Intelligent pressure regulation and control method and system for artery compression

Abstract

The invention discloses an intelligent pressure regulation and control method and system for artery compression, and the method comprises the steps: collecting the arterial blood pressure waveform and vascular elasticity parameters of a puncture part of a patient in real time, and the multi-dimensional pressure data of a pressure sensor; constructing an initial pressure parameter set containing a pressure distribution gradient and a vascular response coefficient through a dynamic feature extraction algorithm; carrying out collaborative analysis on the initial pressure parameter set and blood flow characteristic parameters obtained through blood flow monitoring, and generating a regulation and control matrix containing an optimal compression force range and a time decay function; performing multi-target optimization processing on the regulation and control matrix through a parallel optimization algorithm to generate a real-time pressure control instruction set; based on the deviation value of the real-time pressure execution data fed back by the user interface and the blood flow monitoring data, a grading alarm mechanism is triggered, and a compression strategy is automatically corrected. By utilizing the embodiment of the invention, accurate regulation and control of the artery compression process can be realized, and the hemostatic effect is optimized and the medical quality is improved on the premise of ensuring the safety of a patient.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology, and in particular to an intelligent pressure control method and system for arterial compression. Background Art

[0002] Arterial compression is commonly used in the medical field, particularly during surgical procedures and emergency treatment, to stop bleeding and reduce bleeding risk. However, traditional arterial compression methods often rely on the experience and judgment of medical personnel, making it difficult to ensure that the applied pressure is both effective and safe. Excessive pressure can lead to complications such as vascular damage and tissue ischemia, while too little pressure may fail to achieve the desired hemostatic effect. Therefore, there is an urgent need for an intelligent pressure regulation method to achieve precise control during arterial compression. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent pressure control method and system for arterial compression to address the deficiencies in the existing technology, enable precise control of the arterial compression process, optimize the hemostatic effect, and improve medical quality while ensuring patient safety.

[0004] An embodiment of the present application provides an intelligent pressure control method for arterial compression, the method comprising: output layer: compression force (mmHg) and time

[0005] The patient's arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor are collected in real time at the puncture site. An initial pressure parameter set, including the pressure distribution gradient and vascular response coefficient, is constructed using a dynamic feature extraction algorithm. The dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data.

[0006] Performing a collaborative analysis on the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, nonlinearly coupling the pressure gradient change rate with Doppler ultrasound blood flow velocity data, and generating a control matrix including an optimal compression force range and a time decay function;

[0007] Performing multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set to control the pressure of the compression device used for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining genetic algorithm and fuzzy control;

[0008] The deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface triggers a graded alarm mechanism and autonomously corrects the compression strategy. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude by which the deviation value exceeds the preset threshold, forming a closed-loop control link.

[0009] Optionally, the real-time acquisition of the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data of the pressure sensor at the patient's puncture site, and the construction of an initial pressure parameter set including a pressure distribution gradient and a vascular response coefficient by a dynamic feature extraction algorithm, wherein the dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data, including:

[0010] Based on the multi-dimensional pressure data collected by the pressure sensor array and the non-uniform sampling characteristics of the arterial blood pressure waveform, a dynamic time warping algorithm is used to perform spatiotemporal alignment of vascular elasticity parameters to generate a spatiotemporally synchronized vascular-pressure coupled dataset.

[0011] The spatiotemporally synchronized vascular-pressure coupled dataset is fed into a deep convolutional neural network. Multi-scale spatiotemporal convolution kernels are used to extract the correlation features between vascular wall deformation and pressure conduction. The pressure gradient field is then decomposed using a sparse coding algorithm to generate a pressure distribution gradient map containing local pressure extreme points and diffusion directions.

[0012] According to the pressure distribution gradient map, the attention mechanism is used to adaptively weight the vascular elasticity parameters. The vascular response coefficient is nonlinearly superimposed with the pressure gradient field through the tensor fusion layer to generate an initial pressure parameter set including the dynamic compensation threshold and elasticity correction factor.

[0013] Optionally, the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed, and the pressure gradient change rate is nonlinearly coupled with Doppler ultrasound blood flow velocity data to generate a control matrix containing an optimal compression force range and a time attenuation function, including:

[0014] Based on the pressure gradient change rate in the initial pressure parameter set, the Doppler ultrasound blood flow velocity data is subjected to time-frequency analysis to extract the blood flow pulse phase characteristics. The dynamic time warping algorithm is then used to perform nonlinear matching with the pressure gradient time series to generate a spatiotemporally aligned pressure-blood flow coupling tensor.

[0015] The pressure-blood flow coupling tensor is input into the high-order singular value decomposition model to extract the core feature matrix, and the feature matrix is ​​sparsely processed based on non-negative matrix decomposition to generate the coupling coefficient matrix of blood flow velocity and pressure gradient;

[0016] A radial basis function neural network is used to construct a nonlinear mapping from the coupling coefficient matrix to the compression force. The time decay function is introduced as a dynamic constraint term to generate a control matrix containing the optimal compression force range and the time decay factor.

[0017] Optionally, the control matrix is ​​subjected to multi-objective optimization processing by a parallel optimization algorithm to generate a real-time pressure control instruction set to perform pressure control on the compression device for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining genetic algorithm and fuzzy control, including:

[0018] Based on the compression intensity range of the control matrix, Latin hypercube sampling is used to generate the initial population. The individuals in the population are pre-screened using the probability distribution model of the patient's comfort threshold to obtain a set of candidate solutions that meet the multi-objective constraints.

[0019] The compression force error rate, blood flow index, and comfort threshold are used as fuzzy input variables. Their fuzzy sets are defined through a dynamic membership function. A nonlinear fitness function is constructed based on the Takagi-Sugeno model, and the fuzzy fitness score of each candidate solution is output.

[0020] A genetic algorithm with adaptive crossover and mutation probabilities is used to iteratively optimize candidate solutions. A fuzzy inference mechanism is combined to dynamically adjust the selection pressure and retain the Pareto frontier solution set. In each generation of evolution, a fuzzy controller is used to correct the local search direction to avoid premature convergence.

[0021] The optimized Pareto front solution set is non-dominated sorted, and the solution with the highest comprehensive score is selected by combining the elite retention strategy. Its logical consistency with the blood flow monitoring data is verified through the back propagation neural network, and mapped into a real-time pressure control instruction set to regulate the pressure of the compression device used for arterial compression.

[0022] Optionally, the deviation between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface triggers a graded alarm mechanism and autonomously modifies the compression strategy. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude by which the deviation exceeds a preset threshold, thereby forming a closed-loop control link, including:

[0023] Based on the real-time pressure execution data and blood flow monitoring data fed back by the user interface, the root mean square error and peak-to-peak value of the deviation value are calculated through a sliding time window. The deviation trend is predicted by combining the exponential smoothing method to generate a dynamic deviation window and confidence interval.

[0024] The confidence interval of the dynamic deviation window is compared with the preset threshold at multiple levels, the alarm level is determined by using a fuzzy matching algorithm, and the prior probability distribution of the alarm level is updated through a Bayesian network;

[0025] Based on the correction strategy corresponding to the alarm level, the reverse reinforcement learning algorithm is used to generate a control parameter adjustment plan. The digital twin model is used to virtually verify the corrected parameters using both blood flow and comfort indicators, and a safe correction parameter set is output.

[0026] The safety correction parameter set is synchronized to the control unit, the deviation rate between the execution effect and the expected target is monitored in real time, and the feedback gain of the closed-loop link is adjusted using an incremental PID control algorithm to ensure stability and adaptability during continuous intervention.

[0027] Another embodiment of the present application provides an intelligent pressure regulation system for arterial compression, the system comprising:

[0028] An acquisition module is used to collect, in real time, the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor at the patient's puncture site, and construct an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient using a dynamic feature extraction algorithm. The dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data.

[0029] An analysis module is configured to collaboratively analyze the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, perform nonlinear coupling between the pressure gradient change rate and the output layer: compression intensity (mmHg) and time-decay Doppler ultrasound blood flow velocity data, and generate a control matrix including an optimal compression intensity range and a time-decay function;

[0030] a processing module configured to perform multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set for pressure control of a compression device for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: a compression force error rate, a blood flow index, and a patient comfort threshold, and employs a hybrid optimization strategy combining a genetic algorithm and fuzzy control;

[0031] The feedback module is used to trigger a graded alarm mechanism and autonomously correct the compression strategy based on the deviation value between the real-time pressure execution data and blood flow monitoring data fed back by the user interface. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude of the deviation value exceeding the preset threshold, forming a closed-loop control link.

[0032] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0033] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0034] Compared with the existing technology, the present invention provides an intelligent pressure control method for arterial compression, which collects the arterial blood pressure waveform, vascular elasticity parameters and multi-dimensional pressure data of the pressure sensor at the patient's puncture site in real time, and constructs an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient through a dynamic feature extraction algorithm; the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed to generate a control matrix including the optimal compression force range and time attenuation function; the control matrix is ​​multi-objective optimized through a parallel optimization algorithm to generate a real-time pressure control instruction set; based on the deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, a hierarchical alarm mechanism is triggered and the compression strategy is autonomously corrected, thereby achieving precise control of the arterial compression process, optimizing the hemostatic effect and improving the quality of medical care while ensuring the safety of the patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A hardware structure block diagram of a computer terminal for an intelligent pressure control method for arterial compression provided by an embodiment of the present invention;

[0036] Figure 2 A schematic flow chart of an intelligent pressure control method for arterial compression provided by an embodiment of the present invention;

[0037] Figure 3 A schematic structural diagram of an intelligent pressure regulation system for arterial compression provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0039] The embodiment of the present invention first provides an intelligent pressure control method for arterial compression, which can be applied to electronic devices such as computer terminals, specifically ordinary computers.

[0040] The following describes it in detail by taking running on a computer terminal as an example. Figure 1The hardware structure block diagram of a computer terminal for an intelligent pressure control method for arterial compression provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0041] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can cause the processor to execute any one of the intelligent pressure control methods for arterial compression.

[0042] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0043] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any intelligent pressure control method for arterial compression.

[0044] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0045] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0046] See also Figure 2 , an embodiment of the present invention provides an intelligent pressure control method for arterial compression, which may include the following steps:

[0047] S201, collecting in real time the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor at the patient's puncture site, and constructing an initial pressure parameter set including a pressure distribution gradient and a vascular response coefficient using a dynamic feature extraction algorithm, wherein the dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data;

[0048] Specifically, based on the multi-dimensional pressure data collected by the pressure sensor array and the non-uniform sampling characteristics of the arterial blood pressure waveform, a dynamic time warping algorithm can be used to perform spatiotemporal alignment processing on the vascular elasticity parameters to generate a spatiotemporally synchronized vascular-pressure coupling dataset.

[0049] The pressure sensor array collects multidimensional pressure data (including vertical pressure, shear force, and pressure distribution range) at the patient's puncture site at a sampling frequency of 100 Hz. Simultaneously, the photoelectric pulse wave sensor acquires arterial blood pressure waveforms (systolic pressure, diastolic pressure, and pulse pressure difference) using non-uniform sampling (at intervals of 0.5 to 2 seconds). Because the time axes of the two data are not synchronized, dynamic time warping (DTW) is used to align the timing:

[0050] Time Warping Path Search: This function calculates the cumulative distance matrix between the pressure data sequence (length N = 500) and the blood pressure waveform sequence (length M = 30), and uses dynamic programming to find the optimal alignment path. For example, if the peak of a certain pressure data segment at t = 3.2 seconds corresponds to the peak of the blood pressure waveform at t = 3.5 seconds, DTW automatically warps the time axis to achieve alignment.

[0051] Interpolation compensation: Cubic spline interpolation is performed on the non-uniformly sampled blood pressure waveform to generate a 100Hz sequence synchronized with the pressure data. For example, a blood pressure waveform of 120 / 80 mmHg at t=3 seconds and 118 / 78 mmHg at t=4 seconds will be interpolated to 119 / 79 mmHg at t=3.2 seconds.

[0052] Vascular elasticity parameter fusion: The vascular elastic modulus (in kPa) obtained by ultrasound elastography (such as ARFI, acoustic radiation force impulse imaging) is mapped to the aligned time axis to form a spatiotemporally synchronized 3D dataset (time × pressure × elastic modulus).

[0053] The resulting vascular-pressure coupling dataset contains the following dimensions: Layers: compression force (mmHg) and time decay

[0054] Time axis: 0~5 seconds, step size 0.01 second;

[0055] Spatial axis: 5×5 grid coordinates of the sensor array;

[0056] Characteristic axes: pressure value (0-300 mmHg), elastic modulus (5-50 kPa), blood pressure waveform (systolic pressure / diastolic pressure).

[0057] The spatiotemporally synchronized vascular-pressure coupled dataset is fed into a deep convolutional neural network. Multi-scale spatiotemporal convolution kernels are used to extract the correlation features between vascular wall deformation and pressure conduction. The pressure gradient field is then decomposed using a sparse coding algorithm to generate a pressure distribution gradient map containing local pressure extreme points and diffusion directions.

[0058] The deep convolutional neural network (CNN) uses a multi-branch structure to process spatiotemporal features of different scales:

[0059] The large-scale branch uses a 5×5×3 (space × time) convolution kernel to capture the overall deformation trend of the blood vessel wall. For example, when the compression device applies pressure, the blood vessel wall changes from a circular shape to an elliptical shape. This branch detects the correlation between the deformation amplitude and the pressure distribution.

[0060] The mesoscale branch uses a 3×3×5 convolution kernel to analyze the local pressure transmission path. For example, when the pressure at a sensor node rises from 150 mmHg to 180 mmHg, the pressure at the adjacent node follows suit 0.2 seconds later, forming a transmission delay pattern.

[0061] Small-scale branch: 1×1×7 convolution kernel, focusing on high-frequency pressure fluctuations. For example, it can detect tiny 10Hz pressure oscillations caused by vascular elastic recoil.

[0062] Feature fusion and sparse coding:

[0063] The three-branch output is concatenated into a 256-dimensional feature vector, which is then fed into a K-SVD sparse encoder to learn an overcomplete dictionary (100 basis vectors). For example, a certain basis vector corresponds to the "pressure diffuses from the center to the edge" pattern.

[0064] Decompose the pressure gradient field to generate a pressure distribution gradient map, including:

[0065] Local pressure extreme point: marks the area where the pressure exceeds 250 mmHg (highlighted in red);

[0066] Diffusion direction arrow: shows the pressure conduction path (such as diffusion from coordinates (2,3) to (3,4)).

[0067] Example: At a vascular bifurcation, the map shows that the extreme pressure point is located upstream of the bifurcation, and the diffusion direction extends along the branching vessel, which is consistent with the deformation area of ​​ultrasound elastography.

[0068] According to the pressure distribution gradient map, the attention mechanism is used to adaptively weight the vascular elasticity parameters. The vascular response coefficient is nonlinearly superimposed with the pressure gradient field through the tensor fusion layer to generate an initial pressure parameter set including the dynamic compensation threshold and elasticity correction factor.

[0069] The attention mechanism adopts a channel-space dual-path structure:

[0070] Channel attention: Global average pooling (GAP) is used to calculate the importance weight of each feature channel. For example, the elastic modulus channel has a weight of 0.8 in vascular sclerosis areas (modulus > 30 kPa), while the weight in normal areas (modulus < 20 kPa) is 0.2.

[0071] Spatial attention: A sigmoid function is used to generate a spatial mask to highlight high-pressure areas (e.g., the weight of grid points with pressure > 200 mmHg is set to 1, and the rest are 0.3).

[0072] The tensor fusion layer performs the following operations:

[0073] Calculation of vascular response coefficient: Based on the relationship between elastic modulus (E) and pressure gradient (▽P), the response coefficient matrix is ​​generated: R = α•E + β•▽P, where α = 0.6 and β = 0.4 are empirical coefficients calibrated through experiments;

[0074] Nonlinear superposition: The response coefficient matrix and pressure gradient map are input into the residual block (including two layers of full connection and ReLU activation), and the fused feature tensor is output.

[0075] Dynamic compensation threshold and elastic correction factor:

[0076] Dynamic compensation threshold: adjusts the upper limit of compression intensity based on the diffusion speed of the current pressure extreme point (e.g. 0.5mm / s). For example, when diffusion is too fast, the threshold is reduced from 250mmHg to 230mmHg to prevent tissue damage.

[0077] Elasticity correction factor: Based on the spatial distribution of vascular elastic modulus (e.g., factor = 1.2 for sclerosis area and 1.0 for normal area), it is used for weighted correction of pressure instructions in subsequent optimization algorithms.

[0078] The final structure of the initial pressure parameter set is:

[0079] Parameter 1: Dynamic compensation threshold (200~300mmHg);

[0080] Parameter 2: elastic correction factor matrix (5×5 grid);

[0081] Parameter 3: Pressure diffusion direction vector field.

[0082] S202, collaboratively analyzing the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, nonlinearly coupling the pressure gradient change rate with Doppler ultrasound blood flow velocity data, and generating a control matrix including an optimal compression force range and a time decay function;

[0083] Specifically, the Doppler ultrasound blood flow velocity data can be subjected to time-frequency analysis based on the pressure gradient change rate in the initial pressure parameter set to extract the blood flow pulse phase characteristics. This can then be nonlinearly matched with the pressure gradient time series using a dynamic time warping algorithm to generate a spatiotemporally aligned pressure-blood flow coupling tensor.

[0084] Time-frequency analysis:

[0085] Doppler ultrasound blood flow velocity data were acquired at a sampling rate of 100 frames per second, with each frame containing the spatial distribution of blood flow velocity (e.g., 0.1 mm resolution along the vessel axis). Time-frequency analysis was performed using a short-time Fourier transform (STFT) with a window length of 50 ms (corresponding to 5 frames of data) and an overlap ratio of 75%. The Morlet wavelet transform was used to extract the phase characteristics of the blood flow pulse (e.g., the dominant frequency is concentrated in the 1-5 Hz beat period), and the instantaneous frequency and phase angle were calculated at each time point. For example, a phase angle of 120° at a certain moment indicates that blood flow is in the transition phase from systole to diastole.

[0086] Dynamic Time Warping (DTW):

[0087] The pressure gradient rate of change (for example, the time series from 10 mmHg / s to 25 mmHg / s) is nonlinearly aligned with the blood flow velocity time series. The DTW path cost function uses Euclidean distance and introduces a slope constraint (limiting the path curvature to no more than 45°) to match the delayed characteristics of physiological signals (for example, compression-induced blood flow changes lag by approximately 0.5 seconds). After alignment, each peak point of the pressure gradient (for example, 20 mmHg / s) is precisely associated with the corresponding phase of blood flow (for example, the onset of diastole).

[0088] Coupled tensor construction:

[0089] The resulting three-dimensional tensor has dimensions time × space × features, where:

[0090] Time axis: aligned sampling points (500 in total, time span 5 seconds);

[0091] Spatial axis: 10 regions along the length of the vessel (each 2 cm);

[0092] Characteristic axes: pressure gradient (mmHg / s), blood flow velocity (cm / s), phase angle (degrees).

[0093] For example, the tensor element T[250,5,2]=18.5cm / s indicates that at 2.5 seconds, the blood flow velocity in the fifth blood vessel is 18.5cm / s.

[0094] The pressure-blood flow coupling tensor is input into the high-order singular value decomposition model to extract the core feature matrix, and the feature matrix is ​​sparsely processed based on non-negative matrix decomposition to generate the coupling coefficient matrix of blood flow velocity and pressure gradient;

[0095] Perform modal decomposition on a 3D tensor:

[0096] Time mode: extract the periodic component with a main frequency of 2 Hz (corresponding to a heart rate of 60-120 beats / minute);

[0097] Spatial mode: separates the spatial distribution pattern of the proximal end (high flow velocity area) and the distal end (low flow velocity area) of the blood vessel;

[0098] Characteristic mode: Distinguish between pressure-dominated (such as the initial stage of compression) and blood flow-dominated (compression steady-state period) characteristics.

[0099] The top three core features (contribution rate > 85%) are retained to generate a core feature matrix (dimension 3×3×3). For example, a core matrix C[1,2,3]=0.92 indicates a strong correlation between temporal mode 1, spatial mode 2, and eigenmode 3.

[0100] Non-negative Matrix Factorization (NMF):

[0101] The core feature matrix is ​​expanded into a two-dimensional matrix (9×3) and decomposed into a basis matrix W (9×2) and a coefficient matrix H (2×3) using the NMF algorithm. The sparsity constraint is set to λ = 0.1, forcing the basis vector to retain only key features:

[0102] Basis matrix W: The first column represents the “compression-blood flow inhibition” mode (e.g., blood flow velocity decreases by 30% when the pressure gradient is >15 mmHg / s);

[0103] Basis matrix H: The second row represents the "vascular elasticity compensation" mode (such as blood vessel response coefficient calculation: when the elasticity parameter of the elastic tube is >0.8, the compression efficiency is increased by 20%).

[0104] Coupling coefficient matrix generation:

[0105] The sparsified feature matrix is ​​reconstructed through matrix multiplication W×H, retaining non-negative values ​​and normalized to [0,1]. For example, the element [2,1]=0.75 in the coupling coefficient matrix means that in the second blood vessel segment, for every 1 mmHg / s increase in the pressure gradient, the blood flow velocity decreases by 0.75 cm / s.

[0106] A radial basis function neural network is used to construct a nonlinear mapping from the coupling coefficient matrix to the compression force. The time decay function is introduced as a dynamic constraint term to generate a control matrix containing the optimal compression force range and the time decay factor.

[0107] Radial Basis Function Neural Network (RBFNN) Design:

[0108] Input layer: the expanded vector of the coupling coefficient matrix (length 30, containing 3 coefficients for 10 vascular regions);

[0109] Hidden layer: 20 Gaussian kernel functions, the centers are initialized by K-means clustering (number of clusters K = 20), and the width parameter σ = 0.5;

[0110] Output layer: compression force (mmHg) and time decay factor τ .

[0111] Non-linear mapping training:

[0112] Use historical data (1000 sets of samples) for supervised learning, and the loss function is the mean square error (MSE) plus the time decay constraint:

[0113]

[0114] Among them, the ideal attenuation factor τ_ideal= (This corresponds to the compression force decaying to 37% of the initial value within 10 seconds.) The weights were optimized using the Levenberg-Marquardt algorithm, and the test set error was <5% after 50 iterations.

[0115] Control matrix generation:

[0116] Optimal compression range: The output layer provides recommended pressure for each vascular region (e.g., 40-60 mmHg for the proximal region and 20-30 mmHg for the distal region);

[0117] Time decay function: Exponential form , for example, the initial pressure P0=50mmHg, τ= , the pressure drops to 30 mmHg after 5 seconds.

[0118] Sample Application:

[0119] When the coupling coefficient matrix shows that there is a high pressure gradient-low blood flow conflict in the third segment of the blood vessel (coefficient [3,1] = 0.9), the RBFNN outputs the compression force of this area as 55 mmHg, and the time decay factor τ = , ensuring gradual decompression to the safety threshold within 8 seconds.

[0120] S203, performing multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set to control the pressure of the compression device for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining a genetic algorithm and fuzzy control;

[0121] Specifically, Latin hypercube sampling can be used to generate an initial population based on the compression intensity range of the control matrix, and the probability distribution model of the patient's comfort threshold can be used to pre-screen the individuals in the population to obtain a set of candidate solutions that meet the multi-objective constraints.

[0122] Latin Hypercube Sampling (LHS) is used to uniformly generate initial candidate solutions in the multidimensional solution space across the compression intensity range. Assume that the compression intensity defined by the control matrix ranges from 30 to 100 mmHg (millimeters of mercury), the blood flow index target is 0.8 to 1.2 (dimensionless), and the comfort threshold is 1 to 5 (with 1 being the most comfortable). LHS is used to generate 1,000 initial solutions in the three-dimensional space (compression intensity, blood flow index, and comfort), ensuring that the sample intervals in each dimension do not overlap. For example, the compression intensity dimension is divided into 10 subintervals (30-37, 38-45, etc.), and one sample is randomly selected from each subinterval.

[0123] The probability distribution model for patient comfort thresholds was constructed based on historical data, assuming a beta distribution (α=2, β=5) with a preference for low levels (a 70% probability of comfort levels 1-2). During pre-screening, solutions with comfort levels >3 were eliminated, as were solutions with compression forces exceeding the elastic limit of the vessel (e.g., >90 mmHg, posing a risk of vascular collapse). For example, 600 candidate solutions were retained from an initial pool of 1,000 solutions to ensure compliance with physiological safety margins.

[0124] Example candidate solution:

[0125] Solution A: Compression force 65 mmHg, blood flow index 0.95, comfort level 2;

[0126] Solution B: Compression intensity 72 mmHg, blood flow index 1.1, comfort level 3 (eliminated due to excessive comfort level).

[0127] The compression force error rate, blood flow index, and comfort threshold are used as fuzzy input variables. Their fuzzy sets are defined through a dynamic membership function. A nonlinear fitness function is constructed based on the Takagi-Sugeno model, and the fuzzy fitness score of each candidate solution is output.

[0128] Fuzzy input variable definition:

[0129] Compression force error rate (|actual value - target value| / target value):

[0130] Membership function: low error (0%~5%, trapezoidal function), medium error (5%~15%, triangle), high error (>15%, S-shaped function);

[0131] Blood flow index (target 1.0±0.2):

[0132] Membership function: blocking (<0.8, Z-type function), normal (0.8~1.2, Gaussian function), excessive (>1.2, S-type function);

[0133] Comfort threshold:

[0134] Membership function: comfortable (level 1-2, trapezoidal function), general (level 3, triangle), uncomfortable (level 4-5, S-shaped function).

[0135] Takagi-Sugeno model rule base:

[0136] Rule 1: If the error rate is low, blood flow is normal, and comfort is high → fitness score = 0.7 × error membership + 0.2 × blood flow membership + 0.1 × comfort membership;

[0137] Rule 2: If the error rate is medium, blood flow is obstructed, and comfort is average → score = 0.5 × error + 0.4 × blood flow + 0.1 × comfort;

[0138] There are 27 rules in total (3 variables × 3 states).

[0139] Dynamic fitness calculation:

[0140] For each candidate solution, the corresponding rules are activated based on their parameter values, and a weighted sum is calculated to obtain a fuzzy score. For example, the parameters of solution A activate rule 1, resulting in a score of 0.85 (error rate 3% → membership 0.9, blood flow 0.95 → 0.8, comfort 2 → 0.7).

[0141] A genetic algorithm with adaptive crossover and mutation probabilities is used to iteratively optimize candidate solutions. A fuzzy inference mechanism is combined to dynamically adjust the selection pressure and retain the Pareto frontier solution set. In each generation of evolution, a fuzzy controller is used to correct the local search direction to avoid premature convergence.

[0142] Genetic algorithm parameter adaptation strategy:

[0143] Crossover probability Pc: Initially 0.8, dynamically adjusted based on population diversity (genetic diversity). If diversity decreases (dissimilarity < 30%), Pc is increased to 0.9 to enhance exploration.

[0144] Mutation probability Pm: Initially 0.05. If the optimal solution does not improve after three consecutive generations, Pm is increased to 0.15 to introduce disturbance.

[0145] Fuzzy reasoning adjusts the search direction:

[0146] Input: contemporary population fitness variance (reflecting convergence), and the improvement of the optimal solution;

[0147] Output: Local search direction weight (e.g., bias toward compression strength or blood flow index). For example, if the variance is low and the improvement is small, increase the search weight for comfort.

[0148] Evolution process:

[0149] Selection: Tournament selection (size 3), selecting 300 parents from 600 candidate solutions;

[0150] Crossover: simulates binary crossover (SBX) to generate 300 offspring;

[0151] Mutation: Gaussian mutation (standard deviation = 5% parameter range), perturbing offspring with probability Pm;

[0152] Merge: There are 600 solutions in total for the parent and child generations. They are sorted by fuzzy score and the top 300 are retained for the next generation.

[0153] Pareto front preservation:

[0154] At the end of each generation, the non-dominated sorting algorithm of NSGA-II is used to select the top 50 solutions as elite solutions to ensure multi-objective equilibrium. For example, the Pareto solution set of the 10th generation contains:

[0155] Solution X: Error rate 4%, blood flow 1.0, comfort level 1;

[0156] Solution Y: Error rate 2%, blood flow 0.9, comfort level 2.

[0157] The optimized Pareto front solution set is non-dominated sorted, and the solution with the highest comprehensive score is selected by combining the elite retention strategy. Its logical consistency with the blood flow monitoring data is verified through the back propagation neural network, and mapped into a real-time pressure control instruction set to regulate the pressure of the compression device used for arterial compression.

[0158] Non-dominated sorting and elite screening:

[0159] Fast non-dominated sorting: divide the Pareto solution set into multiple frontier layers (Front 1 is optimal, Front 2 is second-best, etc.);

[0160] Crowding calculation: Calculate the crowding distance between solutions in the target space (error rate, blood flow, comfort), and retain sparsely distributed solutions;

[0161] Elite selection: Select the 20 most crowded solutions from Front 1 to ensure diversity and optimality.

[0162] Back Propagation Neural Network (BPNN) Verification:

[0163] Network structure: 3 input layers (compression intensity, blood flow, comfort), 2 hidden layers (16 nodes, ReLU activation), 1 output layer (blood flow prediction value, Sigmoid activation);

[0164] Training data: 5000 sets of historical compression data, loss function is MAE (mean absolute error);

[0165] Verification logic: Candidate solutions are fed into the BPNN, and the predicted blood flow values ​​are compared with the target value (1.0). Solutions with an error > 10% are discarded. For example, solution X predicts a blood flow of 0.98 with an error of 2%, thus passing verification.

[0166] Instruction set generation:

[0167] Finally, 10 optimal solutions were selected and dynamically selected according to scenario requirements:

[0168] Acute bleeding scenario: prioritize the solution with the lowest error rate (e.g., 1% error);

[0169] Conventional compression scenario: Select the solution with the best comfort level (Comfort Level 1).

[0170] The command is sent to the compression device via the CAN bus to control the air pump pressure and compression duration (e.g. 65 mmHg for 15 minutes).

[0171] S204, based on the deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, triggers the graded alarm mechanism and autonomously corrects the compression strategy, wherein the graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude of the deviation value exceeding the preset threshold, forming a closed-loop control link.

[0172] Specifically, the real-time pressure execution data and blood flow monitoring data fed back by the user interface can be used to calculate the root mean square error and peak-to-peak value of the deviation value through a sliding time window, and the deviation trend can be predicted by combining the exponential smoothing method to generate a dynamic deviation window and confidence interval.

[0173] The sliding time window is used to analyze the deviation between pressure execution data and blood flow monitoring data in real time. The window size is set to 5 seconds (corresponding to 50 sampling points, sampling frequency 10Hz), with a step size of 1 second. The following indicators are calculated for each window:

[0174] Root Mean Square Error (RMSE): measures the overall magnitude of pressure deviation. For example, if the pressure execution value in the current window is [120, 125, 130] mmHg and the monitored value is [115, 120, 125] mmHg, then .

[0175] Peak-to-Peak: Reflects the maximum fluctuation range of the deviation. For example, if the maximum deviation in the window is +8 mmHg and the minimum is -3 mmHg, the peak-to-peak value is 11 mmHg.

[0176] Exponential smoothing is used to predict deviation trends. The smoothing coefficient α is set to 0.3 (historical data weighting 70%, current data weighting 30%), and the prediction formula is:

[0177]

[0178] is the smoothed value at time t, Current deviation value. For example, if the current deviation is 5 mmHg and the previous smoothed value is 4 mmHg, then the new smoothed value = 0.3 × 5 + 0.7 × 4 = 4.3 mmHg.

[0179] Dynamic deviation window confidence intervals are generated by bootstrap resampling:

[0180] Randomly draw 100 samples from the current window (with replacement);

[0181] Calculate the RMSE mean and standard deviation of each group of samples;

[0182] The 95% quantile is used as the upper limit of the confidence interval. For example, if the mean = 5 mmHg and the standard deviation = 1.2 mmHg, the confidence interval is [3.5, 6.5] mmHg.

[0183] The confidence interval of the dynamic deviation window is compared with the preset threshold at multiple levels, the alarm level is determined by using a fuzzy matching algorithm, and the prior probability distribution of the alarm level is updated through a Bayesian network;

[0184] Preset threshold levels (unit: mmHg):

[0185] Level 1 alarm: deviation ≥ 10;

[0186] Second level alarm: 5≤deviation<10;

[0187] Level 3 alarm: 3≤deviation<5;

[0188] Normal range: Deviation <3.

[0189] The fuzzy matching algorithm uses a trapezoidal membership function to handle the fuzziness of the threshold boundary:

[0190] Level 1 alarm membership: when the deviation is ≥8, the membership is 1, and when the deviation is 6≤<8, the membership decreases linearly to 0;

[0191] Secondary alarm membership: when the deviation is ≥5, the membership is 1; when the deviation is 3≤<5, the membership decreases linearly to 0.

[0192] For example, when the deviation = 7 mmHg, the first-level membership = 0.5, the second-level membership = 0.5, and a mixed alarm is triggered.

[0193] Bayesian Network Update:

[0194] Node definition: The parent node is the historical alarm level (such as the number of level 1 alarms in the past 10 minutes), and the child node is the current alarm level;

[0195] Conditional Probability Table (CPT): If there have been three Level 1 alarms in the past 10 minutes, the probability of a Level 1 alarm is increased to 70%;

[0196] Posterior Update: Calculates the likelihood function based on real-time data and combines it with the prior probability to output the posterior distribution. For example, if the current deviation is 9 mmHg and there have been two Level 1 alarms in the past, the posterior probability of a Level 1 alarm is 65%.

[0197] Based on the correction strategy corresponding to the alarm level, the reverse reinforcement learning algorithm is used to generate a control parameter adjustment plan. The digital twin model is used to virtually verify the corrected parameters using both blood flow and comfort indicators, and a safe correction parameter set is output.

[0198] Inverse Reinforcement Learning (IRL):

[0199] Expert demonstration: Collect 100 sets of historical successful correction examples (e.g. reducing pressure by 15% at level 1 alarm);

[0200] Reward function learning: Using the maximum entropy IRL model to learn the implicit reward function, we discovered that the reduction in pressure is positively correlated with the deviation value.

[0201] Policy generation: Based on the learned reward function, the Q-learning algorithm generates an adjustment plan. For example, if the current deviation is 12 mmHg, the policy recommends "reducing the pressure by 18% and extending the pressure for 2 minutes."

[0202] Digital Twin Verification:

[0203] Blood flow permeability index: simulated corrected radial artery blood flow velocity (normal value > 20 cm / s);

[0204] Comfort index: The patient pain score (0-10 points, threshold <3) was calculated based on the pressure distribution model.

[0205] For example, a modified plan reduced the compression force from 150 mmHg to 130 mmHg, increased the blood flow velocity to 22 cm / s, and the pain score was 2.5 points, which was verified.

[0206] The safety correction parameter set is synchronized to the control unit, the deviation rate between the execution effect and the expected target is monitored in real time, and the feedback gain of the closed-loop link is adjusted using an incremental PID control algorithm to ensure stability and adaptability during continuous intervention.

[0207] Incremental PID control algorithm parameter settings:

[0208] Proportional coefficient ( ): 0.8 (fast response deviation);

[0209] Integration coefficient ( ): 0.05 (eliminating steady-state error);

[0210] Differential coefficient ( ): 0.2 (suppress overshoot).

[0211] Control quantity calculation:

[0212]

[0213] in, is the current deviation. For example, if the current deviation is 4 mmHg and the previous deviation is 5 mmHg, then Δu = 0.8 × 4 + 0.05 × 9 + 0.2 × (-1) = 3.2 + 0.45 - 0.2 = 3.45 mmHg.

[0214] Feedback gain dynamic adjustment:

[0215] Overshoot suppression: If the deviation fluctuation is >±2mmHg after 3 consecutive controls, Kp will be automatically reduced to 0.6;

[0216] Steady-state optimization: If the deviation remains <1 mmHg for more than 1 minute, increase Ki to 0.1.

[0217] Technical Effects and Examples

[0218] Graded alarm response: When the deviation value reaches 8mmHg, fuzzy matching triggers a level 1 alarm (sound and light alarm + red flash on the interface). With a Bayesian network confidence level of 85%, the system initiates a pressure reduction strategy within 200ms.

[0219] Digital twin verification: The modified solution "compression force 130mmHg" was simulated to show a blood flow velocity of 25cm / s and a pain score of 2 points, which were better than the safety threshold;

[0220] Closed-loop stability: Incremental PID controls the steady-state deviation within ±1mmHg, and the adaptive adjustment period is <500ms.

[0221] It can be seen that the arterial blood pressure waveform, vascular elasticity parameters and multi-dimensional pressure data of the pressure sensor at the patient's puncture site are collected in real time, and an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient is constructed through a dynamic feature extraction algorithm; the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed to generate a control matrix including the optimal compression force range and time attenuation function; the control matrix is ​​multi-objective optimized through a parallel optimization algorithm to generate a real-time pressure control instruction set; based on the deviation value of the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, a hierarchical alarm mechanism is triggered and the compression strategy is automatically corrected, thereby achieving precise control of the arterial compression process, optimizing the hemostasis effect and improving the quality of medical care while ensuring the safety of the patient.

[0222] Another embodiment of the present invention provides an intelligent pressure control system for arterial compression, see Figure 3 , the system may include:

[0223] Acquisition module 301 is used to collect, in real time, the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor at the patient's puncture site, and construct an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient using a dynamic feature extraction algorithm. The dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data.

[0224] An analysis module 302 is configured to collaboratively analyze the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, perform nonlinear coupling between the pressure gradient change rate and Doppler ultrasound blood flow velocity data, and generate a control matrix including an optimal compression force range and a time decay function;

[0225] a processing module 303 configured to perform multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set for controlling the pressure of the compression device used for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and employs a hybrid optimization strategy combining a genetic algorithm and fuzzy control;

[0226] Feedback module 304 is used to trigger a graded alarm mechanism and autonomously correct the compression strategy based on the deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude of the deviation value exceeding the preset threshold, thereby forming a closed-loop control link.

[0227] It can be seen that the arterial blood pressure waveform, vascular elasticity parameters and multi-dimensional pressure data of the pressure sensor at the patient's puncture site are collected in real time, and an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient is constructed through a dynamic feature extraction algorithm; the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed to generate a control matrix including the optimal compression force range and time attenuation function; the control matrix is ​​multi-objective optimized through a parallel optimization algorithm to generate a real-time pressure control instruction set; based on the deviation value of the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, a hierarchical alarm mechanism is triggered and the compression strategy is automatically corrected, thereby achieving precise control of the arterial compression process, optimizing the hemostasis effect and improving the quality of medical care while ensuring the safety of the patient.

[0228] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0229] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps:

[0230] S201, collecting in real time the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor at the patient's puncture site, and constructing an initial pressure parameter set including a pressure distribution gradient and a vascular response coefficient using a dynamic feature extraction algorithm, wherein the dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data;

[0231] S202, collaboratively analyzing the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, nonlinearly coupling the pressure gradient change rate with Doppler ultrasound blood flow velocity data, and generating a control matrix including an optimal compression force range and a time decay function;

[0232] S203, performing multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set to control the pressure of the compression device for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining a genetic algorithm and fuzzy control;

[0233] S204, based on the deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, triggers the graded alarm mechanism and autonomously corrects the compression strategy, wherein the graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude of the deviation value exceeding the preset threshold, forming a closed-loop control link.

[0234] It can be seen that the arterial blood pressure waveform, vascular elasticity parameters and multi-dimensional pressure data of the pressure sensor at the patient's puncture site are collected in real time, and an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient is constructed through a dynamic feature extraction algorithm; the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed to generate a control matrix including the optimal compression force range and time attenuation function; the control matrix is ​​multi-objective optimized through a parallel optimization algorithm to generate a real-time pressure control instruction set; based on the deviation value of the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, a hierarchical alarm mechanism is triggered and the compression strategy is automatically corrected, thereby achieving precise control of the arterial compression process, optimizing the hemostasis effect and improving the quality of medical care while ensuring the safety of the patient.

[0235] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0236] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0237] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0238] S201, collecting in real time the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor at the patient's puncture site, and constructing an initial pressure parameter set including a pressure distribution gradient and a vascular response coefficient using a dynamic feature extraction algorithm, wherein the dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data;

[0239] S202, collaboratively analyzing the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, nonlinearly coupling the pressure gradient change rate with Doppler ultrasound blood flow velocity data, and generating a control matrix including an optimal compression force range and a time decay function;

[0240] S203, performing multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set to control the pressure of the compression device for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining a genetic algorithm and fuzzy control;

[0241] S204, based on the deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, triggers the graded alarm mechanism and autonomously corrects the compression strategy, wherein the graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude of the deviation value exceeding the preset threshold, forming a closed-loop control link.

[0242] It can be seen that the arterial blood pressure waveform, vascular elasticity parameters and multi-dimensional pressure data of the pressure sensor at the patient's puncture site are collected in real time, and an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient is constructed through a dynamic feature extraction algorithm; the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed to generate a control matrix including the optimal compression force range and time attenuation function; the control matrix is ​​multi-objective optimized through a parallel optimization algorithm to generate a real-time pressure control instruction set; based on the deviation value of the real-time pressure execution data and the blood flow monitoring data fed back by the user interface, a hierarchical alarm mechanism is triggered and the compression strategy is automatically corrected, thereby achieving precise control of the arterial compression process, optimizing the hemostasis effect and improving the quality of medical care while ensuring the safety of the patient.

[0243] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. An intelligent pressure control method for arterial compression, characterized in that: The method comprises: The patient's arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor are collected in real time at the puncture site. An initial pressure parameter set, including the pressure distribution gradient and vascular response coefficient, is constructed using a dynamic feature extraction algorithm. The dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data. Performing a collaborative analysis on the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, nonlinearly coupling the pressure gradient change rate with Doppler ultrasound blood flow velocity data, and generating a control matrix including an optimal compression force range and a time decay function; Performing multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set to control the pressure of the compression device used for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining genetic algorithm and fuzzy control; The deviation value between the real-time pressure execution data and the blood flow monitoring data fed back by the user interface triggers a graded alarm mechanism and autonomously corrects the compression strategy. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude by which the deviation value exceeds the preset threshold, forming a closed-loop control link.

2. The method according to claim 1, characterized in that The method collects arterial blood pressure waveforms, vascular elasticity parameters, and multi-dimensional pressure data of pressure sensors at the patient's puncture site in real time, and constructs an initial pressure parameter set including pressure distribution gradients and vascular response coefficients using a dynamic feature extraction algorithm. The dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data, including: Based on the multi-dimensional pressure data collected by the pressure sensor array and the non-uniform sampling characteristics of the arterial blood pressure waveform, a dynamic time warping algorithm is used to perform spatiotemporal alignment of vascular elasticity parameters to generate a spatiotemporally synchronized vascular-pressure coupled dataset. The spatiotemporally synchronized vascular-pressure coupled dataset is fed into a deep convolutional neural network. Multi-scale spatiotemporal convolution kernels are used to extract the correlation features between vascular wall deformation and pressure conduction. The pressure gradient field is then decomposed using a sparse coding algorithm to generate a pressure distribution gradient map containing local pressure extreme points and diffusion directions. According to the pressure distribution gradient map, the attention mechanism is used to adaptively weight the vascular elasticity parameters. The vascular response coefficient is nonlinearly superimposed with the pressure gradient field through the tensor fusion layer to generate an initial pressure parameter set including the dynamic compensation threshold and elasticity correction factor.

3. The method according to claim 2, characterized in that The initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring are collaboratively analyzed, and the pressure gradient change rate is nonlinearly coupled with Doppler ultrasound blood flow velocity data to generate a control matrix containing an optimal compression force range and a time attenuation function, including: Based on the pressure gradient change rate in the initial pressure parameter set, the Doppler ultrasound blood flow velocity data is subjected to time-frequency analysis to extract the blood flow pulse phase characteristics. The dynamic time warping algorithm is then used to perform nonlinear matching with the pressure gradient time series to generate a spatiotemporally aligned pressure-blood flow coupling tensor. The pressure-blood flow coupling tensor is input into the high-order singular value decomposition model to extract the core feature matrix, and the feature matrix is ​​sparsely processed based on non-negative matrix decomposition to generate the coupling coefficient matrix of blood flow velocity and pressure gradient; A radial basis function neural network is used to construct a nonlinear mapping from the coupling coefficient matrix to the compression force. The time decay function is introduced as a dynamic constraint term to generate a control matrix containing the optimal compression force range and the time decay factor.

4. The method according to claim 3, characterized in that The control matrix is ​​subjected to multi-objective optimization processing by a parallel optimization algorithm to generate a real-time pressure control instruction set to perform pressure control on a compression device for arterial compression. The parallel optimization algorithm simultaneously optimizes three objective functions, namely, compression force error rate, blood flow index, and patient comfort threshold, and adopts a hybrid optimization strategy combining genetic algorithm and fuzzy control, including: Based on the compression intensity range of the control matrix, Latin hypercube sampling is used to generate the initial population. The individuals in the population are pre-screened using the probability distribution model of the patient's comfort threshold to obtain a set of candidate solutions that meet the multi-objective constraints. The compression force error rate, blood flow index, and comfort threshold are used as fuzzy input variables. Their fuzzy sets are defined through a dynamic membership function. A nonlinear fitness function is constructed based on the Takagi-Sugeno model, and the fuzzy fitness score of each candidate solution is output. A genetic algorithm with adaptive crossover and mutation probabilities is used to iteratively optimize candidate solutions. A fuzzy inference mechanism is combined to dynamically adjust the selection pressure and retain the Pareto frontier solution set. In each generation of evolution, a fuzzy controller is used to correct the local search direction to avoid premature convergence. The optimized Pareto front solution set is non-dominated sorted, and the solution with the highest comprehensive score is selected by combining the elite retention strategy. Its logical consistency with the blood flow monitoring data is verified through the back propagation neural network, and mapped into a real-time pressure control instruction set to regulate the pressure of the compression device used for arterial compression.

5. The method according to claim 4, characterized in that The deviation between the real-time pressure execution data and the blood flow monitoring data based on user interface feedback triggers a graded alarm mechanism and autonomously modifies the compression strategy. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit based on the magnitude by which the deviation exceeds the preset threshold, forming a closed-loop control link, including: Based on the real-time pressure execution data and blood flow monitoring data fed back by the user interface, the root mean square error and peak-to-peak value of the deviation value are calculated through a sliding time window. The deviation trend is predicted by combining the exponential smoothing method to generate a dynamic deviation window and confidence interval. The confidence interval of the dynamic deviation window is compared with the preset threshold at multiple levels, the alarm level is determined by using a fuzzy matching algorithm, and the prior probability distribution of the alarm level is updated through a Bayesian network; Based on the correction strategy corresponding to the alarm level, the reverse reinforcement learning algorithm is used to generate a control parameter adjustment plan. The digital twin model is used to virtually verify the corrected parameters using both blood flow and comfort indicators, and a safe correction parameter set is output. The safety correction parameter set is synchronized to the control unit, the deviation rate between the execution effect and the expected target is monitored in real time, and the feedback gain of the closed-loop link is adjusted using an incremental PID control algorithm to ensure stability and adaptability during continuous intervention.

6. An intelligent pressure control system for arterial compression, characterized in that: The system comprises: An acquisition module is used to collect, in real time, the arterial blood pressure waveform, vascular elasticity parameters, and multi-dimensional pressure data from the pressure sensor at the patient's puncture site, and construct an initial pressure parameter set including the pressure distribution gradient and vascular response coefficient using a dynamic feature extraction algorithm. The dynamic feature extraction algorithm uses a deep convolutional neural network to quantitatively model the spatiotemporal distribution characteristics of the pressure data. an analysis module for collaboratively analyzing the initial pressure parameter set and the blood flow characteristic parameters obtained through blood flow monitoring, performing nonlinear coupling between the pressure gradient change rate and Doppler ultrasound blood flow velocity data, and generating a control matrix including an optimal compression force range and a time decay function; a processing module configured to perform multi-objective optimization processing on the control matrix using a parallel optimization algorithm to generate a real-time pressure control instruction set for pressure control of a compression device for arterial compression, wherein the parallel optimization algorithm simultaneously optimizes three objective functions: a compression force error rate, a blood flow index, and a patient comfort threshold, and employs a hybrid optimization strategy combining a genetic algorithm and fuzzy control; The feedback module is used to trigger a graded alarm mechanism and autonomously correct the compression strategy based on the deviation value between the real-time pressure execution data and blood flow monitoring data fed back by the user interface. The graded alarm mechanism dynamically adjusts the alarm level and synchronously updates the control parameter set of the control unit according to the magnitude of the deviation value exceeding the preset threshold, forming a closed-loop control link.

7. The system according to claim 6, characterized in that The acquisition module is specifically used to: Based on the multi-dimensional pressure data collected by the pressure sensor array and the non-uniform sampling characteristics of the arterial blood pressure waveform, a dynamic time warping algorithm is used to perform spatiotemporal alignment of vascular elasticity parameters to generate a spatiotemporally synchronized vascular-pressure coupled dataset. The spatiotemporally synchronized vascular-pressure coupled dataset is fed into a deep convolutional neural network. Multi-scale spatiotemporal convolution kernels are used to extract the correlation features between vascular wall deformation and pressure conduction. The pressure gradient field is then decomposed using a sparse coding algorithm to generate a pressure distribution gradient map containing local pressure extreme points and diffusion directions. According to the pressure distribution gradient map, the attention mechanism is used to adaptively weight the vascular elasticity parameters. The vascular response coefficient is nonlinearly superimposed with the pressure gradient field through the tensor fusion layer to generate an initial pressure parameter set including the dynamic compensation threshold and elasticity correction factor.

8. The system according to claim 7, characterized in that The analysis module is specifically used to: Based on the pressure gradient change rate in the initial pressure parameter set, the Doppler ultrasound blood flow velocity data is subjected to time-frequency analysis to extract the blood flow pulse phase characteristics. The dynamic time warping algorithm is then used to perform nonlinear matching with the pressure gradient time series to generate a spatiotemporally aligned pressure-blood flow coupling tensor. The pressure-blood flow coupling tensor is input into the high-order singular value decomposition model to extract the core feature matrix, and the feature matrix is ​​sparsely processed based on non-negative matrix decomposition to generate the coupling coefficient matrix of blood flow velocity and pressure gradient; A radial basis function neural network is used to construct a nonlinear mapping from the coupling coefficient matrix to the compression force. The time decay function is introduced as a dynamic constraint term to generate a control matrix containing the optimal compression force range and the time decay factor.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when run.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 5.

Images