Digital twinning method for intravascular stent virtual clinical test

Abstract

The invention discloses a digital twinning method for a virtual clinical test of an intravascular stent. The method comprises the following steps: obtaining a preprocessed image after segmentation and preprocessing; performing blood vessel segmentation on the preprocessed image to obtain a blood vessel segmentation image; processing the blood vessel segmentation image to obtain a preprocessed real patient blood vessel model; performing virtual blood vessel sample amplification to generate a virtual blood vessel model; a full-order CFD model is constructed, and hemodynamic parameters are obtained; training the PINN physical neural network to obtain PINN prediction data; and obtaining a preset mechanical condition. Medical images are analyzed and processed through an image processing technology, position detection, segmentation extraction, three-dimensional reconstruction and three-dimensional display of human blood vessels and diseased regions of the human blood vessels are achieved, and hemodynamic changes of specific parts are simulated through different support designs and placement.

Description

Technical Field

[0001] The present application belongs to the field of medical processing technology, and specifically relates to a digital twin method for virtual clinical trials of vascular stents. Background Art

[0002] Cardiovascular disease is close to the first place in the cause of death from human diseases. In recent years, the incidence and mortality of CVD in my country have continued to rise, and it has become the leading cause of death among urban and rural residents in my country. According to statistics, in 2020, CVD accounted for 48.00% and 45.86% of the causes of death among urban and rural residents, respectively; 2 out of every 5 deaths were caused by CVD. It widely invades various types of arteries, and the carotid arteries and cerebral arteries are the main sites of infection, thus causing a variety of serious vascular diseases such as myocardial infarction and cerebral infarction.

[0003] In view of the increasing trend of CVD incidence and mortality, vascular stent interventional treatment is of great significance. At present, vascular stents are divided into cutting stents, braided stents, and degradable stents, and there are many areas for improvement. Vascular stent design and placement can use computer simulation and mathematical modeling technology to conduct virtual clinical trials to generate a large number of virtual patients based on the clinical data of a small number of recruits, so that trial design, data collection and analysis can be carried out faster, shortening the research cycle, and accelerating the development and marketing of new treatments or drugs. In particular, the application of vascular digital twin 3D reconstruction and CFD hemodynamic analysis plays an important role in the work of clinicians and researchers. Compared with traditional clinical trials, it has the advantages of saving time, reducing costs, and reducing clinical trials. Virtual device placement and design can also simulate a variety of different clinical situations and disease progression, helping researchers to more comprehensively evaluate the effectiveness and safety of treatment methods.

[0004] Digital twin (DT) is a simulation process that makes full use of data such as physical models, sensors, and operation history, integrates multi-disciplinary, multi-physical quantities, multi-scales, and multi-probabilities, and completes mapping in virtual space, thereby reflecting the entire life cycle of the corresponding physical equipment. As one of the important applications of digitalization, intelligence, and service, DT breaks through the limitations of time, space, cost, and safety on physical entities, expands and optimizes the relevant functions of physical entities, and enhances the application value of physical entities. Related technologies include augmented reality and virtual reality. By building a digital model of the physical device, engineers can test it multiple times in a virtual environment and optimize and improve it to meet product safety and performance requirements and reduce the time and money spent on physical prototypes. In the clinical trial stage of medical devices, digital twin technology can simulate the performance of the device in the clinical environment and evaluate the safety and performance of the device, thereby reducing clinical trial time, reducing costs, and accelerating the development of medical devices. After the medical device is on the market, digital twin technology can also monitor in real time, collect and analyze the operation data of the device, predict possible failures or performance degradation, and provide data support for equipment maintenance and upgrades. At the same time, DT is also widely used in precision medicine, cancer care, individualized training and other fields. On the one hand, simulation, prediction, and analysis of DT can be used to help patients achieve better treatment outcomes and fewer adverse reactions.

[0005] On the other hand, real clinical data from patients can be used to benchmark, validate, and improve DT modeling. SSM (Statistical Shape Models) are constructed from a set of shapes in a training set, which are usually obtained by collecting the same number of points on all shapes. By calculating the covariance matrix of these points, a statistical model can be derived, including the average shape and each mode (i.e., the main mode of shape change). In the medical field, SSM can help with the three-dimensional reconstruction of anatomical structures, the analysis of anatomical variations, and as part of future digital anatomical atlases, the display of static anatomical structures and their normal or pathological changes. The study at the University of Leeds used a "virtual population" to evaluate the effect of flow diverters in cerebral aneurysms. A virtual population was constructed using real patient data extracted from a clinical database to ensure that anonymous virtual patients were very similar to patients used in clinical trials of real blood flow diversion devices in terms of age, gender, and aneurysm characteristics. After applying the inclusion and exclusion criteria of traditional trials to the computer simulation reference cohort, 82 virtual patients were retained. Through computer models, clinicians and patients can obtain new treatment technologies more safely and promptly. In just a few months, the results of three real clinical trials were replicated, confirming the huge potential of computer virtual human clinical trials. Summary of the invention

[0006] The present application provides a digital twin method for virtual clinical trials of vascular stents to solve the above-mentioned technical problems.

[0007] In order to solve the above technical problems, a technical solution adopted in this application is: a digital twin method for virtual clinical trials of vascular stents, comprising the following steps:

[0008] Based on the real patient vascular CTA image sequence model, obtain the pre-processed image after segmentation and pre-processing;

[0009] Based on the vascular image information and DVT-Unet model, the preprocessed image is modeled and the vascular segmentation is performed automatically to obtain the vascular segmentation image;

[0010] Perform surface processing based on the vascular segmentation image to obtain a pre-processed real patient vascular model;

[0011] Based on the pre-processed real patient vascular model and the statistical shape model, the virtual vascular sample is amplified to generate a virtual vascular model;

[0012] Based on all vascular models including virtual vascular models and real patient vascular models, a full-order CFD model is constructed to design and place virtual plaques and stents, and obtain hemodynamic parameters;

[0013] The hemodynamic parameters are used as training sets to input into the PINN physical neural network for training to obtain PINN prediction data;

[0014] Based on the PINN prediction data and parameter indicators, and based on the hemodynamic information of different parts predicted by the obtained model, the design and positioning of the interventional device are modified and improved to obtain the preset mechanical conditions.

[0015] Furthermore, the method of modeling and fully automatically segmenting blood vessels on the preprocessed image to obtain a blood vessel segmentation image includes:

[0016] Manually annotate the preprocessed images to obtain image data sets;

[0017] Based on the DVT-Unet model, feature learning and blood vessel segmentation are performed to generate simulated blood vessel images;

[0018] Based on the simulated vascular image, the image data set is predicted and segmented, and the weight parameters of the DVT-Unet model are adjusted so that the simulated vascular image reaches the preset performance;

[0019] Based on the simulated blood vessel image that achieves the preset performance, a first image set is obtained.

[0020] Furthermore, a method for amplifying a virtual blood vessel sample and generating a virtual blood vessel model includes:

[0021] Uniformly re-meshing the pre-processed real patient blood vessel model to obtain the original component including the three-dimensional mesh;

[0022] Based on the wrapping module, the template component in the wrapping module is wrapped onto the original component and deformed to obtain the wrapped component;

[0023] Create the same triangulation as the template part and preserve the shape of the deformed part of the cladding part;

[0024] Based on the scaling registration step, the covering component is scaled down so that the covering component is aligned with the original component;

[0025] Based on the statistical shape model, position differences are eliminated to ensure that the shape of the deformed part in the coated component is captured; the instance coefficients in the statistical shape model are updated to obtain the augmented virtual blood vessel model.

[0026] Furthermore, a method for constructing a full-order CFD model, designing and placing virtual plaques and stents, and obtaining hemodynamic parameters includes:

[0027] Based on the virtual vascular model, stents or plaques are fused to the area of ​​interest, boundary conditions are set, and a fluid-solid coupling model is constructed;

[0028] Based on the fluid-solid coupling model, the transient Navier-Stokes equations are solved and the vascular model is simulated by CFD to obtain the vascular data in the form of finite element discretization and the bidirectional fluid-solid coupling CFD model;

[0029] Based on the vascular data, hemodynamic parameters including hemodynamic parameters are obtained by calculating the stress of the plaque area.

[0030] Furthermore, a method for constructing a full-order CFD model, designing and placing virtual plaques and stents, and obtaining hemodynamic parameters includes:

[0031] Based on the virtual vascular model, stents or plaques are fused to the area of ​​interest, boundary conditions are set, and a fluid-solid coupling model is constructed;

[0032] Based on the fluid-solid coupling model, the transient Navier-Stokes equations are solved and the vascular model is simulated by CFD to obtain the vascular data in the form of finite element discretization and the bidirectional fluid-solid coupling CFD model;

[0033] Based on the vascular data, hemodynamic parameters including hemodynamic parameters are obtained by calculating the stress of the plaque area.

[0034] Furthermore, the hemodynamic parameters are input as a training set into the PINN physical neural network for training, and the method for obtaining PINN prediction data includes:

[0035] Based on the bidirectional fluid-structure coupling CFD model, the mechanical distribution column vectors of multiple discrete time nodes are selected to form a sample matrix;

[0036] Construct a PINN physical neural network, where the PINN physical neural network includes an input layer, a hidden layer and an output layer;

[0037] Based on the sample matrix, the PINN physical neural network is trained to obtain the PINN model and PINN prediction data.

[0038] The beneficial effects of the present application are: the present application analyzes and processes medical images through image processing technology, realizes the position detection, segmentation extraction, three-dimensional reconstruction and three-dimensional display of human blood vessels and their lesions, and simulates the hemodynamic changes in specific parts through different stent designs and placements, constructs vascular digital twins based on PINN and SSM, scientifically evaluates virtual test results through qualitative or quantitative analysis, and applies them to virtual vascular clinical trials, which can non-invasively obtain patient vascular parameters similar to those required for real clinical trials, thereby reducing the sample and hemodynamic testing costs required for traditional clinical trials, providing optimization solutions for the planning of vascular stent interventional surgery, and providing important assistance for the research and development of vascular interventional medical devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of an embodiment of a digital twin method for virtual clinical trials of vascular stents of the present application;

[0040] Figure 2 yes Figure 1 A schematic flow chart of an embodiment of step S2 in FIG.

[0041] Figure 3 It is a flowchart of the DVT-Unet model segmentation of blood vessels in the first embodiment of the digital twin method for virtual clinical trials of vascular stents of the present application;

[0042] Figure 4 yes Figure 1 A schematic flow chart of an embodiment of step S4 in FIG.

[0043] Figure 5 yes Figure 1 A schematic flow chart of an embodiment of step S5 in FIG.

[0044] Figure 6 yes Figure 1 A flowchart of an embodiment of step S6 in FIG. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with specific embodiments.

[0046] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0047] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a digital twin method for virtual clinical trials of vascular stents in this application. The method includes the following steps:

[0048] Step S1. Based on a real patient vascular CTA image sequence model, a pre-processed image is obtained after segmentation and pre-processing.

[0049] Specifically, a CT scan is performed on the patient to acquire a high-resolution three-dimensional CTA image sequence model of the patient's blood vessels, and a doctor or a radiologist evaluates the reconstructed blood vessel image to obtain a segmented and pre-processed pre-processed image.

[0050] Step S2. Based on the vascular image information and the DVT-Unet model, the preprocessed image is modeled and the vascular segmentation is performed automatically to obtain a vascular segmentation image.

[0051] For details, see Figure 2 , Figure 2 yes Figure 1 A flow chart of an embodiment of step S2 in FIG. 1 is a flow chart of an embodiment of step S2 in FIG. 1 . Step S2 includes:

[0052] Step S21: Manually annotate the preprocessed image to obtain an image data set.

[0053] Step S22: Perform feature learning and blood vessel segmentation based on the DVT-Unet model to generate a simulated blood vessel image.

[0054] Specifically, the DVT-Unet model is used to perform feature learning on the blood vessel segmentation image, and the blood vessels therein are segmented and extracted to obtain a simulated blood vessel image.

[0055] Step S23: Based on the simulated vascular image, predictive segmentation is performed on the image data set, and the weight parameters of the DVT-Unet model are adjusted so that the simulated vascular image achieves a preset performance.

[0056] Specifically, the model evaluation mode is turned on, the vascular images in the image data set are predictively segmented by simulating vascular images, and the weight parameters of the DVT-Unet model are adjusted so that the simulated vascular images reach the preset performance.

[0057] In this embodiment, the processing process of the DVT-Unet model includes using a labeled blood vessel and plaque dataset in the pre-training stage, enhancing the dataset by random cropping, rotation, scaling, etc. to improve the generalization ability of the model, and using a vision transformer layer for encoding to convert the blood vessel image into a data form that can represent blood vessel characteristics, which may include grayscale values, connectivity, etc.

[0058] like Figure 3 As shown in the figure, the dual ViT encoder uses one path of the Vision Transformer (ViT) to capture global semantic information and establish relationships across pixels through the self-attention mechanism. The other path uses a convolutional neural network (CNN) or a simplified version of ViT to focus on learning local features of the image. The two paths use a feature fusion layer to model the dependency between global semantics and local features. The decoder part in DVT-Unet is similar to the traditional U-Net structure, which is used to gradually restore the spatial resolution of the image and fuse the multi-level information in the encoder.

[0059] During the decoding process, the feature maps in the encoder are fused with the feature maps of the corresponding layers in the decoder to retain more detailed information. During the training process, the DVT-Unet model is trained using the preprocessed and enhanced dataset and the training is simplified by reducing the token size and attention range, which may involve model pruning or optimization of the attention mechanism. The accuracy and robustness of the model are evaluated using an independent test set. Manual verification is performed to ensure the accuracy and medical acceptability of the segmentation results. Finally, the trained DVT-Unet model is obtained and applied to actual medical image analysis tasks, such as vascular segmentation, plaque detection, etc.

[0060] In the DVT-Unet model, we innovatively introduced a dual Vision Transformer (dual ViT) encoder architecture to significantly improve the accuracy and efficiency of vessel segmentation and plaque detection. This unique design fully utilizes ViT's advantages in global semantic understanding and local feature capture through two parallel processing paths.

[0061] The first path in the dual ViT encoder uses a standard Vision Transformer structure, which can capture and integrate global semantic information in the image across pixels with its powerful self-attention mechanism. This capability enables the model to deeply understand the overall structure of the vascular network and the location and characteristics of plaques in the global context. The second path focuses on the local details of the image, which may be combined with the efficient local feature extraction capabilities of convolutional neural networks (CNNs), or use lightweight ViT variants to reduce computational burden. This path ensures that the model can accurately identify the edges of blood vessels, bifurcations, and subtle texture features of plaques.

[0062] The two paths interact with each other in the encoding stage through a specific fusion mechanism (such as feature map addition, concatenation or attention weighting), realizing the organic combination of global semantics and local features. This fusion not only enhances the representation ability of the model, but also promotes the complementarity and enhancement of information.

[0063] In order to simplify the training process and improve efficiency, this application also optimizes the token size and attention range of the dual ViT encoder. By reducing unnecessary computation, the model can achieve faster training and inference speed while maintaining high performance. In the training process, we use finely labeled blood vessel and plaque datasets to enrich data diversity through random cropping, enhancement and other strategies. After training, we use an independent test set to evaluate the model and combine manual verification to ensure the accuracy and medical reliability of the segmentation results. Experimental results show that the DVT-Unet model, with its dual ViT encoder architecture, demonstrates excellent performance and robustness in vascular segmentation and plaque detection tasks.

[0064] Step S24: Based on the simulated blood vessel image that achieves preset performance, a first image set is obtained.

[0065] Specifically, simulated blood vessel images that achieve preset performance are collected to obtain a first image set.

[0066] Step S3: Perform surface processing based on the blood vessel segmentation image to obtain a pre-processed real patient blood vessel model.

[0067] Specifically, a 3D-slicer is used to process, smooth and mesh the vascular model to obtain a pre-processed real patient vascular model.

[0068] Step S4: Based on the pre-processed real patient blood vessel model and the statistical shape model, the virtual blood vessel sample is amplified to generate a virtual blood vessel model.

[0069] Based on the processed real patient vascular model and the statistical shape model SSM (Statistic Shape Model), virtual vascular samples are amplified to generate a large number of virtual vascular models with similar statistical characteristics to the original samples.

[0070] For details, see Figure 4 , Figure 4 yes Figure 1 A flow chart of an embodiment of step S4 in FIG. 4 includes:

[0071] Step S41: uniformly re-grid the pre-processed real patient blood vessel model to obtain an original component including a three-dimensional grid.

[0072] Specifically, the surface mesh is optimized based on the uniform remesh uniform automatic mesh reconstruction operation, the quality and size of the surface triangles of the original parts are controlled, and the original parts are meshed as a whole to create Tet4 and Tet10 three-dimensional meshes in preparation for subsequent finite element analysis. The original parts are the various parts of the vascular model in the real patient vascular model.

[0073] Step S42: Based on the covering module, the template component in the covering module is covered on the original component and deformed to obtain the covering component.

[0074] Specifically, based on the Warp Part module, the template part is wrapped onto the original part to deform it. Create a triangulation with the same shape as the template part, but retain the shape of the wrapped part, ensure point correspondence, and repeat this step for all original parts.

[0075] Step S43: Create a triangulation that is the same as the template component and retains the shape of the deformed portion of the covering component.

[0076] Specifically, a triangulation is created that has the same triangulation as the template part, but preserves the shape of the deformed part in the wrapping part, ensuring that the points of each corresponding part correspond, and repeat this step for all original parts.

[0077] Step S44: Based on the scaling registration step, the covering component is reduced so that the covering component is aligned with the original component.

[0078] Specifically, based on the Registration with Scaling step, the wrapped component and the original component are aligned, and this step is repeated for all the original components to align the original components.

[0079] Step S45. Based on a statistical shape model (SSM), position differences are eliminated to ensure that the shape of the deformed portion of the covering component is captured; instance coefficients in the statistical shape model are updated to obtain an augmented virtual blood vessel model.

[0080] Specifically, based on the statistical shape model SSM (Create SSM), the Procrustes alignment method is used to eliminate position differences to ensure that only shape deformations are captured. The Convert Instance to Part operation recreates the part with the required instance coefficient combination. Before converting the SSM instance to a part, the instance coefficients in the statistical shape model are updated through the operation to obtain the augmented vascular model.

[0081] Step S5. Based on all vascular models including virtual vascular models and real patient vascular models, a full-order CFD model is constructed to perform virtual plaque and stent design (shape, size, memory alloy hardness) and placement, and obtain hemodynamic parameters.

[0082] For details, see Figure 5 , Figure 5 yes Figure 1 A flow chart of an embodiment of step S5 in FIG. 1 is shown in FIG. 1 . Step S5 includes:

[0083] Step S51: Based on the virtual blood vessel model, stents or plaques are fused to the site of interest, boundary conditions are set, and a fluid-solid coupling model is constructed.

[0084] Specifically, take the carotid artery as an example: Assume that the carotid artery wall is a linear elastic material with an elastic modulus of 1.5 MPa, a Poisson's ratio of 0.45, and a density of 1.055 kg / m3. Refer to the flow distribution ratio of the inner and outer carotid arteries in the TF-AHCB model of the carotid bifurcation tube, the wall shear stress of the reflow zone and the diversion zone, and other values. Adjust the parameters based on the normal blood vessel blood flow parameters, change the device structure, and try again and again until there is a feasible device structure that makes the virtual clinical trial feasible, so as to achieve the purpose of promoting the development of clinical trials and build a fluid-solid coupling model.

[0085] Step S52: Based on the fluid-solid coupling model, solve the transient Navier-Stokes equations and perform CFD simulation on the blood vessel model to obtain blood vessel data in the form of finite element discretization and a bidirectional fluid-solid coupling CFD model.

[0086] Step S53: Based on the blood vessel data, hemodynamic parameters are obtained by calculating the stress of the plaque area.

[0087] Specifically, based on the blood flow data, hemodynamic parameters including pressure field, velocity field and wall shear stress (WSS) in the fluid are obtained by calculating the stress in the plaque area.

[0088] Step S6. Input the hemodynamic parameters as a training set into the PINN physical neural network for training to obtain PINN prediction data.

[0089] For details, see Figure 6 , Figure 6 yes Figure 1 Schematic diagram of a flow chart of an embodiment of step S6 in FIG. Step S6 includes:

[0090] Step S61. Based on the bidirectional fluid-solid coupling CFD model, select the mechanical distribution column vectors of multiple discrete time nodes to form a sample matrix.

[0091] Step S62: construct a PINN physical neural network, wherein the PINN physical neural network includes an input layer, a hidden layer and an output layer.

[0092] Specifically, a PINN physical neural network including an input layer, a hidden layer and an output layer is constructed.

[0093] Step S63: Based on the sample matrix, the PINN physical neural network is trained to obtain the PINN model and PINN prediction data.

[0094] Step S7. Based on the PINN prediction data and parameter indicators, and based on the hemodynamic information of different parts predicted by the obtained model, modify and improve the design and positioning of the interventional device to obtain the preset mechanical conditions.

[0095] Specifically, based on the PINN prediction data, referring to the parameter indicators that may cause damage to blood vessels and interventional devices, and based on the hemodynamic information of different parts predicted by the obtained PINN model, the design and positioning of interventional devices are modified and improved to obtain clinically satisfactory mechanical conditions.

[0096] The digital twin method for virtual clinical trials of vascular stents in the present application constructs a vascular digital twin based on PINN and SSM, and applies it to virtual vascular clinical trials. It can non-invasively obtain patient vascular parameters similar to those required for real clinical trials, thereby reducing the cost of samples and hemodynamic testing required for traditional clinical trials, and providing important assistance for the research and development of vascular interventional medical devices.

[0097] The above descriptions are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A digital twin method for virtual clinical trials of vascular stents, characterized in that: The following steps are involved: Based on the real patient vascular CTA image sequence model, obtain the pre-processed image after segmentation and pre-processing; Based on the vascular image information and the DVT-Unet model, modeling and fully automatic vascular segmentation are performed on the preprocessed image to obtain a vascular segmentation image; Performing surface processing based on the blood vessel segmentation image to obtain a preprocessed real patient blood vessel model; Based on the preprocessed real patient blood vessel model and the statistical shape model, amplifying the virtual blood vessel sample to generate a virtual blood vessel model; Based on all vascular models including the virtual vascular model and the real patient vascular model, a full-order CFD model is constructed to design and place virtual plaques and stents, and obtain hemodynamic parameters; Inputting the hemodynamic parameters as a training set into the PINN physical neural network for training to obtain PINN prediction data; Based on the PINN prediction data and parameter indicators, and based on the hemodynamic information of different parts predicted by the obtained model, the design and positioning of the interventional device are modified and improved to obtain the preset mechanical conditions.

2. The method according to claim 1, characterized in that The method of modeling and fully automatically segmenting blood vessels on the preprocessed image to obtain a blood vessel segmentation image comprises: Manually annotate the preprocessed image to obtain an image data set; Based on the DVT-Unet model, feature learning and blood vessel segmentation are performed to generate simulated blood vessel images; Based on the simulated blood vessel image, predictively segment the image data set, and adjust the weight parameters of the DVT-Unet model so that the simulated blood vessel image achieves a preset performance; Based on the simulated blood vessel image that achieves preset performance, a first image set is obtained.

3. The method according to claim 2, characterized in that The method for amplifying a virtual blood vessel sample and generating a virtual blood vessel model comprises: uniformly re-gridding the pre-processed real patient blood vessel model to obtain an original component including a three-dimensional grid; Based on the covering module, the template component in the covering module is covered on the original component and deformed to obtain the covering component; Creating a triangulation that is identical to the template component and retains the shape of the deformed portion of the cladding component; Based on the scaling registration step, the covering component is reduced so that the covering component is aligned with the original component; Based on the statistical shape model, position differences are eliminated to ensure that the shape of the deformed portion of the covering component is captured; and instance coefficients in the statistical shape model are updated to obtain an augmented virtual blood vessel model.

4. The method according to claim 3, characterized in that The method of constructing a full-order CFD model, designing and placing virtual plaques and stents, and obtaining hemodynamic parameters includes: Based on the virtual blood vessel model, stent or plaque fusion is performed on the part of interest, boundary conditions are set, and a fluid-solid coupling model is constructed; Based on the fluid-solid coupling model, solving the transient Navier-Stokes equations and performing CFD simulation on the blood vessel model to obtain blood vessel data in the form of finite element discretization and the bidirectional fluid-solid coupling CFD model; Based on the blood vessel data, hemodynamic parameters are obtained by calculating the stress of the plaque area.

5. The method according to claim 4, characterized in that The method of inputting the hemodynamic parameters as a training set into a PINN physical neural network for training to obtain PINN prediction data comprises: Based on the bidirectional fluid-structure coupling CFD model, the mechanical distribution column vectors of multiple discrete time nodes are selected to form a sample matrix; Constructing a PINN physical neural network, wherein the PINN physical neural network includes an input layer, a hidden layer and an output layer; Based on the sample matrix, the PINN physical neural network is trained to obtain the PINN model and the PINN prediction data.

Images