization processing on the true blood pressure value; Construct a blood pressure prediction model. Use the normalized true blood pressure value as the training label, and use the ultrasonic video as the input of the blood pressure prediction model to train the blood pressure prediction model; Perform initial calibration on the blood pressure of the user to be measured; Use the trained blood pressure prediction model to perform real-time prediction on the blood pressure.

2. The non-invasive continuous blood pressure monitoring method based on patch-type ultrasound according to claim 1, wherein The formula for normalizing the true blood pressure value is: Among them, is the actual measured value, is the maximum value during one measurement process, is the minimum value during one measurement process, is the processed data.

3. The non-invasive continuous blood pressure monitoring method based on patch-type ultrasound according to claim 1, characterized in that Construct a blood pressure prediction model. Use the normalized true blood pressure value as the training label, and use the ultrasonic video as the input of the blood pressure prediction model to train the blood pressure prediction model, including: Use the method based on 3D-UNet to extract the hidden features in the ultrasonic video; design a prediction head to map the image features into continuous blood pressure values; The input of the blood pressure prediction model is the continuous frames of the ultrasonic video, forming a 5D tensor of B × N× C × H × W, where N is the number of frames, set to 30, the frame rate of the ultrasonic image is set to 30, corresponding to collecting 1s of ultrasonic video, H and W are the height and width of the image, reset the original image to 256*256, both H and W are 256, B is the Batch size, the amount of data fed to the network in each batch, and C is the number of channels of the image; The output of the blood pressure prediction model is a blood pressure sequence of B × M × 1; where M is the sequence length of the continuous blood pressure output; M is 1000; Perform normalization processing on the blood pressure label; Randomly crop, rotate, and add noise transformation to each frame of the ultrasound video, and add Gaussian noise with a mean of 0 to the blood pressure signal. to it; Build a 3D-Unet structure, construct an encoder and a decoder; among them, the encoder is 4 layers of 3D convolution + Batch Norm + ReLU + 3D MaxPooling, repeated four times, and the number of channels doubles layer by layer; the output shapes of the four layers are: [B, N, 64, H / 2, W / 2], [B, N, 128, H / 4, W / 4], [B, N, 256, H / 8, W / 8], [B, N, 512, H / 16, W / 16]; the size of the 3d convolution kernel used is [3,3,3], and the size of the pooling layer is [1,2,2]; the decoder is 3D transposed convolution + upsampling + skip connection, repeated 4 times, and the number of channels is halved layer by layer. Finally, the decoder outputs: [B, N, 64, H, W]; Sample the network output vector to [B,N,5,5] through multi-layer CNN convolution; Use the MLP head to transform the network output dimension of the previous step into [B,M,1]; The loss function is Mean Squared Error (MSE), and calculate the mean square error between the predicted blood pressure sequence and the true value. The formula is: wherein is the blood pressure tag value, is the blood pressure value predicted by the network; M is the length of the blood pressure sequence predicted by the network.

4. The non-invasive continuous blood pressure monitoring method based on patch-type ultrasound according to claim 3, characterized in that, The training is set as follows: Use the AdamW optimizer with weight decay, set the initial learning rate: 1e-4, and adopt the CosineAnnealingLR scheduling; Momentum: β1 = 0.9, β2 = 0.999, weight decay: 1e-5 (to prevent overfitting); set early stopping.

5. The non-invasive continuous blood pressure monitoring method based on patch-type ultrasound according to claim 1, characterized in that The initial calibration of the blood pressure of the user to be measured includes: Calibrate according to the patient's diastolic blood pressure and systolic blood pressure; at the first use, use a cuff-type sphygmomanometer or other standard device to obtain the user's reference blood pressure; then add the reference blood pressure value to the result obtained by the network model to obtain the blood pressure value predicted according to the ultrasonic video.

6. Non-invasive continuous blood pressure monitoring system based on patch-type ultrasound, characterized in that, The system includes: A patch-type ultrasonic device for obtaining real-time ultrasonic videos of blood vessels; A monitor for synchronously obtaining the true blood pressure value; A normalization processing unit for normalizing the true blood pressure value; A model construction and training unit for constructing a blood pressure prediction model, using the normalized true blood pressure value as the training label, using the ultrasonic video as the input of the blood pressure prediction model, and training the blood pressure prediction model; A calibration unit for initializing and calibrating the blood pressure of the user to be measured; A prediction unit for performing real-time prediction of blood pressure using the trained blood pressure prediction model.

7. An electronic device, the electronic device includes a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes each step of the non-invasive continuous blood pressure monitoring method based on patch-type ultrasound according to any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, each step of the non-invasive continuous blood pressure monitoring method based on patch-type ultrasound according to any one of claims 1-5 is implemented.

Images