Internal arteriovenous fistula intelligent planning method and system based on hemodynamic simulation

Abstract

The invention relates to the technical field of medical image processing and medical aid decision making, in particular to an internal arteriovenous fistula intelligent planning method and system based on haemodynamic simulation, and the method comprises the steps: firstly obtaining three-dimensional space anatomical data of arteriovenous blood vessels of a patient, and constructing a blood vessel tree three-dimensional model; and inputting the blood vessel parameter data and the hemodynamic parameter data at different blood flow rates into a deep learning model to obtain dynamic shear force data of the anastomotic stoma, and comparing the dynamic shear force data of the anastomotic stoma with preset optimal anastomotic angle data to determine an optimal anastomotic angle. And performing internal arteriovenous fistula operation type three-dimensional modeling on the three-dimensional model of the vascular tree by combining the optimal anastomosis angle to obtain a three-dimensional internal arteriovenous fistula model, and performing a virtual operation based on the model to obtain virtual operation process data and a postoperative hemodynamic state evaluation result. The success rate of one-time operation of internal fistula can be increased to 85%-90% from 60%-65% in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing and medical auxiliary decision-making, and in particular to an intelligent planning method and system for arteriovenous fistula based on hemodynamic simulation, which is particularly suitable for vascular access establishment planning for hemodialysis patients. Background Art

[0002] Arteriovenous fistulas (AVFs) are the preferred form of long-term vascular access for hemodialysis patients. They are created by directly anastomosing the patient's own arteries and veins, causing the veins to dilate and arterialize under the stimulation of high-pressure blood flow, thereby forming a vascular access suitable for repeated punctures. However, in clinical practice, the success rate of AVFs is approximately 60% to 65%, with an early failure rate as high as 30%. Approximately 40% of fistulas require reoperation within a year to maintain function.

[0003] Currently, fistula surgery planning relies primarily on physician experience and simple vascular measurements (such as vessel diameter and blood flow velocity), lacking systematic consideration of hemodynamic factors. Although studies have shown that the hemodynamic characteristics of the anastomotic site, particularly the distribution of wall shear forces, are closely related to fistula maturity and long-term patency, existing technologies have limitations in the following areas:

[0004] 1. Although traditional computational fluid dynamics (CFD) software can simulate blood flow, the calculations are time-consuming and difficult to apply to real-time clinical decision-making.

[0005] 2. Existing methods have difficulty accurately capturing the dynamic process of blood-vessel wall interaction, especially the differences at different anastomotic angles;

[0006] 3. There is a lack of effective methods to combine microscopic shear stress changes with long-term prediction of fistula function;

[0007] 4. It is impossible to achieve individualized fistula planning and optimization, making it difficult to meet the special needs of different patients.

[0008] Therefore, there is an urgent need to develop a fistula planning method that can combine hemodynamic simulation and intelligent prediction to improve the success rate and long-term patency rate of fistula surgery. Summary of the Invention

[0009] The purpose of the present invention is to provide an intelligent planning method and system for arteriovenous fistula based on hemodynamic simulation, aiming to solve the problems in the existing technology such as lack of individualization in fistula planning and difficulty in accurately predicting long-term function.

[0010] The present invention proposes an intelligent planning method for arteriovenous fistula based on hemodynamic simulation, comprising:

[0011] Obtain the three-dimensional anatomical data of the patient's arteries and veins and construct a three-dimensional model of the vascular tree;

[0012] Combining the vascular parameter data of the three-dimensional vascular tree model with the hemodynamic parameter data at different blood flow rates, and inputting the data into a deep learning model to obtain the dynamic shear force data of the anastomosis;

[0013] Comparing the anastomotic dynamic shear force data with the preset optimal anastomotic angle data to determine the optimal anastomotic angle;

[0014] Performing three-dimensional modeling of an arteriovenous fistula procedure on the three-dimensional model of the vascular tree in combination with the optimal anastomosis angle to obtain a three-dimensional arteriovenous fistula model;

[0015] Based on the three-dimensional arteriovenous fistula model, a virtual surgery is performed to obtain virtual surgery process data and postoperative hemodynamic status evaluation results.

[0016] Preferably, obtaining the three-dimensional anatomical data of the patient's arteries and veins and constructing the three-dimensional model of the vascular tree specifically includes:

[0017] Acquire original vascular 3D data from high-resolution ultrasound or computed tomography angiography;

[0018] Inputting the original three-dimensional blood vessel data into a two-stage deformable 3D convolutional neural network based on a self-attention mechanism for processing to obtain the three-dimensional blood vessel tree model;

[0019] Among them, the self-attention mechanism is used to enhance the recognition accuracy of key branch points of blood vessels.

[0020] Preferably, the construction process of the two-stage deformable 3D convolutional neural network based on the self-attention mechanism includes:

[0021] A two-stage deformable 3D convolutional neural network architecture based on the self-attention mechanism is constructed. The first stage obtains the rough segmentation results of blood vessels, and the second stage refines the blood vessel boundaries and branch regions.

[0022] Acquire a training dataset comprising three-dimensional blood vessel data and corresponding labels, wherein the labels include blood vessel boundaries, centerlines, and branch point location information;

[0023] The training data set is used to train the two-stage deformable 3D convolutional neural network based on the self-attention mechanism to obtain a trained neural network model.

[0024] Preferably, the vascular parameter data of the three-dimensional vascular tree model and the hemodynamic parameter data at different blood flow rates are combined and input into a deep learning model to obtain the anastomotic dynamic shear force data, specifically including:

[0025] Dividing the vascular tree three-dimensional model into different regions, including anastomotic core region, anastomotic edge region, transition region, and distal region, and assigning different mesh densities to each region;

[0026] Computational fluid dynamics software is used to calculate the pressure field, velocity field, and turbulence field of the three-dimensional vascular tree model, and hemodynamic parameters under different blood flow rates are set;

[0027] Inputting the hemodynamic parameters under different blood flow rates into an improved smooth particle fluid dynamics method for simulation to obtain shear force data of the anastomotic region under different blood flow rates;

[0028] The vascular parameter data is used as the input of the deep learning model to obtain the anastomotic dynamic shear force data.

[0029] Preferably, the improved smoothed particle hydrodynamics method comprises:

[0030] automatically generating an adaptive particle distribution according to the morphological characteristics of the three-dimensional model of the vascular tree;

[0031] Increasing the particle density in the anastomotic region to improve simulation accuracy;

[0032] Adaptive kernel function is used to dynamically adjust the kernel function radius according to the local particle density;

[0033] An improved virtual particle method is used to process the blood vessel wall boundary to improve the stability of the calculation at the boundary.

[0034] The elastic deformation response of the blood vessel wall is considered to achieve fluid-structure interaction simulation.

[0035] Preferably, comparing the anastomotic dynamic shear force data with preset optimal anastomotic angle data to determine the optimal anastomotic angle specifically includes:

[0036] Construct a deep correlation model between dynamic time-varying shear stress and vascular wall response, including a dual-stream feature extraction network and a multi-scale fusion module;

[0037] Inputting the anastomotic dynamic shear force data into the depth correlation model to obtain long-term vascular wall response prediction results under different anastomotic angles;

[0038] A Bayesian optimization framework enhanced by physical information was used to evaluate each candidate anastomotic angle scheme and calculate a multi-objective score that included surgical success rate, fistula maturation time, and long-term patency rate.

[0039] The optimal anastomotic angle is determined based on the multi-objective score and patient-specific needs.

[0040] Preferably, the Bayesian optimization framework based on physical information enhancement includes:

[0041] Integrating hemodynamic theory and clinical experience, a theoretical correlation model between anastomotic angle and fistula function was established;

[0042] Define physical feasibility constraints, including angular range and spatial position restrictions;

[0043] Construct a multi-objective function and dynamically adjust the weight coefficients of short-term goals, medium-term goals, and long-term goals;

[0044] Initialize candidate anastomosis angle solutions and use Gaussian process regression model to predict the performance of each solution;

[0045] Based on the expected improvement and uncertainty trade-off, the optimization is iterated until it converges to the optimal solution.

[0046] Preferably, performing a virtual surgery based on the three-dimensional arteriovenous fistula model to obtain the virtual surgery process data and the postoperative hemodynamic status assessment results specifically includes:

[0047] constructing a virtual surgical scene based on the optimal anastomosis angle and the three-dimensional model of the vascular tree;

[0048] Simulate the surgical process and generate postoperative vascular morphology;

[0049] Predict the fistula status at different time periods, including one week, one month, and six months later;

[0050] Calculate the predicted value of fistula maturation time and long-term patency rate;

[0051] Generate visual surgical guidance plans, including anastomosis angles, position markers, and key operation points.

[0052] As an advantage, it also includes an adaptive closed-loop feedback optimization step:

[0053] Actual surgical parameters and postoperative follow-up data were recorded;

[0054] Compare the predicted results with the actual results to quantify the accuracy of the prediction;

[0055] Update model parameters based on the difference between prediction and actual results;

[0056] Add new case data to the knowledge base and optimize sampling strategies;

[0057] Perform anomaly detection and analysis on cases of prediction failure to improve system robustness.

[0058] Arteriovenous fistula intelligent planning system based on hemodynamic simulation, including:

[0059] A vascular tree three-dimensional model construction module is used to obtain the three-dimensional spatial anatomical data of the patient's arteries and veins and construct a vascular tree three-dimensional model;

[0060] A hemodynamic simulation module is used to combine the vascular parameter data of the three-dimensional vascular tree model with the hemodynamic parameter data at different blood flow rates, input them into a deep learning model, and obtain anastomotic dynamic shear force data;

[0061] An optimal anastomotic angle determination module is used to compare the anastomotic dynamic shear force data with preset optimal anastomotic angle data to determine the optimal anastomotic angle;

[0062] a three-dimensional modeling module for the arteriovenous fistula procedure, configured to perform three-dimensional modeling of the arteriovenous fistula procedure on the three-dimensional model of the vascular tree in combination with the optimal anastomosis angle to obtain a three-dimensional arteriovenous fistula model;

[0063] A virtual surgery module, configured to perform virtual surgery based on the three-dimensional arteriovenous fistula model, and obtain virtual surgery process data and postoperative hemodynamic status evaluation results;

[0064] Adaptive closed-loop feedback module, used to record actual surgical parameters and postoperative follow-up data, compare predicted results with actual results, update model parameters, and improve system prediction accuracy;

[0065] This invention innovatively integrates hemodynamic simulation, deep learning, and multi-objective optimization technologies to achieve intelligent management of the entire process, from acquiring individual patient vascular data to generating the optimal surgical plan. Specifically, the invention proposes a series of technological innovations in predicting anastomotic shear forces and determining the optimal anastomotic angle, making fistula planning more accurate and efficient.

[0066] The present invention has the following beneficial effects:

[0067] 1. Significantly improve the success rate of fistula surgery from the current 60% to 65% to 85% to 90%, reducing additional surgical trauma for patients;

[0068] 2. Reduce the early failure rate of internal fistula from 30% to less than 15%, thereby improving the efficiency of medical resource utilization;

[0069] 3. Extend the lifespan of the fistula, increasing the one-year patency rate from 60% to over 80%, and reducing the need for maintenance interventions;

[0070] 4. Shorten the time for fistula planning from 2-3 hours in traditional methods to within 30 minutes, thus improving clinical work efficiency;

[0071] 5. Reduce dependence on expert doctors, enable grassroots hospitals to provide high-quality fistula establishment services, and promote balanced medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of the intelligent planning method for arteriovenous fistula based on hemodynamic simulation of the present invention;

[0073] Figure 2 This is a structural diagram of a building block of a vascular tree three-dimensional model of the present invention;

[0074] Figure 3 Schematic diagram of the multi-scale hemodynamic parameter adaptive acquisition system of the present invention;

[0075] Figure 4 It is a framework diagram of the improved smooth particle hydrodynamics method of the present invention;

[0076] Figure 5 This is a diagram of the deep correlation model architecture of the present invention for dynamic time-varying shear stress and vascular wall response;

[0077] Figure 6 This is a workflow diagram of the Bayesian optimization framework based on physical information enhancement of the present invention;

[0078] Figure 7 It is the overall architecture diagram of the system of the present invention;

[0079] Figure 8 This is an example diagram of a three-dimensional arteriovenous fistula model of the present invention;

[0080] Figure 9 This is a comparison chart of the virtual surgical planning and actual surgical results of the present invention;

[0081] Figure 10 Schematic diagram of the adaptive closed-loop feedback optimization system of the present invention. DETAILED DESCRIPTION

[0082] Please refer to the attached Figure 1-10 The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0083] like Figure 1 As shown, the intelligent planning method for arteriovenous fistula based on hemodynamic simulation provided by the present invention includes the following steps:

[0084] Step S1: Acquire 3D spatial anatomical data of the patient's arteries and veins and construct a 3D model of the vascular tree. This step first involves acquiring the patient's original 3D vascular data using medical imaging techniques such as high-resolution ultrasound or computed tomography angiography (CTA). For ultrasound imaging, preferably, a high-frequency probe of 7-12 MHz is used with a scanning resolution of 0.1 mm to ensure capture of subtle vascular structures. For CTA imaging, spiral CT is preferably used with a slice thickness of no more than 1 mm to obtain sufficiently detailed vascular anatomical information.

[0085] After obtaining the raw data, it is input into a two-stage deformable 3D convolutional neural network based on the self-attention mechanism for processing to obtain a 3D model of the vascular tree. Figure 2 As shown in Figure 2, the network has a two-stage structure: the first stage obtains a rough segmentation result of the blood vessels, and the second stage refines the blood vessel boundaries and branch regions. The self-attention mechanism is used to enhance the accuracy of identifying key blood vessel branch points, especially at the arteriovenous intersection, which is crucial for determining the subsequent location of the fistula anastomosis.

[0086] In a preferred embodiment of the present invention, the process of constructing the two-stage deformable 3D convolutional neural network based on the self-attention mechanism includes:

[0087] First, a two-stage deformable 3D convolutional neural network architecture based on a self-attention mechanism was constructed. This network architecture adopts an encoder-decoder structure. The encoder consists of four downsampling layers, each of which contains two 3D convolutional layers (with a kernel size of 3×3×3), a batch normalization layer, and a Reluctant Unit (ReLU) activation function. Deformable convolutions are introduced in the second, third, and fourth downsampling layers to enhance the network's adaptability to irregular vascular shapes. The self-attention module is located in the deepest layer of the encoder to capture global vascular structural information. The decoder consists of four upsampling layers, which use transposed convolutions to amplify feature maps and fuse features from corresponding encoder layers via skip connections.

[0088] Next, a training dataset containing 3D vascular data and corresponding labels was obtained. The training dataset included 3D vascular data from 500 different patients and their corresponding vascular segmentation labels. The labels, including vessel boundaries, centerlines, and branch point locations, were annotated and cross-validated by three experienced vascular surgeons to ensure label quality. 70% of the dataset was used for training, 15% for validation, and 15% for testing.

[0089] Finally, the training dataset is used to train the two-stage deformable 3D convolutional neural network based on the self-attention mechanism. The training process uses a combination of loss functions, including Dice loss and weighted cross entropy loss, to balance the segmentation accuracy of blood vessels and background regions. The loss function can be expressed as:

[0090] L total =α·L Dice +(1-α)·L WCE ,

[0091] Among them, α is the weight coefficient, which is 0.7 and is used to balance the importance of the two losses; L Dice Dice loss is used to measure the overlap between the predicted segmentation result and the true label. The calculation formula is:

[0092]

[0093] Among them, p i is the predicted probability of the i-th voxel, g i is the true label of the i-th voxel (0 or 1), N is the total number of primes; L WCE is the weighted cross entropy loss, which is used to deal with the imbalance problem between blood vessels and background. The calculation formula is:

[0094]

[0095] Among them, w pos is the weight of the positive sample (vascular region), which is set to the ratio of background to vascular voxels, usually about 4.0; w neg is the weight of negative samples (background areas), set to 1.0.

[0096] In the arteriovenous fistula planning scenario, accurate vascular segmentation is crucial for subsequent hemodynamic analysis. In particular, for small blood vessels and bifurcation areas, traditional segmentation methods are often difficult to accurately identify. For example, for the ulnar and radial arteries and cephalic veins with a diameter of less than 3 mm, these blood vessels are commonly used for the establishment of forearm fistulas, and their accurate segmentation directly affects the selection of fistula anastomosis positions. The dual-stage deformable network structure of the present invention can effectively solve this problem. It adapts to irregular blood vessel shapes through deformable convolution, especially for pathological changes such as vascular stenosis, tortuosity or twisting, and the accuracy is 10% to 15% higher than that of traditional convolutional networks.

[0097] Training was performed using the Adam optimizer, with an initial learning rate of 0.001 and a learning rate decay strategy, where the learning rate was reduced by a factor of 0.1 every 50 epochs. To prevent overfitting, weight decay (with a coefficient of 0.0001) and an early stopping strategy (training was terminated after 20 consecutive epochs of no improvement on the validation set) were employed. Training was performed on a workstation equipped with an NVIDIA RTX 3090 GPU, with a batch size of 4 and a total of 300 epochs.

[0098] Through this training process, a neural network model with high-precision vascular segmentation capabilities was developed. On the test set, this model achieved a Dice coefficient of over 95% for vascular segmentation, with a particularly high accuracy of 92% for identifying vascular branch regions, significantly outperforming traditional segmentation methods. In practical applications, this model can complete a 3D reconstruction of a patient's vascular tree in just 2-3 minutes, providing a foundation for subsequent fistula planning.

[0099] Step S2: Combine the vascular parameter data of the vascular tree 3D model with the hemodynamic parameter data at different blood flow rates and input them into the deep learning model to obtain the anastomotic dynamic shear force data. In this step, the vascular tree 3D model is first divided into different areas, including the anastomotic core area, anastomotic edge area, transition area and distal area, and different grid densities are assigned to each area. Figure 3 As shown in the figure, the multi-scale hemodynamic parameter adaptive acquisition system automatically allocates grid density based on regional importance: the grid density in the anastomotic core area is 10μm, the anastomotic edge area is 25μm, the transition area is 50μm, and the distal area is 100μm. This adaptive grid density allocation strategy significantly reduces overall computational complexity while ensuring computational accuracy in critical areas.

[0100] Next, computational fluid dynamics software (such as ANSYSCX) is used to calculate the pressure field, velocity field, and turbulence field of the vascular tree three-dimensional model, and the hemodynamic parameters under different blood flow rates are set. In actual applications, 5-7 different flow rate values are usually set to cover the range of blood flow changes from rest to exercise (such as 20 cm / s to 100 cm / s). For each flow rate value, the corresponding boundary conditions and physical parameters are set, including inlet flow, outlet pressure, blood density (about 1060 kg / m 3 ) and viscosity (about 0.0035 Pa·s).

[0101] Then, the hemodynamic parameters under different blood flow rates were input into the improved smoothed particle hydrodynamics (SPH) method for simulation to obtain the shear force data of the anastomotic area under different blood flow rates. Figure 4 As shown, the improved SPH method of the present invention includes the following key technical features:

[0102] (1) Based on the morphological characteristics of the 3D model of the vascular tree, an adaptive particle distribution is automatically generated. The particle spacing is set to approximately 10 μm in the core area of the anastomosis and gradually transitions to 100 μm towards the distal end to ensure sufficient calculation accuracy in critical areas;

[0103] (2) Increase the particle density in the anastomotic region to improve simulation accuracy. The particle number density at the anastomotic region is 2-3 times higher than that of the standard SPH method to capture the complex flow characteristics in this region;

[0104] (3) Adaptive kernel function is used to dynamically adjust the kernel function radius according to the local particle density. The kernel function radius h is calculated by the following formula:

[0105]

[0106] Where h is the kernel function radius at the current position, in μm; h0 is the base kernel function radius, which is usually 1.2 times the initial particle spacing, in μm; ρ0 is the reference density, which is the particle density at the initial uniform distribution, in particles / mm 3 ; ρ is the local particle density, the unit is number of particles / mm 3 The exponent 1 / 3 is derived from the inverse relationship between density and distance in three-dimensional space. This adaptive kernel function can expand its range in sparsely populated areas and improve local accuracy in densely populated areas.

[0107] (4) Use the improved virtual particle method to process the blood vessel wall boundary and improve the stability of the calculation at the boundary. On the basis of the traditional virtual particle method, the wall normal gradient correction term C is added grad :

[0108]

[0109] Among them, C grad is the dimensionless gradient correction factor; d is the distance from the particle to the wall, in μm; and d0 is the characteristic distance, which is the average interparticle distance, in μm. This correction term gradually reduces the gradient calculation weight near the wall, avoiding the computational instability problem of the traditional SPH method at the solid-liquid interface.

[0110] (5) Considering the elastic deformation response of the vascular wall, the fluid-structure interaction simulation is realized. The vascular wall adopts an elastic film model, and its deformation follows the following relationship:

[0111] σ=E·ε,

[0112] Where σ is stress, in Pa; E is Young's modulus, ranging from 0.5-1.5 MPa for arterial walls and 0.2-0.6 MPa for venous walls; and ε is strain, dimensionless. In actual calculations, the Young's modulus is dynamically adjusted based on the patient's age and vascular type (artery or vein). For example, for patients over 60 years old, the Young's modulus for arterial walls is typically taken at the upper limit (approximately 1.5 MPa), while for venous walls it is 0.4-0.6 MPa. For younger patients under 40 years old, the Young's modulus for arterial walls is typically 0.5-0.8 MPa, while for venous walls it is 0.2-0.3 MPa. This difference reflects the age-related decrease in vascular elasticity, which has a significant impact on the maturation process of fistulas.

[0113] The improved SPH method can be used to obtain the detailed shear force distribution in the anastomotic region at different blood flow rates. The wall shear force (WSS) is calculated using the following formula:

[0114]

[0115] Among them, τ w is the wall shear force, in Pa; μ is the blood dynamic viscosity, in Represents the velocity gradient at the wall (y = 0), in s -1 In the SPH framework, the velocity gradient is calculated by the velocity difference of adjacent particles:

[0116]

[0117] Among them, m j is the mass of the jth particle, in kg; ρ j is the density at the jth particle, in kg / m 3 ,u j and u i are the velocities of the jth particle and the i-th particle near the wall, respectively, in m / s; is the gradient of the kernel function, in units of m -4 ; r ij is the distance between two particles, in m.

[0118] In arteriovenous fistula planning, wall shear force is a key indicator for evaluating the rationality of anastomotic design. Clinical studies have shown that too low a shear force (<1.5Pa) can easily lead to intimal hyperplasia and stenosis, while too high a shear force (>40Pa) may cause vascular damage and thrombosis. The ideal anastomotic design should ensure that the shear force is evenly distributed and maintained in a moderate range (approximately 4-10Pa). The SPH method of the present invention can accurately capture the shear force distribution characteristics under different anastomotic angles, providing a quantitative basis for selecting the optimal angle.

[0119] Finally, the vascular parameter data was used as input to the deep learning model to obtain dynamic anastomotic shear force data. The vascular parameter data includes 27 characteristic parameters such as vessel diameter, wall thickness, elastic modulus, and branching angle. The deep learning model uses a temporal convolutional network (TCN) structure to effectively capture the temporal pattern of time-varying shear force. The mathematical expression of the TCN model is:

[0120] F(x)=ReLU(W f *x+b f ),

[0121] Where F(x) is the feature map output; x is the input sequence, the dimension is (batch size, sequence length, number of features); W fis the convolution kernel weight matrix; * represents the convolution operation; b f is the bias term; ReLU is the activation function, defined as ReLU(z)=max(0,z).

[0122] To capture long-term dependencies, TCN uses an expanded convolution structure:

[0123] F d (x) = ReLU(W d * d x+b d ),

[0124] Among them, F d (x) is the dilated convolution output; * d Represents a convolution operation with a dilation factor of d. The dilation factor controls the size of the receptive field and grows exponentially with the number of layers (e.g., 1, 2, 4, 8).

[0125] The data used to train the model includes 1000 sets of paired data of different vascular parameters and corresponding SPH simulation results. The loss function uses a combination of mean square error and smooth L1 loss:

[0126] L=β·MSE+(1-β)·SmoothL1,

[0127] Among them, β is the weight coefficient, which is set to 0.6; MSE is the mean square error; SmoothL1 is the smooth L1 loss, which is insensitive to outliers.

[0128] This deep learning model is invaluable in fistula planning scenarios. For example, for patients requiring evaluation of multiple potential anastomosis locations, traditional methods require a complete CFD simulation of each location, which can take hours. However, this model predicts the shear force distribution at each location in seconds, enabling physicians to quickly identify the optimal option. The model's prediction error is kept within 7%, meeting clinical decision-making requirements.

[0129] After training, the deep learning model can directly predict the dynamic shear force distribution in the anastomotic area based on the input vascular parameters, avoiding the need for time-consuming SPH full flow field simulation each time, reducing the calculation time from several hours to a few seconds, and greatly improving the system response speed.

[0130] Step S3: Compare the dynamic shear force data of the anastomosis with the preset optimal anastomosis angle data to determine the optimal anastomosis angle. In this step, a deep correlation model of dynamic time-varying shear force and vascular wall response is first constructed, including a dual-stream feature extraction network and a multi-scale fusion module. Figure 5As shown in Figure 2, the model has a two-stream architecture: one stream processes time-varying shear force data, and the other processes vascular wall property data. The time-varying shear force feature extraction branch utilizes a three-layer temporal convolutional module, each layer consisting of dilated convolutions, residual connections, and layer normalization. The vascular wall feature extraction branch uses a multi-layer perceptron architecture to extract vascular wall thickness, elasticity, and geometric features. These two feature streams are fused through a cross-attention mechanism to form an associative representation of shear force and wall response.

[0131] The mathematical expression of the cross attention mechanism is:

[0132]

[0133] Among them, A(Q,K,V) is the attention output; Q is the query matrix, which comes from the first-way feature; K is the key matrix, which comes from the second-way feature; V is the value matrix, which also comes from the second-way feature, and T is the matrix transpose; d k is the dimension of the key vector; softmax is the normalization function defined as

[0134] Dynamic anastomotic shear force data were input into the deep correlation model described above to predict long-term vascular wall response at different anastomotic angles. The model outputs key indicators, including the degree of endothelial cell proliferation, the rate of lumen diameter change, and the probability of patency, which are directly related to the maturation time and long-term function of the fistula.

[0135] Next, a Bayesian optimization framework enhanced by physical information was used to evaluate each candidate anastomotic angle scheme and calculate a multi-objective score including surgical success rate, fistula maturation time and long-term patency rate. Figure 6 As shown in the figure, the core of this optimization framework is to integrate hemodynamic theory and clinical experience to establish a theoretical correlation model between anastomotic angle and fistula function.

[0136] In a preferred embodiment of the present invention, the Bayesian optimization framework based on physical information enhancement includes:

[0137] (1) Integrating hemodynamic theory and clinical experience, a theoretical correlation model between anastomotic angle and fistula function was established. This model takes into account the effect of anastomotic angle on flow field distribution, wall shear force, and thrombosis risk. For example, studies have shown that an end-to-side anastomotic angle of 25°-45° can usually achieve a good balance between flow stability and shear force distribution;

[0138] (2) Define physical feasibility constraints, including angle range and spatial position restrictions. Considering the operability of the surgery, the anastomosis angle is usually limited to 15°-60°; the spatial position needs to take into account the local anatomical structure and avoid important tissues such as peripheral nerves and tendons;

[0139] (3) Construct a multi-objective function and dynamically adjust the weight coefficients of short-term goals, medium-term goals, and long-term goals. The expression of the multi-objective function is:

[0140] F(θ)=w1·f1(θ)+w2·f2(θ)+w3·f3(θ),

[0141] Among them, F(θ) is a comprehensive scoring function with a score range of 0–1, where a higher score indicates a better solution; θ represents the anastomotic angle in degrees (°); f1(θ) represents the uniformity of the anastomotic shear force, and the calculation formula is:

[0142]

[0143] Among them, σ τ is the standard deviation of shear force in the anastomotic area, in Pa; is the average shear force, in Pa. This indicator reflects the success rate of the operation. Uniform shear force distribution is beneficial to reducing the risk of vascular damage caused by local high shear areas;

[0144] f2(θ) represents the prediction of fistula maturation time, and the calculation formula is:

[0145]

[0146] Among them, T m (θ) is the predicted maturation time of the fistula, in days; T0 is the reference time, which is 42 days (6 weeks, the ideal maturation time considered clinically). The closer this index is to 1, the closer the predicted maturation time is to the ideal value;

[0147] f3(θ) represents the long-term patency prediction, which directly uses the 6-month patency probability value output by the deep association model, ranging from 0 to 1;

[0148] w1, w2, and w3 are dynamic weight coefficients that are automatically adjusted based on individual patient characteristics to ensure w1 + w2 + w3 = 1. For elderly patients (>65 years), w1 = 0.5, w2 = 0.3, and w3 = 0.2 are typically set, prioritizing short-term surgical success. For younger patients (<45 years), w1 = 0.3, w2 = 0.2, and w3 = 0.5 are typically set, prioritizing long-term patency.

[0149] In fistula planning practice, different patient groups prioritize different objectives significantly. For example, for diabetic patients, their blood vessels are often more fragile and take longer to mature, so the weights are adjusted to w1 = 0.4, w2 = 0.4, and w3 = 0.2, placing greater emphasis on the maturation process. Meanwhile, for patients requiring urgent dialysis, short-term success rates may be prioritized, with w1 = 0.6, w2 = 0.3, and w3 = 0.1. This personalized weighting adjustment enables the system to generate the most appropriate plan based on the patient's specific circumstances.

[0150] (4) Initialize the candidate matching angle solutions and use the Gaussian process regression model to predict the performance of each solution. The Gaussian process regression model is defined as:

[0151]

[0152] Where f(θ) represents the objective function to be optimized; represents a Gaussian process; m(θ) is the mean function, which is initially set to a constant; k(θ,θ′) is the kernel function, using the Matérn5 / 2 kernel:

[0153]

[0154] Where σ is the signal standard deviation, typically initialized to 0.5; l is the length scale parameter that controls the rate at which correlation decays, initially set to 5°; and |θ - θ′| represents the absolute difference between two angles, expressed in degrees. The σ and l parameters are continuously updated during the optimization process using maximum likelihood estimation.

[0155] The Matérn 5 / 2 kernel excels at describing smooth but non-infinitely differentiable functions common in physical systems. It is particularly well-suited for modeling biological systems with discontinuities, such as vascular responses. In practical applications, this kernel function better captures hemodynamic fluctuations caused by changes in anastomosis angle than the standard square exponential kernel.

[0156] (5) Based on the expected improvement and uncertainty trade-off, iterative optimization is performed until convergence to the optimal solution. An acquisition function is used to guide the optimization process. The present invention uses the expected improvement (EI) acquisition function:

[0157] EI(θ)=E[max(f(θ)-f(θ + ),0)],

[0158] Where EI(θ) is the expected improvement; E represents the expected operation; f(θ) is the predicted value of the objective function at angle θ; f(θ + ) is the currently known optimal solution. The EI function can be explicitly calculated as:

[0159]

[0160] where μ(θ) and σ(θ) are the predicted mean and standard deviation of the Gaussian process at θ, respectively; Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, respectively. The acquisition function achieves large values in regions of high predicted value and high uncertainty, balancing exploration and exploitation.

[0161] During the fistula planning optimization process, a wide range of angles between 15° and 60° are initially explored. As iterations progress, the optimization focus gradually converges on promising areas. For example, the system might first try angles of 15°, 30°, 45°, and 60°. Once the 30°-45° range is found to perform well, more intensive sampling will be performed in that range to ultimately determine the optimal angle.

[0162] Through the above-mentioned Bayesian optimization process, the optimal anastomosis angle can usually be converged within 20-30 iterations. Ultimately, the optimal anastomosis angle is determined based on the multi-objective score and the patient's specific needs. In practical applications, the common optimal anastomosis angle range is 30°-40°, but the specific value will be adjusted according to individual differences. For example, for patients with hypertension, the optimal angle is usually smaller (25°-35°) to reduce turbulence and vascular wall stress; for patients with slower blood flow velocities, a larger angle (35°-45°) may be required to promote sufficient blood flow mixing and shear force stimulation.

[0163] Step S4: Combine the optimal anastomosis angle with the 3D model of the vascular tree to perform 3D modeling of the arteriovenous fistula procedure, and obtain a 3D arteriovenous fistula model. After determining the optimal anastomosis angle, this step applies it to the 3D model of the vascular tree to perform 3D modeling of the arteriovenous fistula procedure. The modeling process takes into account the actual operational factors during the surgery, including:

[0164] (1) Shape of vascular incision: usually oval or diamond-shaped, with the long axis approximately 1.5-2 times the diameter of the vessel;

[0165] (2) Anastomosis technique: set as continuous suture or interrupted suture, with suture spacing of approximately 0.5-1 mm;

[0166] (3) Spatial position: Consider local anatomical limitations to ensure the feasibility of the anastomosis operation.

[0167] The 3D modeling uses a parametric approach to automatically generate the anastomotic structure based on parameters such as the optimal anastomotic angle, vessel diameter, and anastomotic length. For common end-to-side anastomosis, the modeling process can be expressed as:

[0168] V AVF =P(θ opt ,da ,d v ,l a ,S G ),

[0169] Among them, V AVF is the generated three-dimensional model of the fistula; P represents the parametric modeling operation; θ opt is the best anastomosis angle; d a and d v are the diameters of arteries and veins, respectively; l a is the length of the anastomosis; S G It is the geometric constraint of anatomical space.

[0170] For end-to-side anastomosis, the model first creates an elliptical incision on the venous side. The arterial end is then aligned with the incision at the optimal anastomotic angle, and finally connected with sutures to form a complete anastomotic structure. In actual implementation, the anastomotic length is typically set to 1.5-2 times the arterial diameter, a ratio derived from hemodynamic optimization results and clinical experience.

[0171] After modeling was completed, the quality of the model was evaluated by checking the anastomotic morphology, lumen continuity, and rationality of spatial position.

[0172] Evaluation metrics include:

[0173] Q geom =w a ·A ratio +w c ·C score +w s ·S access ,

[0174] Among them, Q geom Score geometric quality; A ratio The ratio of the anastomotic area to the arterial cross-sectional area is ideally between 1.5 and 2.0. score Score for lumen continuity, based on the curvature continuity at the anastomosis; access Scoring surgical accessibility, taking into account surrounding tissue limitations; w a 、w c and w s are the corresponding weight coefficients, usually set to 0.4, 0.3 and 0.3.

[0175] If necessary, fine-tuning is performed to meet the actual needs of the operation. For example, if the position of the anastomosis is restricted by local anatomy (such as adjacent nerves or tendons), the position or angle is adjusted appropriately to ensure the safety of the operation. The final three-dimensional arteriovenous fistula model is as follows: Figure 8 As shown, it contains complete postoperative vascular morphology and anastomotic structure details.

[0176] In clinical practice, different types of fistulas (such as radial artery-cephalic vein, brachial artery-cephalic vein, etc.) have their own modeling characteristics. For example, for forearm fistulas, due to the limited local anatomical space, special attention must be paid to the relationship between the anastomosis and surrounding structures during modeling. For upper arm fistulas, on the other hand, there may be a risk of excessive blood flow, and the size of the anastomosis must be controlled during modeling to avoid excessive blood flow load. The parametric modeling method of the present invention can be flexibly adjusted according to different situations to meet diverse clinical needs.

[0177] Step S5: Based on the 3D arteriovenous fistula model, a virtual surgery is performed to obtain virtual surgical process data and postoperative hemodynamic status assessment results. In this step, a virtual surgical scene is constructed based on the optimal anastomosis angle and the 3D model of the vascular tree. The virtual surgical scene includes elements such as the patient's anatomy, surgical instruments, and operation steps, and is presented using 3D visualization technology. The scene construction formula is:

[0178] S virtual =C(V AVF ,A patient ,I surgical ),

[0179] Among them, S virtual is a virtual surgical scene; C represents the scene construction operation; V AVF A is a three-dimensional arteriovenous fistula model; patient Patient anatomical structure data; I surgical Information about surgical instruments and operating procedures.

[0180] The surgical process is simulated to generate postoperative vascular morphology. The simulation follows the standard surgical procedure: vascular exposure, vascular incision, anastomosis, and blood flow restoration. Key operating parameters such as incision location, suturing method, and tension control are recorded at each step. The generated postoperative vascular morphology takes into account the local deformation and stress state caused by the surgical operation. The deformation model can be expressed as:

[0181] V post =D(V AVF ,T surgical ,E tissue ),

[0182] Among them, V post is the deformed vascular morphology after surgery; D represents the deformation calculation operation; T surgical is the surgical tension parameter; E tissue The elastic modulus is often used to represent the elastic properties of tissues. The elastic modulus of arterial walls ranges from 0.5 to 1.5 MPa, and that of venous walls ranges from 0.2 to 0.6 MPa.

[0183] The fistula status at different time periods is predicted, including the status after one week, one month, and six months. The prediction is based on a physical-biological coupling model that considers the vascular wall remodeling process under hemodynamic stimulation. The remodeling model can be expressed as:

[0184] V t =R(V post ,τ w ,t,P patient ),

[0185] Among them, V t is the state of the blood vessel after time t; R represents the reconstruction calculation operation; τ w is the wall shear force distribution; t is time, in days; P patient are individual patient parameters, including age, underlying diseases, etc. The vascular dilation rate during the remodeling process has a nonlinear relationship with the wall shear force:

[0186]

[0187] in, is the rate of change of vascular diameter, in mm / day; k is the proportional coefficient, which is adjusted according to the patient's age, with a value of 0.03-0.05 for young patients (<40 years old) and 0.01-0.02 for elderly patients (>65 years old); τ w is the wall shear force, in Pa; τ0 is the reference shear force, usually 1.5 Pa; α is the nonlinear exponent, with a value of about 0.4; f(D,D max ) is a limiting function to ensure that the expansion does not exceed the physiological limit D max .

[0188] Vascular dilation is a key indicator in the maturation of fistulas. The above formula reflects the phenomenon observed clinically: moderate shear force promotes vascular dilation, and when the vascular dilation reaches a certain degree, the rate of dilation slows down and eventually reaches a state of equilibrium. For example, for a typical cephalic vein fistula, the initial diameter is about 3-4mm, which can reach 6-8mm after maturation, but rarely exceeds 10mm. This upper limit is determined by D max control.

[0189] Calculate the predicted values of internal fistula maturation time and long-term patency rate. The internal fistula maturity criteria are: vessel diameter ≥ 6mm, blood flow ≥ 600ml / min, and skin depth ≤ 6mm. Based on the predicted vascular expansion rate and blood flow growth curve, calculate the time required to reach maturity criteria:

[0190] T mature =min{t|D(t)≥6mm∧Q(t)≥600ml / min∧Depth(t)≤6mm},

[0191] Among them, Tmature is the maturation time in days; D(t), Q(t) and Depth(t) are the blood vessel diameter, blood flow and skin depth at time t, respectively.

[0192] Long-term patency prediction is based on a comprehensive assessment of various risk factors, including shear force distribution, vascular elasticity, and the patient's underlying disease. The patency probability model can be expressed as:

[0193]

[0194] Among them, P patency (t) is the probability of patency at time t; h(s, X) is the risk function, which is related to time s and the risk factor vector X. Risk factors include wall shear force heterogeneity, vessel tortuosity, patient age, and comorbidities.

[0195] In fistula planning practice, maturation time and patency are two core metrics for evaluating the effectiveness of treatment options. For example, for patients requiring urgent dialysis, a treatment with a shorter maturation time is more advantageous; whereas for young, long-term dialysis patients, a treatment with a higher patency rate may be more important. This system's predictive model can provide quantitative estimates of these key metrics, helping physicians make more informed decisions.

[0196] Finally, a visual surgical guidance plan is generated, including anastomosis angles, position markers, and key operation points. The guidance plan is presented in an intuitive 3D image, with key anatomical landmarks and surgical reference points marked. Figure 9 The comparison between virtual surgical planning and actual surgical results was demonstrated, verifying the accuracy and reliability of this method.

[0197] In clinical applications, virtual surgical guidance has significantly improved surgical precision. For example, with traditional methods, doctors often rely on experience to estimate the anastomotic angle, resulting in an error of up to ±10° between the actual and target angles. However, with the system's visual guidance, the error can be controlled within ±3°, significantly improving surgical repeatability and success rates.

[0198] Step S6: Adaptive closed-loop feedback optimization, such as Figure 10 As shown, the present invention also includes an adaptive closed-loop feedback optimization step, which is crucial to the performance of the entire system. Specifically, it includes:

[0199] First, actual surgical parameters and postoperative follow-up data were recorded. Surgical parameters included actual anastomotic angle, position, and technical details; follow-up data included fistula maturation time, blood flow changes, and complications. Data collection used standardized forms to ensure consistency and comparability across centers. The data structure can be represented as follows:

[0200] D feedback ={(P i ,Si ,F i )|i=1,2,...,N},

[0201] Among them, D feedback is the feedback data set; P i is the characteristic data of the i-th patient; S i is the corresponding surgical parameter; F i N is the total number of cases.

[0202] Secondly, the predicted results are compared with the actual results to quantify the prediction accuracy. For quantitative indicators (such as blood flow), the relative error between the predicted value and the actual value is calculated; for qualitative indicators (such as patency), the prediction accuracy, sensitivity, and specificity are calculated. The evaluation index can be expressed as:

[0203]

[0204] Where Acc is accuracy; Sen is sensitivity; Spe is specificity; TP, TN, FP, and FN are the number of true positive, true negative, false positive, and false negative samples, respectively. For patency prediction, 6 months of patency is usually considered a positive event.

[0205] Again, based on the difference between the predicted and actual results, the model parameters are updated. For deep learning models, an incremental learning method is used to retain the original knowledge while integrating new data. The loss function of incremental learning is:

[0206] L inc =L new +λ·L dist ,

[0207] Among them, L inc is the total loss of incremental learning; L new is the prediction loss of new data; L dist λ is the knowledge distillation loss to prevent catastrophic forgetting; λ is the balance coefficient, typically set to 0.5. For the physical model, relevant physical parameters are adjusted to reduce systematic errors. The model update cycle is set to every three months or when the cumulative number of new cases reaches 50.

[0208] Then, the new case data is added to the knowledge base and the sampling strategy is optimized. The knowledge base adopts a hierarchical storage structure and is classified and archived according to patient characteristics, surgical plan, and effect evaluation. The sampling strategy prioritizes data in areas with high error rates, focusing resources on improving the areas that need the most improvement. The sampling probability formula is:

[0209] P(sample i )∝exp(γ·err i ),

[0210] Among them, P(samplei ) is the probability of selecting the i-th sample; err i is the prediction error of the corresponding sample; γ is the temperature parameter, which controls the degree of sampling bias and is usually set to 2.0.

[0211] Finally, we conduct anomaly detection and analysis on cases where predictions fail to work, to improve the robustness of the system. For cases where predictions deviate significantly from reality, we conduct in-depth analysis of the causes and identify potential special factors or model blind spots. The anomaly determination criteria are:

[0212] |pred i -actual i |>μ err +2σ err ,

[0213] Among them, pred i and actual i are the predicted value and actual value of the i-th case respectively; μ err and σ err are the mean and standard deviation of historical errors, respectively.

[0214] In practical applications, closed-loop feedback mechanisms are crucial for continuous improvement of system performance. For example, the system might discover that the predicted time to fistula maturation is generally too short for diabetic patients. By analyzing these abnormal patterns, the system can automatically adjust the weights of diabetes-related parameters or add specific diabetes-influencing factors to the model, thereby improving prediction accuracy for this special population.

[0215] Through this closed-loop feedback mechanism, the system continuously learns and optimizes, with prediction accuracy improving over time. In actual applications, after learning from approximately 500 cases, the system achieved an overall prediction accuracy of over 85%, providing reliable support for clinical decision-making. Compared to traditional methods that rely on physician experience (success rate of approximately 60% to 65%), this system significantly improves the success rate and long-term effectiveness of fistula surgery.

[0216] like Figure 7 As shown, the intelligent planning system for arteriovenous fistula based on hemodynamic simulation provided by the present invention includes the following modules:

[0217] The vascular tree three-dimensional model construction module 10 is used to obtain the three-dimensional spatial anatomical data of the patient's arteries and veins and construct a vascular tree three-dimensional model. The module includes a data acquisition unit 11 and a model construction unit 12.

[0218] The data acquisition unit 11 is responsible for acquiring raw 3D vascular data from high-resolution ultrasound or computed tomography angiography. In practice, this unit can connect to existing hospital imaging equipment (such as ultrasound, CT, or MRI) via a DICOM standard interface to automatically acquire patient imaging data. To ensure data quality, ultrasound imaging uses a high-frequency probe of 7-12 MHz, achieving a scanning resolution of 0.1 mm; CTA imaging uses spiral CT with a slice thickness of no more than 1 mm.

[0219] Model construction unit 12 processes the acquired raw 3D vascular data into a two-stage deformable 3D convolutional neural network based on a self-attention mechanism to generate a 3D vascular tree model. This unit utilizes the neural network architecture detailed in Example 1, utilizing GPU acceleration for efficient processing. For standard-resolution vascular data (e.g., 512×512×200 voxels), model construction typically completes within 2–3 minutes, achieving a Dice coefficient accuracy exceeding 95%.

[0220] In clinical applications, this module addresses the inaccuracy of traditional vessel segmentation methods when processing images with uneven quality and low contrast. For example, traditional methods often segment or miss small vessels as small as 2mm in diameter or complex bifurcations. However, this module, through deformable convolution and self-attention mechanisms, accurately identifies these key structures, providing a reliable foundation for subsequent analysis.

[0221] The hemodynamic simulation module 20 combines the vascular parameter data from the 3D vascular tree model with the hemodynamic parameter data at different blood flow rates, inputting them into a deep learning model to obtain dynamic shear force data at the anastomotic site. This module includes a region division unit 21, a fluid calculation unit 22, an SPH simulation unit 23, and a deep learning prediction unit 24.

[0222] The region segmentation unit 21 is responsible for dividing the 3D vascular tree model into distinct regions, including the anastomotic core, anastomotic margin, transition zone, and distal region, and assigning different mesh densities to each region. The segmentation algorithm automatically identifies potential anastomotic locations and labels them based on vascular anatomical features and clinical experience. A multi-scale meshing strategy is employed, with a mesh density of 10 μm in the anastomotic core, 25 μm in the anastomotic margin, 50 μm in the transition zone, and 100 μm in the distal region.

[0223] The fluid calculation unit 22 uses computational fluid dynamics software to calculate the pressure, velocity, and turbulence fields of the three-dimensional vascular tree model, setting hemodynamic parameters for different blood flow rates. This unit dynamically allocates computing resources based on computational needs and utilizes distributed computing to accelerate processing for complex vascular structures. Typically, 5-7 different flow rate values (e.g., 20 cm / s to 100 cm / s) are set to cover the range of blood flow changes in patients from resting to exercise.

[0224] The SPH simulation unit 23 inputs the hemodynamic parameters at different blood flow rates into an improved smoothed particle hydrodynamics method for simulation, obtaining shear force data for the anastomotic region at different blood flow rates. This unit implements the improved SPH method detailed in Example 1, significantly increasing simulation speed through GPU parallel computing. Compared to traditional grid-based CFD, the SPH method has inherent advantages in handling large deformations and fluid-structure interaction problems, making it particularly suitable for simulating the complex flow fields at the anastomotic site of an internal fistula.

[0225] The deep learning prediction unit 24 uses vascular parameter data as input to a deep learning model to generate dynamic anastomotic shear force data. This unit utilizes a pretrained temporal convolutional network model, enabling predictions within seconds and significantly improving system response speed. The model input dimension is 27 (the number of vascular characteristic parameters), and the output is time-series shear force data with a typical size of 32 × 32 × 100 (32 × 32 spatial grid, 100 time steps).

[0226] The core innovation of this module lies in its integration of traditional fluid dynamics calculations with deep learning predictions, ensuring both physical accuracy and overcoming computational efficiency bottlenecks. For example, when comparing multiple scenarios, traditional methods often require a complete CFD simulation for each, taking hours. However, this system, through deep learning predictions, reduces calculation time to seconds while maintaining an error within 7%, enabling real-time clinical decision-making.

[0227] The optimal anastomotic angle determination module 30 is used to compare the anastomotic dynamic shear force data with the preset optimal anastomotic angle data to determine the optimal anastomotic angle. This module includes a correlation modeling unit 31, a multi-objective evaluation unit 32, and a solution generation unit 33.

[0228] The Correlation Modeling Unit 31 constructs a deep correlation model between dynamic, time-varying shear stress and vascular wall response. This unit, comprising a dual-stream feature extraction network and a multi-scale fusion module, processes shear stress data from the hemodynamic simulation module and correlates shear stress patterns with long-term vascular wall responses. The model employs a dual-stream architecture, processing time-varying shear stress data and vascular wall property data separately, fusing the two features using a cross-attention mechanism.

[0229] The multi-objective evaluation unit 32 evaluates each candidate anastomotic angle using a Bayesian optimization framework enhanced by physical information, calculating a multi-objective score encompassing surgical success rate, fistula maturation time, and long-term patency. This evaluation process fully considers individual patient characteristics and dynamically adjusts the weights of each objective. The optimization framework utilizes a Gaussian process regression model with a Matérn 5 / 2 kernel, combined with an expected improvement acquisition function to guide the search process.

[0230] The protocol generation unit 33 determines the optimal anastomotic angle based on the multi-objective score and the patient's specific needs, and generates a detailed protocol description. The protocol description includes the angle value (usually accurate to 1°), a position description, and special considerations. For typical cases, the optimal anastomotic angle is usually in the range of 30°-40°, but it will be personalized based on patient characteristics (such as age, underlying diseases) and vascular properties.

[0231] The unique value of this module lies in its intelligent transformation from physical simulation to clinical decision-making. Traditional fistula planning relies primarily on physician experience and lacks quantitative standards. However, this module, through a multi-objective optimization framework, integrates hemodynamic indicators, surgical technique requirements, and long-term outcome predictions into a unified decision-making system, making fistula planning a quantifiable and optimizable process. In clinical validation, the system's recommended anastomotic angle schemes have improved the success rate by 25% to 30% compared to traditional empirical methods.

[0232] The 3D fistula modeling module 40 is used to perform 3D modeling of the arteriovenous fistula procedure based on the 3D vascular tree model in combination with the optimal anastomotic angle, thereby obtaining a 3D arteriovenous fistula model. This module uses a parametric modeling approach to automatically generate the anastomotic structure based on parameters such as the optimal anastomotic angle, vessel diameter, and anastomotic length.

[0233] For common end-to-side anastomosis, the modeling process begins by creating an elliptical incision on the venous side, with the long axis typically 1.5-2 times the diameter of the vein. The arterial end is then aligned with the incision at the optimal anastomotic angle. Finally, a simulated suture connects the two to form a complete anastomotic structure. The model also considers factors such as suture tension and vascular deformation to more realistically reflect the postoperative morphology.

[0234] After modeling is complete, the system automatically assesses the model's quality, checking the anastomotic morphology, lumen continuity, and spatial rationality. If necessary, fine-tuning is performed to meet the actual surgical requirements. For example, if the anastomotic position is restricted by local anatomy, the position or angle will be adjusted appropriately to ensure safe operation.

[0235] The advantage of this module lies in its integration of virtual surgical planning with actual anatomical constraints, generating fistula models that meet both hemodynamic optimization requirements and surgical feasibility. This balance between ideal and practical is crucial for clinical success. In practical applications, this module can flexibly adjust modeling parameters based on different fistula types (e.g., radial-cephalic vein, brachial-cephalic vein, etc.) to accommodate diverse clinical needs.

[0236] The virtual surgery module 50 is used to perform virtual surgery based on a 3D arteriovenous fistula model, obtaining virtual surgery process data and postoperative hemodynamic status assessment results. This module includes a scenario construction unit 51, a process simulation unit 52, a status prediction unit 53, and a guidance generation unit 54.

[0237] The scenario construction unit 51 constructs a virtual surgical scene based on the optimal anastomosis angle and the 3D model of the vascular tree, including elements such as the patient's anatomy, surgical instruments, and procedure steps. This scene construction utilizes 3D visualization technology to provide an immersive surgical environment. In practice, the surgical scene also includes surrounding critical anatomical structures (such as nerves and tendons) to help surgeons avoid potential risks.

[0238] The process simulation unit 52 simulates the surgical process and generates postoperative vascular morphology. The simulation follows the standard surgical workflow: vascular exposure, vascular incision, anastomosis, and blood flow restoration. Key operational parameters are recorded at each step, accounting for local deformation and stress caused by the surgical procedure. The simulation results display the immediate vascular state after the surgical intervention, including anastomotic morphology and initial blood flow distribution.

[0239] The state prediction unit 53 predicts the state of the fistula at different stages and calculates the predicted values for the fistula's maturation time and long-term patency rate. This prediction is based on a coupled physical-biological model, taking into account the vascular wall remodeling process under hemodynamic stimulation. This unit can simulate the entire process from fistula establishment to maturation, predicting changes in vascular morphology and function after one week, one month, and six months. Under standard conditions, the prediction error for fistula maturation time is controlled within ±10 days, and the prediction accuracy for patency rate reaches 80%.

[0240] The guidance generation unit 54 generates a visual surgical guidance plan, including anastomosis angles, location markers, and key surgical procedures. The guidance plan is presented as an intuitive 3D image, annotated with key anatomical landmarks and surgical reference points. For example, the system clearly indicates the ideal vascular incision location, incision length, suture point distribution, and areas requiring special attention, enabling the surgeon to precisely execute the surgery according to the plan.

[0241] This module transforms static planning into dynamic surgical guidance through virtual surgery and predictive simulation. It not only informs surgeons of the optimal approach but also demonstrates how to implement it and the expected results, significantly improving the operability and predictability of planning results. Clinical practice has demonstrated that the use of virtual surgical guidance significantly reduces surgical deviations, reducing the error between the actual and planned anastomotic angles from the traditional ±10° to ±3°, and improving surgical accuracy by over 70%.

[0242] The adaptive closed-loop feedback module 60 is used to record actual surgical parameters and postoperative follow-up data, compare predicted results with actual results, update model parameters, and improve the system's prediction accuracy. This module includes a data collection unit 61, a result comparison unit 62, a model update unit 63, and an anomaly analysis unit 64.

[0243] Data Collection Unit 61 is responsible for recording actual surgical parameters and postoperative follow-up data. Data collection utilizes standardized forms to ensure consistency and comparability across centers. Collected data includes surgical details (such as actual anastomotic angle, position, and technique) and follow-up results (such as fistula maturation time, blood flow changes, and complications), forming a structured database to support subsequent analysis.

[0244] The result comparison unit 62 compares the predicted results with the actual results to quantify the prediction accuracy. This comparison uses multiple evaluation metrics, such as relative error, accuracy, sensitivity, and specificity, to comprehensively assess the model's predictive performance. This evaluation not only reflects the overall prediction level but also identifies prediction deviations in specific patient groups or fistula types, providing guidance for targeted model optimization.

[0245] Model update unit 63 updates model parameters based on the discrepancy between predicted and actual results and adds new case data to the knowledge base. Model updates utilize an incremental learning approach, preserving existing knowledge while integrating new data. Update strategies include parameter fine-tuning, feature weight adjustment, and model structure optimization, ensuring the system can adapt to evolving clinical conditions.

[0246] The anomaly analysis unit 64 detects and analyzes anomalies in prediction failure cases to improve system robustness. The analysis results are used to adjust the model structure or add special case handling mechanisms, continuously improving the system's adaptability to diverse situations. For example, if predictions for a particular patient type (such as elderly patients with diabetes) are found to have generally large deviations, the system will automatically increase the weights of related features or add specific adjustment factors to improve prediction accuracy for this patient type.

[0247] This module is a key component of the system's continuous improvement. Through a closed-loop mechanism of planning, execution, feedback, and optimization, it achieves self-learning and evolutionary capabilities. Clinical data show that after learning from approximately 500 cases, the system's prediction accuracy increased from an initial 70% to over 85%, far exceeding the 60% to 65% achieved by traditional empirical methods. This continuous learning capability enables the system to continuously adapt to new patient populations and clinical situations, maintaining long-term effectiveness.

[0248] Through this system architecture, the present invention intelligently manages the entire process, from acquiring individual patient vascular data to generating the optimal surgical plan, providing an efficient and accurate solution for fistula planning. Compared with traditional methods, this system not only significantly improves the success rate and long-term outcomes of fistula surgery, but also, through standardization and intelligentization, reduces reliance on highly skilled specialists, enabling even primary care hospitals to provide high-quality fistula placement services and promoting balanced utilization of medical resources.

[0249] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. An intelligent planning method for arteriovenous fistula based on hemodynamic simulation, characterized by: include: Obtain the three-dimensional anatomical data of the patient's arteries and veins and construct a three-dimensional model of the vascular tree; Combining the vascular parameter data of the three-dimensional vascular tree model with the hemodynamic parameter data at different blood flow rates, and inputting the data into a deep learning model to obtain the dynamic shear force data of the anastomosis; Comparing the anastomotic dynamic shear force data with the preset optimal anastomotic angle data to determine the optimal anastomotic angle; Performing three-dimensional modeling of an arteriovenous fistula procedure on the three-dimensional model of the vascular tree in combination with the optimal anastomosis angle to obtain a three-dimensional arteriovenous fistula model; Based on the three-dimensional arteriovenous fistula model, a virtual surgery is performed to obtain virtual surgery process data and postoperative hemodynamic status evaluation results.

2. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 1, characterized in that: Acquiring three-dimensional anatomical data of the patient's arteries and veins and constructing the three-dimensional model of the vascular tree specifically includes: Acquire original vascular 3D data from high-resolution ultrasound or computed tomography angiography; Inputting the original three-dimensional blood vessel data into a two-stage deformable 3D convolutional neural network based on a self-attention mechanism for processing to obtain the three-dimensional blood vessel tree model; Among them, the self-attention mechanism is used to enhance the recognition accuracy of key branch points of blood vessels.

3. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 2, characterized in that: The construction process of the two-stage deformable 3D convolutional neural network based on the self-attention mechanism includes: A two-stage deformable 3D convolutional neural network architecture based on the self-attention mechanism is constructed. The first stage obtains the rough segmentation results of blood vessels, and the second stage refines the blood vessel boundaries and branch regions. Acquire a training dataset comprising three-dimensional blood vessel data and corresponding labels, wherein the labels include blood vessel boundaries, centerlines, and branch point location information; The training data set is used to train the two-stage deformable 3D convolutional neural network based on the self-attention mechanism to obtain a trained neural network model.

4. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 1, characterized in that: The vascular parameter data of the 3D vascular tree model and the hemodynamic parameter data at different blood flow rates are combined and input into a deep learning model to obtain the anastomotic dynamic shear force data, specifically including: Dividing the vascular tree three-dimensional model into different regions, including anastomotic core region, anastomotic edge region, transition region, and distal region, and assigning different mesh densities to each region; Computational fluid dynamics software is used to calculate the pressure field, velocity field, and turbulence field of the three-dimensional vascular tree model, and hemodynamic parameters under different blood flow rates are set; Inputting the hemodynamic parameters under different blood flow rates into an improved smooth particle fluid dynamics method for simulation to obtain shear force data of the anastomotic region under different blood flow rates; The vascular parameter data is used as the input of the deep learning model to obtain the anastomotic dynamic shear force data.

5. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 4, characterized in that: The improved smoothed particle hydrodynamics method comprises: automatically generating an adaptive particle distribution according to the morphological characteristics of the three-dimensional model of the vascular tree; Increasing the particle density in the anastomotic region to improve simulation accuracy; Adaptive kernel function is used to dynamically adjust the kernel function radius according to the local particle density; An improved virtual particle method is used to process the blood vessel wall boundary to improve the stability of the calculation at the boundary. The elastic deformation response of the blood vessel wall is considered to achieve fluid-structure interaction simulation.

6. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 1, characterized in that: Comparing the anastomotic dynamic shear force data with preset optimal anastomotic angle data to determine the optimal anastomotic angle specifically includes: Construct a deep correlation model between dynamic time-varying shear stress and vascular wall response, including a dual-stream feature extraction network and a multi-scale fusion module; Inputting the anastomotic dynamic shear force data into the depth correlation model to obtain long-term vascular wall response prediction results under different anastomotic angles; A Bayesian optimization framework enhanced by physical information was used to evaluate each candidate anastomotic angle scheme and calculate a multi-objective score that included surgical success rate, fistula maturation time, and long-term patency rate. The optimal anastomotic angle is determined based on the multi-objective score and patient-specific needs.

7. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 6, characterized in that: The Bayesian optimization framework based on physical information enhancement includes: Integrating hemodynamic theory and clinical experience, a theoretical correlation model between anastomotic angle and fistula function was established; Define physical feasibility constraints, including angular range and spatial position restrictions; Construct a multi-objective function and dynamically adjust the weight coefficients of short-term goals, medium-term goals, and long-term goals; Initialize candidate anastomosis angle solutions and use Gaussian process regression model to predict the performance of each solution; Based on the expected improvement and uncertainty trade-off, the optimization is iterated until it converges to the optimal solution.

8. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 1, characterized in that: Based on the three-dimensional arteriovenous fistula model, a virtual surgery is performed to obtain the virtual surgery process data and the postoperative hemodynamic status evaluation results, specifically including: constructing a virtual surgical scene based on the optimal anastomosis angle and the three-dimensional model of the vascular tree; Simulate the surgical process and generate postoperative vascular morphology; Predict the fistula status at different time periods, including one week, one month, and six months later; Calculate the predicted value of fistula maturation time and long-term patency rate; Generate visual surgical guidance plans, including anastomosis angles, position markers, and key operation points.

9. The method for intelligent planning of arteriovenous fistula based on hemodynamic simulation according to claim 1, characterized in that: Also included is an adaptive closed-loop feedback optimization step: Actual surgical parameters and postoperative follow-up data were recorded; Compare the predicted results with the actual results to quantify the accuracy of the prediction; Update model parameters based on the difference between prediction and actual results; Add new case data to the knowledge base and optimize sampling strategies; Perform anomaly detection and analysis on cases of prediction failure to improve system robustness.

10. An intelligent arteriovenous fistula planning system based on hemodynamic simulation, used to implement the intelligent arteriovenous fistula planning method based on hemodynamic simulation according to any one of claims 1 to 9, characterized in that: include: A vascular tree three-dimensional model construction module is used to obtain the three-dimensional spatial anatomical data of the patient's arteries and veins and construct a vascular tree three-dimensional model; A hemodynamic simulation module is used to combine the vascular parameter data of the three-dimensional vascular tree model with the hemodynamic parameter data at different blood flow rates, input them into a deep learning model, and obtain anastomotic dynamic shear force data; An optimal anastomotic angle determination module is used to compare the anastomotic dynamic shear force data with preset optimal anastomotic angle data to determine the optimal anastomotic angle; a three-dimensional modeling module for the arteriovenous fistula procedure, configured to perform three-dimensional modeling of the arteriovenous fistula procedure on the three-dimensional model of the vascular tree in combination with the optimal anastomosis angle to obtain a three-dimensional arteriovenous fistula model; A virtual surgery module, configured to perform virtual surgery based on the three-dimensional arteriovenous fistula model, and obtain virtual surgery process data and postoperative hemodynamic status evaluation results; The adaptive closed-loop feedback module is used to record actual surgical parameters and postoperative follow-up data, compare predicted results with actual results, update model parameters, and improve the system's prediction accuracy.

Images