Blood vessel intervention method driven by hemodynamics

Abstract

The invention provides a blood vessel intervention method driven by hemodynamics, which comprises the following steps: constructing a blood vessel fractal topology model according to multi-modal image fusion, and generating an initial intubation path according to the blood vessel fractal topology model; according to the initial intubation path, the embolism catheter is intubated to the first-stage blood vessel branch, and the blood vessel wall contact stress tensor is collected in real time; inputting the blood vessel wall contact stress tensor into an LSTM-GAN prediction network for catheter position calibration; after calibration is completed, the embolism catheter continues to be inserted into the target blood vessel according to an impedance self-adaptive control algorithm, and fluid mechanics parameters of the head of the embolism catheter are collected in real time; calculating to obtain the target blood vessel outlet resistance and the target blood vessel bed embolism proportion according to the fluid mechanics parameters; based on the target blood vessel outlet resistance and the target blood vessel bed embolism proportion, optimal injection parameters are obtained through a quantum particle swarm optimization algorithm. According to the method, through quantum optimization, multi-mode biological sensing and multi-physical field coupling technologies, the intubation positioning precision is improved, and the blood flow regurgitation occurrence rate is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of assisted interventional therapy, and particularly to a hemodynamic-driven vascular interventional method. Background Art

[0002] In current vascular interventional therapy, catheter navigation mainly relies on preoperative static image reconstruction, but it is difficult to fuse dynamic factors such as the elastic deformation of the blood vessel wall and the change of hemodynamic parameters during the operation in real time, resulting in the mismatch between the catheter advancement path and the anatomical structure and the risk of blood vessel perforation.

[0003] Most of the existing embolization parameter optimization methods adopt empirical decision-making or traditional optimization algorithms (such as genetic algorithms), which have problems of slow convergence speed and easy to fall into local optimum in complex blood flow-tissue coupling scenarios and cannot meet the real-time requirements during the operation. In addition, conventional embolization agent injection adopts a constant speed / constant pressure mode, ignoring the influence of blood non-Newtonian properties, temperature field distribution and eddy current effect on the diffusion of the embolization agent. Therefore, it is very necessary to design a hemodynamic-driven vascular interventional method. Summary of the Invention

[0004] The purpose of the present invention is to provide a hemodynamic-driven vascular interventional method, which realizes real-time optimal decision-making of target blood vessel intubation and embolization agent injection parameters by integrating quantum intelligent optimization technology, multi-modal biomechanical sensing technology and multi-physical field coupling modeling technology, so as to improve the intubation positioning accuracy and reduce the incidence of blood flow reflux.

[0005] To achieve the above purpose, the present invention provides the following scheme:

[0006] A hemodynamic-driven vascular interventional method includes the following steps:

[0007] Construct a vascular fractal topology model according to multi-modal image fusion, and generate an initial intubation path according to the vascular fractal topology model;

[0008] Insert an embolization catheter into a primary blood vessel branch according to the initial intubation path, and collect the vascular wall contact stress tensor in real time;

[0009] Input the vascular wall contact stress tensor into an LSTM-GAN prediction network for catheter position calibration;

[0010] After calibration, continue to insert the embolization catheter into the target blood vessel according to the impedance adaptive control algorithm, and collect the hydrodynamic parameters at the head of the embolization catheter in real time;

[0011] Calculate the target blood vessel outlet resistance and the target blood vessel bed embolization ratio according to the hydrodynamic parameters;

[0012] Based on the resistance at the target vessel outlet and the embolization ratio of the target vessel bed, the optimal injection parameters are obtained through the quantum particle swarm optimization algorithm.

[0013] Optionally, a vascular fractal topology model is constructed according to multimodal image fusion, and an initial intubation path is generated based on the vascular fractal topology model, including:

[0014] Performing image registration on the DSA dynamic angiography sequence through optical coherence tomography technology, and calculating the vascular fractal dimension;

[0015] Constructing a vascular fractal topology model based on the vascular fractal dimension, and embedding a hemodynamic feature matrix at the fractal nodes of the vascular fractal topology model to obtain a path optimization function;

[0016] Solving the path optimization function through the A-Star algorithm to obtain a gradient navigation path;

[0017] When the vascular fractal dimension is greater than a preset dimension threshold, adding a path curvature constraint condition to the gradient navigation path to obtain an initial intubation path.

[0018] Optionally, the specific steps for calculating the vascular fractal dimension include:

[0019] Performing vascular segmentation on the DSA dynamic angiography sequence to obtain a vascular skeleton model;

[0020] Performing three-dimensional discrete wavelet transform on the vascular skeleton model to obtain the Hurst exponent;

[0021] Performing singular value decomposition on the Hurst exponent to obtain singular values;

[0022] Through the calculation formula calculate to obtain the vascular fractal dimension D f , where σ i is the weight of the i-th singular value, β is the nonlinear gain factor, H s is the Hurst exponent, W j is the modulus value of the j-th wavelet coefficient, E m is the local elastic modulus, and E0 is the reference elastic modulus.

[0023] Optionally, according to the initial intubation path, inserting the embolization catheter into the first-level vascular branch, and collecting the vascular wall contact stress tensor in real time, including:

[0024] Obtaining the original signal of the three-dimensional contact stress distribution through the flexible piezoresistive microsensor array preset at the head of the embolization catheter at a sampling frequency of 500 Hz;

[0025] Performing noise reduction processing on the original signal through empirical mode decomposition, and separating the viscoelastic response characteristic frequency band of the vascular wall;

[0026] Construct a contact dynamic equation according to the characteristic frequency band of the viscoelastic response of the blood vessel wall;

[0027] Solve the contact dynamic equation through the Lyapunov stability criterion to obtain the contact stress tensor of the blood vessel wall.

[0028] Optionally, the expression of the contact dynamic equation is: where ρ is the density of the blood vessel wall, is the displacement vector field, is the macroscopic equivalent modulus, F is the deformation gradient tensor, S is the stress tensor, S v is the viscous stress tensor, τ is the relaxation time, η is the fractal correction factor, γ is the contact stiffness coefficient, erf(·) is the Gaussian error function, ∈ is the critical strain threshold, β is the nonlinear gain factor, is the blood flow velocity field.

[0029] Optionally, input the contact stress tensor of the blood vessel wall into the LSTM-GAN prediction network for catheter position calibration, including:

[0030] Synchronize and align the spatio-temporal characteristics of the contact stress tensor of the blood vessel wall, the positioning signal of the embolization catheter, and the blood vessel fractal topological model, and then generate a spatio-temporal feature encoding matrix through the spatio-temporal attention mechanism;

[0031] Encode the blood vessel bifurcation angle, curvature radius, and blood flow impulse vector extracted from the blood vessel fractal topological model into a topological feature vector;

[0032] Perform a tensor splicing operation on the topological feature vector and the spatio-temporal feature encoding matrix to obtain the input feature;

[0033] Extract the motion time-series features of the input feature through the bidirectional LSTM sub-network of the LSTM-GAN prediction network;

[0034] Perform loss compensation on the motion time-series features through the multi-physical field coupling loss to obtain the compensated motion features; the multi-physical field coupling loss includes: hemodynamic loss, biomechanical loss, and anatomical constraint loss;

[0035] Calibrate the position of the embolization catheter according to the compensated motion features.

[0036] Optionally, after calibration, continue to insert the embolization catheter into the target blood vessel according to the impedance adaptive control algorithm, and collect the hydrodynamic parameters at the head of the embolization catheter in real time, including:

[0037] Calculate the impedance gradient according to the hydrodynamic parameters; the hydrodynamic parameters include: three-dimensional blood flow velocity vector field, time-varying pressure gradient, pulsation amplitude, vorticity of the flow field, apparent viscosity of blood, and temperature field distribution;

[0038] Based on the change in impedance gradient, the path adjustment amount that satisfies the minimum energy consumption is obtained through the quantum-derived obstacle avoidance algorithm, and the intubation position of the embolization catheter is dynamically adjusted according to the path adjustment amount.

[0039] Optionally, the calculation formula for the resistance at the target vessel outlet is: where η a is the time-averaged apparent viscosity, is the pressure gradient modulus, is the maximum blood flow velocity, ρ b is the blood density, <ω> is the time-averaged vorticity modulus, is the pulsating pressure amplitude, λ is the turbulence characteristic scale, and T is the time.

[0040] Optionally, the calculation formula for the embolization ratio of the target vessel bed is: where <ω> is the time-averaged vorticity modulus, τ r is the blood flow residence time, K is the embolizing agent diffusion coefficient, is the time-averaged pressure, α is the thermodynamic coupling coefficient, and ΔT is the temperature field range.

[0041] Optionally, based on the resistance at the target vessel outlet and the embolization ratio of the target vessel bed, the optimal injection parameters are obtained through the quantum particle swarm optimization algorithm, including:

[0042] Map the injection parameters to the superposition state of quantum bits, and mix the superposition state and the injection parameters into the initial population;

[0043] Construct a fitness function according to the resistance at the target vessel outlet and the embolization ratio of the target vessel bed;

[0044] Update and iterate the initial population through the quantum tunneling effect to obtain a quantum population;

[0045] Collapse the quantum population into the initial population through the fitness function for fitness evaluation to obtain the fitness, and perform reverse update on the quantum population through the fitness to obtain the optimized population;

[0046] Solve the optimized population through the Pareto optimal algorithm to obtain the optimal injection parameters.

[0047] According to the specific embodiments provided by the present invention, the following technical effects are disclosed: The hemodynamic-driven vascular intervention method provided by the present invention includes: constructing a vascular fractal topological model based on multimodal image fusion, and generating an initial intubation path according to the vascular fractal topological model; inserting an embolization catheter into the primary vascular branch according to the initial intubation path, and collecting the vascular wall contact stress tensor in real time; inputting the vascular wall contact stress tensor into the LSTM-GAN prediction network for catheter position calibration; after calibration, continuing to insert the embolization catheter into the target blood vessel according to the impedance adaptive control algorithm, and collecting the hydrodynamic parameters at the head of the embolization catheter in real time; calculating the target blood vessel outlet resistance and the target blood vessel bed embolization ratio according to the hydrodynamic parameters; based on the target blood vessel outlet resistance and the target blood vessel bed embolization ratio, obtaining the optimal injection parameters through the quantum particle swarm optimization algorithm. This method realizes the real-time optimal decision-making of target blood vessel intubation and embolization agent injection parameters by integrating quantum intelligent optimization technology, multimodal biomechanical perception technology, and multi-physical field coupling modeling technology, improves the intubation positioning accuracy, and reduces the incidence of blood flow reflux. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0049] Figure 1 It is a flowchart of the vascular intervention method of the present invention;

[0050] Figure 2 It is a flowchart of generating the initial intubation path of the present invention;

[0051] Figure 3 It is a flowchart of inserting the embolization catheter of the present invention;

[0052] Figure 4 It is a flowchart of catheter position calibration of the present invention;

[0053] Figure 5 It is a flowchart of inserting the catheter into the target blood vessel of the present invention;

[0054] Figure 6 It is a flowchart of the quantum particle swarm optimization algorithm of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0056] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0057] As Figure 1 shown, the present invention provides a hemodynamic-driven vascular intervention method, including the following steps:

[0058] Step 100: Construct a vascular fractal topology model based on multimodal image fusion, and generate an initial intubation path according to the vascular fractal topology model; the specific steps are as Figure 2 shown, including:

[0059] Step 101: Perform image registration on the DSA dynamic angiography sequence through optical coherence tomography technology, and calculate the vascular fractal dimension;

[0060] Specifically, by synchronously collecting the dynamic sequence of digital subtraction angiography (DSA) and the high-resolution tomographic data of optical coherence tomography (OCT), and adopting a hybrid algorithm combining affine transformation and elastic registration to achieve precise alignment of multimodal images. In the rigid registration stage during the alignment process, the main blood vessels are globally matched by maximizing mutual information to eliminate macroscopic offsets caused by patient body position movement; in the non-rigid registration stage, a free deformation field is constructed based on the vascular wall microstructure features extracted by OCT and local sub-pixel-level alignment is completed through B-spline interpolation. Subsequently, the DSA image is segmented for the vascular lumen through the Otsu adaptive threshold algorithm, and small holes are filled by combining morphological closing operations to generate a continuous and smooth vascular binary mask, and the Zhang-Suen skeletonization algorithm is used to extract the vascular skeleton model. Then, the vascular skeleton model is decomposed by three-dimensional discrete wavelet using Haar wavelet basis functions. Through four-layer wavelet decomposition, wavelet coefficient matrices of different scales are obtained respectively, and the energy distribution of each scale is calculated. Then, the logarithmic energy-logarithmic scale curve is fitted by the least square method, and the slope of the curve is used as the Hurst exponent. Then, the Hurst exponent matrix is expanded into a second-order tensor, the first k principal components are extracted to construct a low-dimensional feature space and the singular values are calculated. Finally, through the calculation formula:

[0061]

[0062] the vascular fractal dimension D is calculated f , where σi is the weight of the i-th singular value, β is the non-linear gain factor, which is 0.17 in this embodiment, and H s is the Hurst exponent, and W j is the modulus value of the j-th wavelet coefficient, and E m is the local elastic modulus, and E0 is the reference elastic modulus.

[0063] It should be noted that through multi-modal image fusion and non-linear feature extraction, the accurate quantification of the vascular fractal dimension is realized, and the calculation error rate is reduced.

[0064] Step 102: Construct a vascular fractal topology model according to the vascular fractal dimension, and embed the hemodynamic feature matrix at the fractal nodes of the vascular fractal topology model to obtain a path optimization function;

[0065] Specifically, a recursive algorithm is used to construct a self-similar vascular branch structure (vascular fractal topology model) in combination with the vascular fractal dimension and the geometric parameters (bifurcation angle, branch length, etc.) of the vascular centerline. The branching rule of each fractal node is determined by the vascular fractal dimension. The higher the vascular fractal dimension, the more complex the branching and the larger the number of micro-bifurcations.

[0066] Identify the key fractal nodes (such as the first-level bifurcation point, the starting point of the stenosis segment) in the topology model, and extract their geometric attributes (coordinates, branch angle, etc.). Obtain the blood flow velocity v at the node through fluid dynamics simulation (CFD) a , the wall shear stress τ w and the pressure gradient and combine them into a local hemodynamic feature vector. Expand the hemodynamic feature vector into a 3×3 hemodynamic feature matrix M at each fractal node, where ω is the vorticity, Re is the Reynolds number, Wo is the Womersley number, and κ c is the local curvature. Then compress the matrix into a 32-dimensional feature vector and splice it with the topological attributes of the fractal node to generate the node-level fusion feature F n . Then construct a path optimization function, and the expression is where b is the curvature penalty coefficient, which is 0.8 in this embodiment, and λ1 + λ2 = 1. In this embodiment, λ1 = 0.6 and λ2 = 0.4.

[0067] Step 103: Solve the path optimization function through the A-Star algorithm to obtain the gradient navigation path;

[0068] Step 104: When the vascular fractal dimension is greater than the preset dimension threshold, add a path curvature constraint condition to the gradient navigation path to obtain the initial intubation path.

[0069] Specifically, when D f > the preset dimensional threshold of 2.6, a curvature continuity constraint is imposed on the gradient navigation path. This constraint is obtained by calculating the curvature of the path through the Frenet - Serret formula, and the Lagrange multiplier method is used to correct the curvature continuity constraint to obtain the initial intubation path.

[0070] Step 200: Insert the embolization catheter into the first - level blood vessel branch according to the initial intubation path, and collect the vascular wall contact stress tensor in real - time. The specific steps are as Figure 3 shown, including:

[0071] Step 201: Obtain the original signal of the three - dimensional contact stress distribution through the flexible piezoresistive microsensor array preset at the head of the embolization catheter at a sampling frequency of 500 Hz;

[0072] Specifically, the flexible piezoresistive microsensor array uses a polyimide substrate (with a thickness of 5 μm in this embodiment) and a carbon nanotube composite sensitive layer (with a sensitivity of 0.1 kPa in this embodiment -1 ), and it has 128 channels. When the embolization catheter contacts the vascular wall, the three - dimensional stress distribution causes a change in the sensor resistance. At the same time, the voltage signals of each channel are synchronously collected at a sampling frequency of 500 Hz, and the differential signal transmission technology is used to improve the signal - to - noise ratio of the voltage signals.

[0073] Step 202: Denoise the original signal through empirical mode decomposition and separate the characteristic frequency band of the vascular wall viscoelastic response;

[0074] Specifically, first decompose the original signal into 8 intrinsic mode functions (IMFs). Among them, IMF1 - 3 (1 - 30 Hz) contains the main components of the vascular wall viscoelastic response, and IMF4 - 8 contains high - frequency noise (frequency > 100 Hz) and low - frequency motion artifacts (frequency < 0.5 Hz). Then calculate the kurtosis and entropy of the intrinsic mode functions, retain the IMF components with kurtosis value > 3.5 and entropy value < 1.2, and set the remaining components to zero. Finally, superimpose the retained IMF components to generate the denoised signal.

[0075] Step 203: Construct a contact dynamic equation according to the characteristic frequency band of the vascular wall viscoelastic response;

[0076] Specifically, within the characteristic frequency band of 1 - 30 Hz, generate the dynamic response of the vascular wall through the Kelvin - Voigt model, regard the contact between the catheter and the blood vessel as a dynamic boundary condition, and introduce the Hertz contact theory correction term F c to form a contact coupling equation, and the expression is:

[0077]

[0078] Where R is the radius of curvature of the catheter head, E is the elastic modulus of the blood vessel wall, δ is the insertion depth, and ζ is the Poisson's ratio of the blood vessel wall, which is 0.45 in this embodiment. Finally, the dynamic response and contact coupling equations are combined to obtain the contact dynamic equation, which is expressed as follows:

[0079]

[0080] Where ρ is the density of the blood vessel wall, is the displacement vector field, is the macroscopic equivalent modulus, F is the deformation gradient tensor, S is the stress tensor, S v is the viscous stress tensor, τ is the relaxation time, η is the fractal correction factor, and γ is the contact stiffness coefficient, which is 1.2×10 6 Pa / m, erf(·) is the Gaussian error function, ∈ is the critical strain threshold, which is 0.08 in this embodiment, β is the nonlinear gain factor, which is 0.6 in this embodiment, and its value is the ratio of blood flow density to vascular wall density, is the blood flow velocity field.

[0081] Step 204: Solve the contact dynamic equation using the Lyapunov stability criterion to obtain the vascular wall contact stress tensor.

[0082] Specifically, we first define the energy-type Lyapunov function Where δ is the real-time displacement of the catheter head in the normal direction of the blood vessel wall, k p is the feedback gain of displacement and velocity, which is 1.8×10 3 N / m, and then calculate the time derivative V' of V to verify the global asymptotic stability condition of the contact dynamic equation to ensure that there is no oscillation divergence caused by energy accumulation during the contact process. Then, based on the Lyapunov function, the contact dynamic equation is iteratively solved in combination with the Newmark-β method. In each iteration, the solution vector is updated through the Jacobian matrix until the residual norm is less than 10 -6 , and the final output is the vascular wall contact stress tensor including normal stress and shear stress.

[0083] It should be noted that by deeply integrating high-density flexible sensor arrays with nonlinear dynamics, real-time and accurate perception of vascular wall contact stress is achieved, the resolution of stress measurement is improved, and the contact stress peak detection error is reduced.

[0084] Step 300: Input the vascular wall contact stress tensor into the LSTM-GAN prediction network to calibrate the catheter position; the specific steps are as follows Figure 4 As shown, including:

[0085] Step 301: Synchronize the spatio-temporal features of the vascular wall contact stress tensor, the positioning signal of the embolization catheter, and the vascular fractal topological model, and then generate a spatio-temporal feature encoding matrix through the spatio-temporal attention mechanism;

[0086] Specifically, use a bidirectional gated recurrent unit to perform time synchronization processing on the vascular wall contact stress tensor, the electromagnetic positioning signal of the embolization catheter (including three-dimensional coordinates and attitude angles), and the bifurcation node geometric parameters of the vascular fractal topological model. At the same time, use the gating mechanism of the bidirectional gated recurrent unit to eliminate the sampling frequency differences of multi-source data and generate fused features with aligned timestamps. Then map the vascular fractal topological model to the DSA image coordinate system and unify the spatial reference through a rotation matrix and a translation vector. Then obtain a spatio-temporal feature encoding matrix with dimensions of B×T×512 through the adaptive feature weighting operation of the multi-head attention mechanism (B is the batch size and T is the time step).

[0087] Step 302: Encode the vascular bifurcation angle, curvature radius, and blood flow impulse vector extracted from the vascular fractal topological model into a topological feature vector;

[0088] Specifically, first calculate the included angle between the centerline tangent vectors of adjacent branches (i.e., the vascular bifurcation angle) in the vascular fractal topological model, and based on the discrete point sequence of the vascular centerline in the vascular fractal topological model, locally fit an arc using the three-point method and solve for the curvature radius at the same time. Then calculate the momentum impulse passing through the vascular cross-section per unit time. The calculation formula is:

[0089]

[0090] where ρ b is the blood density, A is the vascular cross-sectional area, is the cross-section normal vector, is the blood flow velocity. Finally, hierarchically aggregate the vascular bifurcation angle, curvature radius, and blood flow impulse vector through a graph convolutional network and output a topological feature vector with dimensions of B×256.

[0091] Step 303: Perform a tensor concatenation operation on the topological feature vector and the spatio-temporal feature encoding matrix to obtain the input feature;

[0092] Specifically, expand the dimension of the topological feature vector to B×T×256 to complete the dimension matching with the spatio-temporal feature encoding matrix. Then concatenate the two matched matrices along the feature axis to generate an input feature tensor and perform layer normalization processing to eliminate the dimension difference.

[0093] Step 304: Extract the motion time series features of the input feature through the bidirectional LSTM sub-network of the LSTM-GAN prediction network;

[0094] Specifically, the hidden state of the input features is updated at each time step through forward propagation, while the hidden state is updated according to the future moment dependence captured by backward propagation, and then the forward and backward hidden states are concatenated to obtain the motion time series features.

[0095] Step 305: Perform loss compensation on the motion time series features through the multi-physics coupling loss to obtain the compensated motion features;

[0096] Specifically, the multi-physics coupling loss includes: hemodynamic loss biomechanical loss and anatomical constraint loss The expressions are respectively:

[0097]

[0098] Among them, is the wall shear stress predicted by the LSTM-GAN network, is the true wall shear stress, ||·|| represents the Euclidean distance calculation, is the component of the vascular wall contact stress tensor, is the component of the vascular wall stress tensor obtained by finite element method simulation, Dice(·) is the Dice similarity coefficient calculation, M p is the binary mask of the initial intubation path, M t is the true vascular lumen mask. Then, these three losses are weighted and summed with weights of 0.5, 0.3, and 0.2 respectively, and the network parameters of the motion time series features are updated through backpropagation to obtain the compensated motion features.

[0099] Step 306: Perform position calibration on the embolization catheter position according to the compensated motion features.

[0100] Specifically, a six-degree-of-freedom calibration vector of the compensated motion features is generated through a three-dimensional transposed convolutional layer, and the position and attitude of the embolization catheter are updated in real time according to the calibration vector, so as to achieve real-time position calibration.

[0101] Step 400: After calibration, continue to insert the embolization catheter into the target blood vessel according to the impedance adaptive control algorithm, and collect the hydrodynamic parameters at the head of the embolization catheter in real time; the specific steps are as Figure 5 shown, including:

[0102] Step 401: Calculate the impedance gradient according to the hydrodynamic parameters; the hydrodynamic parameters include: three-dimensional blood flow velocity vector field, time-varying pressure gradient, pulsation amplitude, vorticity of the flow field, blood apparent viscosity, and temperature field distribution;

[0103] Specifically, the calculation formula of the impedance gradient is:

[0104]

[0105] Among them, η a is the time-averaged apparent viscosity, is the blood flow velocity, is the pressure gradient modulus, and ΔT is the time difference.

[0106] The three-dimensional blood flow velocity vector field is obtained by measuring the three-dimensional flow velocity distribution through pulse opposed-phase encoding technology; the time-varying pressure gradient is obtained by calculating the gradient field through the central difference method; the pulsation amplitude is obtained by performing Hilbert-Huang transform on the blood pressure signal; the vorticity of the flow field is calculated based on the velocity field data by using the second-order central difference method to calculate the curl; the temperature field distribution is obtained by infrared imaging technology.

[0107] Step 402: Based on the change amount of the impedance gradient, obtain the path adjustment amount that satisfies the minimum energy consumption through the quantum-derived obstacle avoidance algorithm, and dynamically adjust the intubation position of the embolization catheter according to the path adjustment amount.

[0108] Specifically, first, based on the impedance gradient field calculated in step 401, discretize the blood vessel lumen into hexahedral grids with a side length of 0.2 mm, and store the three-component data of the normalized impedance gradient in three-dimensional space in each voxel. Construct a spatial obstacle avoidance potential field through the exponential repulsive potential field mapping, where the potential energy in the high-gradient region is significantly enhanced. Subsequently, encode the catheter movement path as a quantum bit combined state, and define a Hamiltonian that includes an energy consumption operator and an obstacle constraint operator, where the energy consumption operator is determined by the convolution operation of the path length and the local blood viscosity energy consumption, and the obstacle constraint operator realizes quantum tunneling suppression through the Heaviside function of the blood vessel wall distance field. Then, adiabatically evolve the initial uniform superposition state (i.e., the initial value of the Hamiltonian) to the Ising model. Finally, perform quantum measurement decoding on the Ising model to obtain the optimal path node coordinates, thereby adjusting the intubation position.

[0109] Step 500: Calculate the resistance at the outlet of the target blood vessel and the embolization ratio of the target blood vessel bed according to the hydrodynamic parameters;

[0110] Specifically, the calculation formula for the resistance at the outlet of the target blood vessel is:

[0111]

[0112] Among them, η a is the time-averaged apparent viscosity, is the pressure gradient modulus, is the maximum blood flow velocity, ρ b is the blood density, <ω> is the time-averaged vorticity modulus, is the pulsating pressure amplitude, λ is the turbulence characteristic scale, and T is the time.

[0113] Specifically, the calculation formula for the embolization ratio of the target vascular bed is as follows:

[0114]

[0115] Among them, <ω> is the time-averaged vorticity modulus, τ r is the blood flow retention time, K is the embolizing agent diffusion coefficient, is the time-averaged pressure, α is the thermodynamic coupling coefficient, and ΔT is the temperature field range.

[0116] Step 600: Based on the target vascular outlet resistance and the embolization ratio of the target vascular bed, obtain the optimal injection parameters through the quantum particle swarm optimization algorithm. The specific steps are as Figure 6 shown, including:

[0117] Step 601: Map the injection parameters to the superposition state of quantum bits, and mix the superposition state and the injection parameters into the initial population;

[0118] Step 602: Construct a fitness function according to the target vascular outlet resistance and the embolization ratio of the target vascular bed;

[0119] Step 603: Update and iterate the initial population through the quantum tunneling effect to obtain the quantum population;

[0120] Step 604: Collapse the quantum population into the initial population through the fitness function for fitness evaluation to obtain the fitness, and perform reverse update on the quantum population through the fitness to obtain the optimized population;

[0121] Step 605: Solve the optimized population through the Pareto optimal algorithm to obtain the optimal injection parameters.

[0122] The beneficial effects of the present invention are as follows:

[0123] 1) By collaborating the real-time calibration of the LSTM-GAN network with the quantum obstacle avoidance algorithm, the accuracy of catheter head positioning is improved;

[0124] 2) The perforation risk during the operation is reduced through the contact stress control mechanism;

[0125] 3) The success rate of path planning is improved through the vascular fractal dimension, and the intubation time for narrow blood vessels can be shortened;

[0126] 4) By converging to the optimal solution through the quantum particle swarm algorithm, the parameter optimization efficiency is improved, the dosage of the embolizing agent is saved, and the operation cost is reduced.

[0127] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0128] In the present invention, specific examples are used to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A hemodynamically driven vascular intervention method, characterized in that: The steps include: constructing a vascular fractal topology model according to multimodal image fusion, and generating an initial cannulation path according to the vascular fractal topology model; inserting the embolization catheter into the primary vascular branch according to the initial insertion path, and collecting the vascular wall contact stress tensor in real time; Inputting the vascular wall contact stress tensor into the LSTM-GAN prediction network to calibrate the catheter position; After the calibration is completed, the embolization catheter is further inserted into the target blood vessel according to the impedance adaptive control algorithm, and the fluid mechanics parameters of the embolization catheter head are collected in real time; Calculate the target vessel outlet resistance and the target vessel bed embolism ratio according to the fluid mechanics parameters; Based on the target vascular outlet resistance and the target vascular bed embolism ratio, the optimal injection parameters are obtained by quantum particle swarm optimization algorithm.

2. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: A vascular fractal topology model is constructed according to multimodal image fusion, and an initial cannulation path is generated according to the vascular fractal topology model, including: Optical coherence tomography was used to register the DSA dynamic angiography sequences and calculate the vascular fractal dimension. Constructing the vascular fractal topology model according to the vascular fractal dimension, and embedding the hemodynamic characteristic matrix at the fractal nodes of the vascular fractal topology model to obtain a path optimization function; Solving the path optimization function by using the A-Star algorithm to obtain a gradient navigation path; When the blood vessel fractal dimension is greater than a preset dimension threshold, a path curvature constraint condition is added to the gradient navigation path to obtain the initial cannulation path.

3. The hemodynamically driven vascular intervention method according to claim 2, characterized in that: The specific steps of calculating the fractal dimension of the blood vessel include: Performing blood vessel segmentation on the DSA dynamic angiography sequence to obtain a blood vessel skeleton model; Performing a three-dimensional discrete wavelet transform on the blood vessel skeleton model to obtain a Hurst index; Performing singular value decomposition on the Hurst index to obtain a singular value; By calculating the formula The blood vessel fractal dimension D is calculated f , where σ i is the i-th singular value weight, β is the nonlinear gain factor, H s is the Hurst index, W j is the modulus of the jth wavelet coefficient, E m is the local elastic modulus, and E0 is the reference elastic modulus.

4. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: The embolization catheter is inserted into the primary vascular branch according to the initial insertion path, and the vascular wall contact stress tensor is collected in real time, including: The original signal of the three-dimensional contact stress distribution is obtained at a sampling frequency of 500 Hz by using a flexible piezoresistive microsensor array preset on the head of the embolization catheter; Performing noise reduction processing on the original signal by empirical mode decomposition, and separating the characteristic frequency band of the viscoelastic response of the blood vessel wall; Constructing a contact dynamic equation according to the characteristic frequency band of the viscoelastic response of the blood vessel wall; The contact dynamic equation is solved by Lyapunov stability criterion to obtain the blood vessel wall contact stress tensor.

5. The hemodynamically driven vascular intervention method according to claim 4, characterized in that: The expression of the contact dynamic equation is: Where ρ is the density of the blood vessel wall, is the displacement vector field, is the macroscopic equivalent modulus, F is the deformation gradient tensor, S is the stress tensor, S v is the viscous stress tensor, τ is the relaxation time, η is the fractal correction factor, γ is the contact stiffness coefficient, erf(·) is the Gaussian error function, ∈ is the critical strain threshold, β is the nonlinear gain factor, is the blood flow velocity field.

6. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: The vascular wall contact stress tensor is input into the LSTM-GAN prediction network for catheter position calibration, including: After performing spatiotemporal feature synchronization alignment on the vascular wall contact stress tensor, the positioning signal of the embolization catheter and the vascular fractal topology model, a spatiotemporal feature encoding matrix is ​​generated through a spatiotemporal attention mechanism; encoding the vascular bifurcation angle, curvature radius and blood flow impulse vector extracted from the vascular fractal topological model into a topological feature vector; Performing a tensor concatenation operation on the topological feature vector and the spatiotemporal feature encoding matrix to obtain input features; Extracting the motion timing features of the input features through the bidirectional LSTM sub-network of the LSTM-GAN prediction network; The motion timing characteristics are compensated by multi-physical field coupling loss to obtain compensated motion characteristics; the multi-physical field coupling loss includes: hemodynamic loss, biomechanical loss and anatomical constraint loss; The position of the embolization catheter is calibrated according to the compensated motion characteristics.

7. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: After the calibration is completed, the embolization catheter is inserted into the target blood vessel according to the impedance adaptive control algorithm, and the fluid mechanics parameters of the embolization catheter head are collected in real time, including: Obtaining impedance gradient according to the fluid mechanics parameters and calculations; the fluid mechanics parameters include: three-dimensional blood flow velocity vector field, time-varying pressure gradient, pulsation amplitude, flow field vorticity, blood apparent viscosity and temperature field distribution; Based on the change in the impedance gradient, a path adjustment amount that satisfies the minimum energy consumption is obtained through a quantum-derived obstacle avoidance algorithm, and the insertion position of the embolization catheter is dynamically adjusted according to the path adjustment amount.

8. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: The calculation formula of the target blood vessel outlet resistance is: Among them, η a is the time-averaged apparent viscosity, is the pressure gradient modulus, is the maximum blood flow velocity, ρ b is the blood density, <ω> is the time-averaged vorticity modulus, Δp p is the pulsating pressure amplitude, λ is the characteristic scale of turbulence, and T is time.

9. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: The calculation formula for the target vascular bed embolism ratio is: Where <ω> is the time-averaged vorticity modulus, τ r is the blood flow retention time, K is the diffusion coefficient of the embolic agent, is the time-averaged pressure, α is the thermodynamic coupling coefficient, and ΔT is the extreme difference in temperature field.

10. The hemodynamically driven vascular intervention method according to claim 1, characterized in that: Based on the target vascular outlet resistance and the target vascular bed embolism ratio, the optimal injection parameters are obtained by quantum particle swarm optimization algorithm, including: Mapping the injection parameters into a superposition state of quantum bits, and mixing the superposition state and the injection parameters into an initial population; Constructing a fitness function according to the outlet resistance of the target blood vessel and the embolism ratio of the target blood vessel bed; The initial population is updated and iterated through the quantum tunneling effect to obtain a quantum population; The quantum population is collapsed into the initial population by the fitness function to perform fitness evaluation to obtain the fitness, and the quantum population is reversely updated by the fitness to obtain the optimized population; The optimized population is solved by a Pareto optimal algorithm to obtain the optimal injection parameters.

Images