Blood pressure monitoring positioning method and device

Abstract

The present invention provides a blood pressure monitoring positioning method and device, comprising the following steps: obtaining multiple sets of pulse wave signal data sets including sensor position values and constructing a time-series pulse waveform graph; using a CNN module to extract global features from the time-series pulse waveform graph, as well as local features based on the MiMaA algorithm; concatenating the global and local features and inputting them into an LSTM module, connecting them to a BN layer, and outputting a feature graph; inputting the feature graph into a time-step-based TA attention module; using a Softmax function to perform positioning prediction and output a position correction prompt. The present invention utilizes a deep neural network to determine the changing trend of waveform intensity and its layered morphology to determine sensor position. This method can be effectively applied to the intelligent positioning of non-invasive wearable devices, helping to improve the accuracy of biosignal recognition.

Description

Technical Field

[0001] The present invention relates to the field of vital sign detection, and in particular to a blood pressure monitoring and positioning method. The present invention also relates to a device using the method. Background Art

[0002] Current blood pressure measurement methods are generally categorized as invasive and non-invasive. Invasive blood pressure measurement involves inserting a catheter into an artery or vein, or sometimes even into the heart. Invasive methods are primarily used to save critically ill patients and are strictly prohibited for use at home. Furthermore, invasive blood pressure measurement methods are highly invasive and can easily lead to subsequent infection.

[0003] Non-invasive blood pressure detection methods mainly include volume compensation method, pulse wave velocity measurement method, photoplethysmography method, applanation tension method, etc.

[0004] The basic principle of the volume compensation method is that when the pressure on the blood vessel wall equals the internal pressure, the diameter of the blood vessel wall remains unchanged with blood pressure fluctuations. In this constant volume state, the applied external pressure and the internal pressure of the blood vessel reach equilibrium, becoming equal. This method typically uses the fingertip as the measurement site. However, this method has drawbacks: prolonged measurement can cause venous congestion, and temperature and circulating blood volume may affect arterial blood pressure at the fingertips, reducing measurement accuracy.

[0005] The pulse wave velocity method, also known as the transit time method, is based on the principle that the rate at which the pulse wave propagates through blood vessels changes with changes in arterial pressure, and that pulse wave velocity is positively correlated with arterial blood pressure. However, this method suffers from individual variability, making it difficult to accurately establish a mathematical model linking pulse wave velocity and arterial blood pressure. Currently, there is no recognized standard algorithm.

[0006] The basic principle of photoplethysmography (PPT) is that blood volume in peripheral microvasculature fluctuates pulsatingly with cardiac pulsation. This is based on the fact that light absorption by arterial blood varies with arterial pulsation, and is obtained through photoplethysmography. However, this method suffers from a major drawback: signal acquisition, which contains a significant amount of noise. Denoising this signal, particularly removing physiological noise, is a major obstacle.

[0007] The basic principle of the applanation tonometry method is to apply pressure to the blood vessel, making the surface of the vessel appear flat. The internal pressure of the vessel and the external pressure reach equilibrium. The measured external pressure value at this time is the blood pressure value. This method is highly accurate and can achieve non-invasive and continuous blood pressure measurement for a long time. The applanation tonometry method often collects signals at the radial artery. Its disadvantage is that the pressure sensor must be precisely positioned directly above the radial artery, and position deviation will affect the measurement accuracy. Summary of the Invention

[0008] In view of this, the present invention proposes a blood pressure monitoring positioning method based on the applanation tension method to improve the accuracy of blood pressure monitoring positioning.

[0009] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0010] A blood pressure monitoring positioning method, the method comprising the following steps:

[0011] Acquire multiple sets of pulse wave signal data sets including sensor position values and construct a time-series pulse waveform graph; use a CNN module to extract global features from the time-series pulse waveform graph, as well as local features based on the MiMaA algorithm; perform concatenated feature fusion on the global and local features and input them into an LSTM module, connect the BN layer, and output a feature graph; input the feature graph into a time-step-based TA attention module; use a Softmax function to perform positioning prediction, and output a position correction prompt.

[0012] Furthermore, the pulse wave signal data set is acquired using a piezoresistive sensor array attached to the radial artery at the "guan" position.

[0013] Furthermore, after the pulse wave signal data set is acquired, the noise introduced by the pulse wave signal data set during the detection and transmission process is filtered out by a hardware filtering circuit, the pulse wave signal data set is normalized, and then a time-series pulse waveform diagram is constructed.

[0014] Furthermore, the MiMaA algorithm evenly divides the pulse waveform into several image blocks, multiplies the pixel values of the image blocks containing the extreme points of the pulse waveform by a weight value of 1, multiplies the pixel values of the blank image blocks by a weight value of 0, and multiplies the pixel values of the image blocks that neither contain the extreme points of the pulse waveform nor are blank by a weight value of 0.7 to obtain the local feature code of the pulse waveform.

[0015] Furthermore, the time-step-based TA attention module adopts a gating mechanism to assign adaptive weight values to different time segments.

[0016] Furthermore, the algorithm steps of the time-step-based TA attention module include: performing a global average pooling operation on the feature map output after the LSTM module calculation and BN fusion to obtain The expression is: in It is the feature map output after LSTM and BN operation, H is Pulse wave length;

[0017] Then in The gate mechanism composed of two fully connected layers is connected, and Relu and Sigmoid activation functions are used respectively to obtain the feature weight vector of each time segment. and Calculated by the following formula: Where R is the threshold width and δ is the Relu activation function;

[0018] Calculated by the following formula: Where σ is the Sigmoid activation function, H1 and H2 are the weight matrices of the two fully connected layers respectively;

[0019] The weight vector and Perform weighted operations to obtain the output results.

[0020] The present invention also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned blood pressure monitoring and positioning method when executing the program.

[0021] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned blood pressure monitoring and positioning method is implemented.

[0022] The present invention further proposes a continuous non-invasive blood pressure detection device, including a pressure sensor, a wristband, a host, a display, a data cable and a power supply. The continuous non-invasive blood pressure detection device can provide data collection, program storage and program running functions for implementing the above-mentioned blood pressure monitoring and positioning method when executing the program.

[0023] The blood pressure monitoring and positioning method of the present invention collects multiple sets of pulse wave signal data sets to construct a time-series pulse waveform diagram; then, the multiple sets of pulse wave signal data sets are spliced according to the position order of the sensors to form pulse wave data with multiple blocks as raw pulse waveform data; the raw pulse waveform data is preprocessed using the MiMaA algorithm, and the raw pulse waveform data and the waveform data obtained by the MiMaA algorithm are used as two inputs of a parallel feature extraction module; the parallel feature extraction module fuses global and local features and inputs them into an LSTM module, connects to a BN layer, and learns the feature dependencies between waveform time segments based on a time step TA attention module; finally, the current sensor position is classified and a position correction prompt is output. The method of the present invention can accurately determine the sensor position through a P-TA-LSTM network, and will provide a position calibration direction prompt for an incorrect position, thereby realizing intelligent calibration and positioning of a non-invasive blood pressure detection device, and can realize precise positioning of a non-invasive blood pressure detection device based on a flat tension method. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0025] Figure 1 Schematic diagram of the flow of a blood pressure monitoring and positioning method according to an embodiment of the present invention;

[0026] Figure 2 Schematic diagram of a contact array of a piezoresistive sensor according to an embodiment of the present invention;

[0027] Figure 3 is a schematic diagram of the three-dimensional coordinate axes of the wrist according to an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of the position of the sensor contacts attached to the radial artery according to an embodiment of the present invention;

[0029] Figure 5 A timing pulse waveform diagram according to an embodiment of the present invention;

[0030] Figure 6 A local feature code of a pulse waveform diagram according to an embodiment of the present invention;

[0031] Figure 7 Schematic diagram of the algorithm steps of the time-step-based TA attention module according to an embodiment of the present invention;

[0032] Figure 8 Schematic diagram of the overall architecture of the deep network model P-TA-LSTM according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. It should be noted that the embodiments and features of the embodiments of the present invention may be combined with each other unless there is a conflict.

[0034] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" to another element, it may be directly connected to the other element, or there may be intermediate elements. In addition, the "connection" used here may include wireless connection. The term "and / or" used here includes all or any units and all combinations of one or more associated listed items.

[0035] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0036] This embodiment relates to a blood pressure monitoring and positioning method, which specifically includes the following steps:

[0037] Multiple pulse wave signal datasets, including sensor position values, are acquired to construct a time-series pulse waveform graph. A CNN module is used to extract global features from the time-series pulse waveform graph, as well as local features based on the MiMaA algorithm. The global and local features are concatenated and fused into an LSTM module, connected to a BN layer, and a feature graph is output. The feature graph is then input into a time-step-based TA attention module. A Softmax function is used for location prediction, and a position correction prompt is output.

[0038] Based on the above idea, this embodiment uses a sensor to collect pulse wave signals, and the sensor contacts can be as follows: Figure 2 The piezoresistive sensor array shown in FIG1 has one row and five columns. The pulse waves collected are five sets of pulse wave signal data sets. The five pulse waveforms are then fused and spliced to construct the following data set: Figure 5 Of course, it can also be expanded to an array of m rows and n columns of piezoresistive sensors. In this case, the pulse waves collected are m*n sets of pulse wave signal data sets, and the subsequent steps are to fuse and splice the m*n pulse waveforms.

[0039] During the collection, a medical expert manually places the sensor on the wrist and fixes it, with the sensor contact just above the "guan" position in the "cunguanchi" of the radial artery. Figure 4 As shown. Establish a three-dimensional coordinate axis at the wrist, as shown Figure 3As shown, taking the right hand as an example, the direction parallel to the arm direction is the x-axis direction, the y-axis direction is perpendicular to the x-axis, and forms a plane parallel to the palm with the x-axis. The z-axis is perpendicular to the x-axis and the y-axis respectively, and the three coordinate axes form a three-dimensional coordinate system. The sensor position value is defined in the three-dimensional coordinate system. The sensor position value may include: thumb direction, thumb opposite direction, wrist inner direction, wrist outer direction, wrist loose direction, wrist normal position, where "thumb direction" is defined as the positive direction of the x-axis; "thumb opposite direction" is defined as the negative direction of the x-axis; "wrist inner" direction is the negative direction of the y-axis, that is, the inner direction of the elbow; "wrist outer" direction is the positive direction of the y-axis, that is, the outer direction of the elbow; "wrist loose" means that the contact still cannot fit well with the wrist skin when pressure is applied, and the contact position is in the positive direction of the z-axis; "wrist normal position" means that the sensor contact array direction is parallel to the x-axis, and the third contact in the array is close to the "off" position, as shown in FIG. Figure 4 The sensor position value can be used to output the enhanced reference information when positioning blood pressure monitoring using this method.

[0040] The pulse wave amplitude over time, as captured by the sensor, is recorded. Noise that may have been introduced during the detection and transmission process can also be filtered out using a hardware filter circuit, preferably before digital filtering is performed to normalize the signal. After normalizing the five pulse wave signal data sets, the strongest signal is determined. Around the fifth second of the strongest pulse wave, the lowest point of the pulse wave is determined. This point is the starting point of a pulse cycle. Using this location as the starting point, multiple vibration cycles are captured to form a continuous time-series pulse waveform diagram. Because the signal data collected approximately five seconds before the start of the pulse cycle is too noisy, the signal data during this period can be removed.

[0041] Using the start and end points of the continuous sequential pulse waveforms determined above as time intervals, a total of sequential pulse waveforms matching the number of sensor contacts can be obtained. In this embodiment, five waveforms of the same time period are collected. The five acquired identical sequential pulse waveforms are spliced from top to bottom in the order of contacts to form five sets of two-dimensional waveforms with the same timing. The horizontal axis of this waveform is the time sequence, and the vertical axis is the waveform amplitude data of the five sensor contacts at the same time. The sequential pulse waveform is constructed, as shown in FIG. Figure 5 shown.

[0042] Then, a CNN module is used to extract global features of the time-series pulse waveform, as well as local features based on the MiMaA algorithm, specifically including the following methods.

[0043] The two parallel CNN modules of the deep network model are used to extract the features of the above-mentioned time series pulse waveform. One of the CNN modules is used to extract the features of the time series pulse waveform. Figure 5The original time-series pulse waveform image is used for global feature extraction, and another parallel CNN module is used to extract local features of the image based on the MiMaA algorithm.

[0044] The basic principle of the MiMaA algorithm is to evenly divide the pulse waveform graph and obtain multiple local graphs in the pulse waveform graph. In this embodiment, the waveform graph is evenly divided according to 5*5 pixels. It can also be divided into 3*3, 7*7 or other convenient a*a division methods, which are not specifically limited here. The local graph is divided into three categories of graphic block information, and different weight information is applied to different three categories of graphic blocks to process the original image. Among these three types of graphics, the first category is graphic blocks containing extreme points; the second category is graphic blocks that neither contain extreme points nor are blank; and the third category is blank blocks. Multiply the pixel values of the image blocks containing extreme points by the weight value 1, multiply the pixel values of the blank image blocks by the weight value 0, and multiply the pixel values of the image blocks that neither contain extreme points of the pulse waveform graph nor are blank by the weight value 0.7 to obtain the local feature encoding of the pulse waveform graph, such as Figure 6 As shown in Figure 2, the local feature encoding of the pulse waveform is used as the input data of the second parallel CNN.

[0045] After the two parallel modules extract global and local features respectively, they perform Concate feature fusion. The fused data is used as the LSTM input data, connected to the BN layer, and outputs the feature map.

[0046] To further explore the dependencies within the pulse waveform time series, the output feature map, after running the LSTM module and connecting it to the BN layer, is fed into the time-step-based TA attention module. This module uses a gating mechanism to assign adaptive weights to different time segments, enabling it to learn the dependencies between different time segments within the feature map.

[0047] like Figure 7 As shown, the algorithm steps of the time-step-based TA attention module include: performing a global average pooling operation on the feature map output after the LSTM module calculation and BN fusion to obtain The expression is: in It is the feature map output after LSTM and BN operation, H is The pulse wave time length. Then The gate mechanism composed of two fully connected layers is connected, and Relu and Sigmoid activation functions are used respectively to obtain the feature weight vector of each time segment. and Calculated by the following formula: Where R is the threshold width and δ is the Relu activation function; Calculated by the following formula: Where σ is the Sigmoid activation function, H1 and H2 are the weight matrices of the two fully connected layers. and Perform weighted operation to obtain output result P1.

[0048] Finally, the output result P1 is subjected to a fully connected Softmax function for six-category prediction, and the position correction prompt is output. The six-category prediction includes: wristband position is normal, wristband is loose, please move the wristband toward the thumb, please move the wristband away from the thumb, please move the wristband outward, please move the wristband inward. Softmax function definition: where z q is the output of the qth node, and Q is the number of output nodes. In this embodiment, the value of Q is 6. The Softmax function converts the output values of the six categories in this embodiment into six probability distribution values in the range [0, 1], with a sum of 1.

[0049] In summary, if Figure 8 The figure shows the overall architecture of the deep network model P-TA-LSTM of this embodiment. This embodiment uses a multi-channel piezoresistive sensor to collect multiple sets of pulse wave signals from the radial artery, preprocesses them to obtain segmented waveforms, and then splices the multiple sets of pulse wave signals in the order of sensor positions to form pulse wave data with multiple blocks as raw pulse waveform data, thereby constructing a time-series pulse waveform graph. The raw pulse waveform graph data is preprocessed using the MiMaA algorithm. The raw pulse waveform data and the waveform data obtained by the MiMaA algorithm are used as the two inputs of the parallel feature extraction module. The parallel feature extraction module then fuses global and local features and inputs them into the LSTM module, connects to the BN layer, and learns the feature dependencies between waveform time segments based on the time step TA attention module. Finally, the current sensor position is classified and output.

[0050] The present invention uses a deep neural network to determine the changing trend and layered form of waveform intensity and make a judgment on the sensor position. This method can be effectively applied to the intelligent positioning of non-invasive wearable devices and help improve the accuracy of biosignal recognition.

[0051] Example 1

[0052] A piezoresistive sensor is placed on the wrist and fixed. Five pulse wave signals are collected using the sensor for over 20 seconds. The collected signals are normalized using hardware filtering circuits and digital filtering. The lowest point of the pulse wave is found around the 5th second, which is used as the starting point of a complete pulse wave. Multiple consecutive complete pulse waveform data are captured, but the first 5 seconds or so of data are discarded due to excessive noise.

[0053] Construct five sets of waveform graphs in parallel in the same time period as the original pulse waveform graph data.

[0054] The original pulse waveform and the waveform processed by the MiMaA algorithm are used as the input of the parallel feature acquisition module. The two parallel CNN modules respectively collect global features and local features enhanced by the MiMaA algorithm.

[0055] The global features and local features are concatenated and fused as the input of the LSTM module and connected to the BN layer.

[0056] Then, a time-step-based TA attention module is used to learn the dependencies between different time segments.

[0057] Finally, the fully connected Softmax function is used to achieve classification output.

[0058] The classification output result is preset to one of six situations, which are: wristband position is normal, wristband is loose, please move the wristband toward the thumb, please move the wristband toward the opposite direction of the thumb, please move the wristband outward, please move the wristband inward.

[0059] In this example, a labeled dataset of 6,000 cases was collected, with data from six categories evenly distributed. The dataset was divided into a ratio of 7:2:1, with 70% used as a training set, 20% as a cross-validation set, and 10% as a test set. The trained model can be directly used to determine the location of non-invasive arterial blood pressure sensors.

[0060] Example 2

[0061] This embodiment relates to an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned blood pressure monitoring and positioning method is implemented.

[0062] Example 3

[0063] This embodiment relates to a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned blood pressure monitoring and positioning method is implemented.

[0064] Example 4

[0065] This embodiment relates to a continuous non-invasive blood pressure detection device, including a pressure sensor, a wristband, a host, a display, a data cable and a power supply. The continuous non-invasive blood pressure detection device can provide data collection, program storage and program running functions for implementing the above-mentioned blood pressure monitoring and positioning method when executing the program.

[0066] The continuous non-invasive blood pressure detection device of this embodiment attaches the pressure sensor to the radial artery of the wrist through a wristband. The pressure sensor collects pulse fluctuations, converts the pulse fluctuations into electrical signals and transmits them to the host. The host stores the program of the above-mentioned blood pressure monitoring and positioning method. The host runs the program on the pulse fluctuations and outputs position correction prompt information to the display through the data line.

[0067] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the scope of protection of the present invention.

[0068] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A blood pressure monitoring and positioning method, characterized in that: The method comprises the following steps: Acquire multiple sets of pulse wave signal data sets including sensor position values and construct a time-series pulse waveform graph; A CNN module is used to extract global features of the time-series pulse waveform, and a MiMaA algorithm is used to extract local features; The global features and the local features are concatenated and fused and input into the LSTM module, connected to the BN layer, and the feature map is output; Input the feature map into the time-step-based TA attention module; Use the Softmax function for positioning prediction and output position correction prompts; The MiMaA algorithm evenly divides the pulse waveform into several image blocks, multiplies the pixel values of the image blocks containing the extreme points of the pulse waveform by a weight value of 1, multiplies the pixel values of the blank image blocks by a weight value of 0, and multiplies the pixel values of the image blocks that neither contain the extreme points of the pulse waveform nor are blank by a weight value of 0.7 to obtain the local feature code of the pulse waveform; The algorithm steps of the time-step-based TA attention module include: performing a global average pooling operation on the feature map output after calculation by the LSTM module and BN fusion, and obtaining The expression is: in It is the feature map output after LSTM and BN operation, H is Pulse wave length; Then The gate mechanism composed of two fully connected layers is connected, and Relu and Sigmoid activation functions are used respectively to obtain the feature weight vector of each time segment. and Calculated by the following formula: Where R is the threshold width and δ is the Relu activation function; Calculated by the following formula: Where σ is the Sigmoid activation function, H1 and H2 are the weight matrices of the two fully connected layers respectively; The weight vector and Perform weighted operations to obtain the output results.

2. The blood pressure monitoring and positioning method according to claim 1, wherein: The pulse wave signal data set is acquired using a piezoresistive sensor array attached to the radial artery at the "guan" position.

3. The blood pressure monitoring and positioning method according to claim 1, wherein: After the pulse wave signal data set is acquired, the noise introduced by the pulse wave signal data set during detection and transmission is filtered out by a hardware filtering circuit, the pulse wave signal data set is normalized, and then a time-series pulse waveform diagram is constructed.

4. The blood pressure monitoring and positioning method according to claim 1, wherein: The time-step-based TA attention module adopts a gating mechanism to assign adaptive weight values to different time segments.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the blood pressure monitoring and positioning method according to any one of claims 1 to 4 is implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the blood pressure monitoring and positioning method according to any one of claims 1 to 4 is implemented.

7. A continuous non-invasive blood pressure detection device, characterized in that: It includes a pressure sensor, a wristband, a host, a display, a data cable and a power supply, and is characterized in that the continuous non-invasive blood pressure detection device can provide data acquisition, program storage, and program running functions for implementing the blood pressure monitoring and positioning method described in any one of claims 1-4 when executing the program.

Images