A system for aortic arch pathology patient-specific hemodynamic analysis assessment

Abstract

The application provides an aortic arch disease patient-specific hemodynamic analysis and evaluation system, comprising: a patient data acquisition module for acquiring 4DFlow MRI original data or static CT / MRI data of a patient; a first boundary condition generation subsystem for constructing a patient-specific aortic inlet super-resolution velocity field data set from the 4DFlow MRI original data, and converting to obtain a CFD simulation inlet boundary condition; a second boundary condition generation subsystem for generating a CFD simulation inlet boundary condition according to the static CT / MRI data of the patient in combination with large-scale population statistical rules; and a blood flow dynamic analysis subsystem for preoperative virtual planning, parameterized modeling and postoperative evaluation according to the CFD simulation inlet boundary condition to obtain an evaluation result. The application can realize preoperative simulation and postoperative evaluation of a covered stent micro-hole fenestration procedure, and fully utilize existing clinical resources.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a hemodynamic analysis and evaluation system specific to patients with aortic arch lesions. Background Technology

[0002] In endovascular repair of aortic arch lesions, the microporous fenestration technique using a covered stent involves creating an opening area and arranging a micropore array on the stent cover. This allows blood flow from the stent lumen through the micropores to the branch arteries or the lesion cavity, ensuring blood flow isolation in the main vessel while maintaining laminar perfusion in the branch arteries, reducing branch artery pressure, and promoting dissection healing. An existing patent (authorization announcement number CN223614976U) discloses a covered stent with an opening area containing multiple micropores (diameter 1.5-1.8 mm, density 25-36 micropores per 2 square centimeters, circumferential angle range of 45°-360°, axial length not exceeding 12 cm). By rationally arranging the diameter and density of the micropores, laminar blood flow in the branch arteries is ensured, avoiding turbulence that increases the risk of thrombosis, arteriosclerosis, arterial stenosis, and cerebral infarction. Simultaneously, it reduces branch artery blood pressure and ensures branch artery blood flow and perfusion.

[0003] However, existing technologies have the following systemic shortcomings regarding preoperative planning and postoperative hemodynamic assessment for this type of procedure: Firstly, the quality of patient-specific raw data is insufficient: 4D Flow MRI, as a non-invasive method for acquiring the three-dimensional blood flow velocity field of the aorta, is limited by scanning time and signal-to-noise ratio, resulting in a spatial resolution of only 2-4 mm and limited temporal resolution. Phase wrapping artifacts are prone to occur in high-velocity regions. Traditional interpolation cannot recover high-frequency blood flow details, and general image super-resolution networks do not fully consider the hydrodynamic characteristics of 4D Flow MRI data and the coupling relationship between multi-channel velocity components, leading to insufficient accuracy of directly extracted patient-specific inlet velocity field data, making it difficult to directly drive high-fidelity CFD (Computational Fluid Dynamics) simulations.

[0004] Secondly, the boundary conditions lack individualized generation capabilities: Existing CFD simulations mostly use empirical formulas, statistical averages from literature, or idealized pulsating inlet conditions as boundary conditions, failing to reflect the correlation between individual patient anatomy and blood flow function, leading to significant deviations between simulation results and the patient's actual physiological state. Furthermore, routine clinical CTA / MRA images do not contain functional information such as flow velocity, lacking the technical means to generate dynamic, personalized boundary conditions from static anatomical images. In particular, existing statistical shape models (SSM) technology is mainly applied to the statistical analysis of aortic geometry and has not been fully extended to the generation of hemodynamic boundary conditions. It lacks key technical aspects such as SSM construction for time-varying three-dimensional velocity profiles at the ascending aortic inlet, physiological rationality screening of synthetic profiles, and personalized mapping based on the patient's static anatomical morphology.

[0005] Third, there is a lack of a dedicated assessment toolchain for the procedure: existing stent selection and blood flow analysis software can only simulate traditional solid covered stents or large fenestrated stents, and cannot perform parameterized configuration and preoperative virtual planning for the location of the opening area (circumferential angle range, axial length), micropore diameter (1.5-1.8mm), and density (25-36 per 2cm2) of the aforementioned microporous covered stents; at the same time, existing CFD simulations either treat the covered stent as a completely closed wall (ignoring the micropore array) or simplify the opening area as a large-scale equivalent opening, which cannot analyze the local flow details at the micropore array level, and cannot quantitatively assess the Reynolds number, laminar flow maintenance rate, and transmembrane pressure difference in the micropore outlet area; existing postoperative assessment indicators (such as wall shear stress and eddy current volume) are mostly used for traditional stent assessment, and no specific functional assessment parameter set for microporous fenestrated procedures has been established, such as branch artery laminar flow maintenance index, micropore turbulence risk coefficient, false lumen / aneurysm pressure reduction rate, and branch perfusion guarantee rate. Summary of the Invention

[0006] Therefore, it is necessary to provide a hemodynamic analysis and evaluation system specific to patients with aortic arch lesions to address the aforementioned technical problems.

[0007] A patient-specific hemodynamic analysis and evaluation system for aortic arch lesions includes: a patient data acquisition module, a first boundary condition generation subsystem, a second boundary condition generation subsystem, and a hemodynamic analysis subsystem. Both the first and second boundary condition generation subsystems are connected to the patient data acquisition module and the hemodynamic analysis subsystem, respectively. The patient data acquisition module acquires raw 4DFlow MRI data or static CT / MRI data of the patient. The first boundary condition generation subsystem constructs a super-resolution velocity field dataset of the patient-specific aortic inlet based on the patient's raw 4DFlow MRI data and converts it to obtain CFD simulation inlet boundary conditions. The second boundary condition generation subsystem generates CFD simulation inlet boundary conditions based on the patient's static CT / MRI data and combines large-scale population statistical patterns. The hemodynamic analysis subsystem performs preoperative virtual planning, parametric modeling, and postoperative evaluation of the covered stent based on the CFD simulation inlet boundary conditions to obtain evaluation results.

[0008] In one embodiment, the first boundary condition generation subsystem includes: a standard input data generation module for preprocessing the raw 4DFlow MRI data and extracting the ascending aortic inlet region, and normalizing it to obtain standard input data; a synthetic training data generation module for generating synthetic training data based on CFD simulation physical perception; a super-resolution network construction module for constructing a deep residual super-resolution network and inputting the standard input data to obtain a super-resolution velocity field dataset; a synthetic training data adjustment module for pre-training and fine-tuning the synthetic training data with clinical data; a boundary condition conversion module for converting the patient-specific ascending aortic inlet super-resolution velocity field dataset into CFD simulation inlet boundary conditions; and a reconstruction quality assessment module for assessing reconstruction quality and setting reconstruction pass conditions, and transmitting the CFD simulation inlet boundary conditions to the hemodynamic learning module when the reconstruction pass conditions are met.

[0009] In one embodiment, the standard input data generation module includes: a phase unwrapping unit, used to unwrap the original 4DFlow MRI data using a spatial continuity constraint-based unwrapping algorithm or a deep learning unwrapping algorithm to obtain the true velocity encoding value; a phase image conversion unit, used to convert the phase image into physical velocity values ​​according to the velocity encoding value to generate a three-dimensional velocity component map; an entrance region extraction unit, used to extract the three-dimensional geometric mask of the ascending aorta based on the amplitude image or phase-contrast magnetic resonance angiography image using threshold segmentation, random walk segmentation, or a deep learning segmentation network, locate the entrance interface of the ascending aorta according to anatomical landmarks, and extract the volume of interest in the entrance region, wherein the volume of interest covers at least 1.5 times the pipe diameter upstream and at least 2 times the pipe diameter downstream of the entrance section; and a normalization processing unit, used to normalize the velocity components and amplitude image in the entrance region to generate standardized input data.

[0010] In one embodiment, the deep residual super-resolution network includes: an input layer, a low-resolution feature encoder, an upsampling layer, a high-resolution feature encoder, and an output layer. The input layer adopts a dual-path parallel input structure, including an anatomical path and a velocity path. The anatomical path inputs PC-MRA, amplitude images, and their product image to provide vascular anatomical information. The velocity path inputs a low-resolution three-dimensional velocity component map, carrying hemodynamic information to be reconstructed. The low-resolution feature encoder is composed of multiple stacked residual blocks, set in a low-resolution space. Each residual block contains two 3×3×3 three-dimensional convolutional layers and skip connections. The upsampling layer is located in the middle of the network and uses sub-pixel convolution or transposed convolution to improve spatial resolution. The high-resolution feature decoder is composed of multiple stacked residual blocks, set in a high-resolution space, for refined reconstruction. The output layer outputs a three-dimensional high-resolution velocity component map and has a global skip connection in the input layer to add the result of upsampling the input low-resolution velocity field after trilinear interpolation to the network residual output to obtain a super-resolution velocity field dataset.

[0011] In one embodiment, the boundary condition transformation module includes: a patching inference unit, used to input preprocessed patient 4D Flow MRI entry region data into a trained depth residual super-resolution network and perform patching inference; a divergence correction unit, used to perform divergence correction on the super-resolution velocity field dataset using a projection method or a divergence-free radial basis function method, set the velocity values ​​in non-fluid regions to zero based on a PC-MRA mask, and denormalize them to physical velocity units; and a boundary condition transformation unit, used to map the corrected super-resolution velocity field dataset to CFD entry section grid nodes using trilinear or higher-order interpolation, map the time series to the CFD simulation time step, obtain the CFD simulation entry boundary conditions, and calculate the spatial average velocity, peak velocity, flow waveform, and Reynolds number time-varying curve for comparison and verification with clinical reference values, and output the results in a preset standard format if the verification is successful.

[0012] In one embodiment, the second boundary condition generation subsystem includes: an ascending aorta geometry model reconstruction module, used to acquire the patient's static CT / MRI data and reconstruct an ascending aorta geometry model based on the static CT / MRI data; an anatomical landmark alignment module, used to perform patient-specific inlet and outlet plane localization and anatomical landmark alignment based on the reconstructed ascending aorta geometry model; a velocity profile extraction module, used to construct a large-scale clinical cohort 4D FlowMRI database, extract 4D FlowMRI velocity profiles from the clinical cohort, and perform spatiotemporal alignment; and a statistical model construction module, used to construct a 4D model based on principal component analysis based on the spatiotemporally aligned 4D FlowMRI velocity profiles. The system includes: a velocity profile statistical shape model; a velocity profile set construction module for generating synthetic velocity profiles and selecting acceptance criteria to construct a synthetic 4D velocity profile dataset that meets a preset size; a velocity profile generation module for generating patient-specific velocity profiles and associating them with anatomical morphology based on the synthetic 4D velocity profile dataset; an entry boundary condition generation module for mapping patient-specific boundary condition space and adapting to the CFD interface to generate CFD simulation entry boundary conditions; and an uncertainty quantification module for quantifying the uncertainty of the CFD simulation entry boundary conditions and performing sensitivity analysis.

[0013] In one embodiment, the anatomical landmark alignment module includes: an inlet positioning unit for locating the inlet and inlet plane required for CFD simulation using an anatomical landmark-driven method based on the reconstructed ascending aorta geometry model; an outlet plane setting unit for setting the outlet plane at the distal end of the descending aorta or at aortic arch branch vessels according to the CFD simulation domain definition; a three-dimensional coordinate recording unit for triangulating and discretizing the inlet and outlet planes into a triangulated mesh and recording the three-dimensional coordinates of each node in the patient coordinate system; and a rigid transformation unit for performing a rigid transformation on the inlet plane to ensure that the inlet plane meets a set standard spatial orientation.

[0014] In one embodiment, the inlet boundary condition generation module includes: an inverse rigid transformation unit, used to perform the inverse operation of the rigid transformation on the patient-specific velocity profile, restoring the velocity vector and node coordinates from the standard coordinate system to the patient's original coordinate system; a velocity vector mapping unit, used to map the fixed node velocity vectors on the unit disk to the original triangulated mesh nodes of the patient inlet plane through inverse radial basis function interpolation, keeping the extravascular node velocities zero for non-circular sections; and a velocity profile mapping unit, used to map the standard cardiac cycle t∈[0, The velocity profile of the set frame within [1] is mapped to the physical time axis according to the patient's actual heart rate; the velocity field dataset output unit is used to output the time-varying three-dimensional velocity field dataset of the inlet plane in a set boundary condition format to obtain the CFD simulation inlet boundary conditions; the outlet boundary condition generation unit, based on the inlet velocity profile and the principle of mass conservation, combined with the patient-specific aortic geometry model, derives the pressure-flow relationship at the distal outlet of the descending aorta through a one-dimensional hemodynamic model or empirical proportional relationship, and generates the outlet boundary conditions using the Windkessel model or impedance boundary condition model; the inlet flow waveform verification unit is used to verify the consistency between the inlet flow waveform and the patient's clinical reference value. If the deviation exceeds the preset threshold, a feedback correction mechanism is triggered to adjust the modal weights and regenerate the profile.

[0015] In one embodiment, the hemodynamic analysis subsystem includes: a vessel wall geometry model construction module, used to acquire preoperative or postoperative imaging data of the patient and construct a vessel wall geometry model; a virtual surgical plan generation module, used to generate a virtual surgical plan for a covered stent micropore fenestration procedure based on the vessel wall geometry model; a coupling geometry module generation module, used to establish a parametric geometric model including the stent body skeleton, the covering layer, and the micropore array in the fenestration area according to the virtual surgical plan, and to perform registration and Boolean fusion based on anatomical landmarks and the vessel wall geometry model to generate a stent-vessel-micropore coupled geometric model; and a transmembrane flow coupling module, used to combine the inlet boundary conditions of the CFD simulation with the patient's systolic blood pressure... A hemodynamic control equation based on microporous transmembrane flow is established using either a Windkessel three-element model calibrated to diastolic pressure or lumped parameter outflow boundary conditions, resulting in transmembrane flow boundary conditions. A mesh generation module is used to perform unstructured tetrahedral and hexahedral hybrid mesh generation on the stent-vessel-micropore coupled geometric model, and adaptive mesh refinement is applied. A surgical outcome evaluation module is used to extract postoperative hemodynamic parameters and evaluate the surgical outcome, obtaining evaluation results. These hemodynamic parameters include at least one of the following: micropore outlet Reynolds number, branch artery laminar flow maintenance index, transmembrane pressure difference, micropore transmural flow rate, branch artery perfusion guarantee rate, branch artery pressure reduction rate, and false lumen / aneurysm pressure reduction rate.

[0016] In one embodiment, the transmembrane flow coupling module includes: a kinetic control equation establishment unit, used to obtain the inlet boundary conditions of the CFD simulation, and combine the Windkessel three-element model or lumped parameter outflow boundary conditions based on the patient's systolic / diastolic blood pressure calibration to establish a hemodynamic control equation considering microporous transmembrane flow. The control equation introduces transmembrane flow boundary conditions or equivalent porous media source terms in the micropore region. The transmembrane flow boundary conditions calculate the local flow resistance based on the micropore diameter and micropore density, so that when blood flows from the stent lumen through the micropores to the outside, it satisfies the laminar flow constraint; and a micropore array equivalent unit, used to use an equivalent porous wall model to represent the micropore array as a thin film boundary with specific permeability and inertial resistance when the micropore scale is smaller than a set multiple of the grid scale.

[0017] Compared to existing technologies, the advantages and beneficial effects of this invention are as follows: A patient data acquisition module collects raw 4DFlow MRI data or static CT / MRI data from the patient; a first boundary condition generation subsystem constructs a super-resolution velocity field dataset of the patient-specific aorta based on the patient's raw 4DFlow MRI data; and converts this dataset to obtain the CFD simulation entry boundary conditions, thereby transforming the low-resolution 4DFlow MRI data into a super-resolution velocity field dataset of the aorta. The raw MRI data is upscaled to a usable high-resolution patient-specific velocity field, overcoming traditional hardware limitations and fully utilizing existing clinical data resources. A second boundary condition generation subsystem generates CFD simulation entry boundary conditions based on the patient's static CT / MRI data and large-scale population statistical patterns. This allows for the generation of dynamic, personalized boundary conditions from conventional CT / MRI images, significantly improving the patient specificity and clinical feasibility of CFD simulation. The two boundary condition generation subsystems complement each other, covering different data availability scenarios. A hemodynamic analysis subsystem is used to perform preoperative virtual planning, parametric modeling, and postoperative evaluation of covered stents based on the CFD simulation entry boundary conditions obtained from either boundary condition generation subsystem. The evaluation results enable preoperative simulation and postoperative evaluation of covered stent micropore fenestration procedures, optimizing preoperative stent selection and micropore layout, and providing hemodynamic evidence for procedure safety assessment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a patient-specific hemodynamic analysis and assessment system for aortic arch lesions in one embodiment.

[0019] Figure 2 This is a schematic diagram of the structure of the first boundary condition generation subsystem in one embodiment.

[0020] Figure 3 This is a schematic diagram of the structure of the second boundary condition generation subsystem in one embodiment.

[0021] Figure 4 This is a schematic diagram of the hemodynamic analysis subsystem in one embodiment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0023] like Figure 1As shown, a patient-specific hemodynamic analysis and evaluation system for aortic arch lesions is provided, comprising: a patient data acquisition module 10, a first boundary condition generation subsystem 20, a second boundary condition generation subsystem 30, and a hemodynamic analysis subsystem 40. Both the first and second boundary condition generation subsystems are connected to the patient data acquisition module 10 and the hemodynamic analysis subsystem 40. The patient data acquisition module 10 is used to acquire the patient's raw 4DFlow MRI data or static CT (Computed Tomography) / MRI (Magnetic Resonance Imaging) data. The first boundary condition generation subsystem 20 is used to construct a patient-specific aortic inlet blood flow velocity field dataset based on the patient's raw 4DFlow MRI data and convert it to obtain CFD simulation inlet boundary conditions. The second boundary condition generation subsystem 30 is used to generate CFD simulation inlet boundary conditions based on the patient's static CT / MRI data and combined with large-scale population statistical patterns. The hemodynamic analysis subsystem 40 is used to perform preoperative virtual planning, parametric modeling, and postoperative evaluation of the covered stent based on the CFD simulation inlet boundary conditions to obtain evaluation results.

[0024] In this embodiment, the patient data acquisition module 10 collects the patient's 4DFlow MRI raw data or static CT / MRI data. The first boundary condition generation subsystem 20 constructs a patient-specific aortic super-resolution velocity field dataset based on the patient's 4DFlow MRI raw data and converts it to obtain the CFD simulation inlet boundary conditions, thus transforming the low-resolution 4DFlow... The raw MRI data is upscaled to a usable high-resolution patient-specific velocity field, overcoming traditional hardware limitations and fully utilizing existing clinical data resources. The second boundary condition generation subsystem 30 generates CFD simulation entry boundary conditions based on the patient's static CT / MRI data and large-scale population statistical patterns. This allows for the generation of dynamic, personalized boundary conditions from conventional CT / MRI images, significantly improving the patient specificity and clinical feasibility of CFD simulation. The two boundary condition generation subsystems complement each other, covering different data availability scenarios. The hemodynamic analysis subsystem 40 uses the CFD simulation entry boundary conditions obtained from either boundary condition generation subsystem to perform preoperative virtual planning, parametric modeling, and postoperative evaluation of covered stents. This yields evaluation results, enabling preoperative simulation and postoperative evaluation of covered stent micropore fenestration procedures. It optimizes preoperative stent selection and micropore layout, and provides hemodynamic evidence for procedure safety assessment.

[0025] Specifically, the first boundary condition generation subsystem 20 and the second boundary condition generation subsystem 30 complement each other to cover different data availability scenarios. Both provide entry boundary conditions to the hemodynamic analysis subsystem 40, forming a closed-loop technology chain of data augmentation, boundary condition generation and special surgical procedure evaluation, providing key technical support for precision medicine.

[0026] like Figure 2 As shown, the first boundary condition generation subsystem 20 includes: a standard input generation and acquisition module 21, used to preprocess the raw 4DFlow MRI data and extract the ascending aortic inlet region, and normalize it to obtain standard input data; a synthetic training data generation module 22, used to generate synthetic training data based on CFD simulation physical perception; a super-resolution network construction module 23, used to construct a deep residual super-resolution network and input standard input data to obtain a super-resolution velocity field dataset; a synthetic training data adjustment module 24, used to pre-train the synthetic training data and fine-tune it with clinical data; a boundary condition conversion module 25, used to convert the patient-specific ascending aortic inlet super-resolution velocity field dataset into CFD simulation inlet boundary conditions; and a reconstruction quality assessment module 26, used to assess reconstruction quality and set reconstruction pass conditions. When the reconstruction pass conditions are met, the CFD simulation inlet boundary conditions are transmitted to the hemodynamic learning module.

[0027] Specifically, the first boundary condition generation subsystem 20 is used to reconstruct a high-resolution, high-fidelity patient-specific ascending aortic inlet blood flow velocity field dataset from low-resolution 4D Flow MRI raw data, providing a patient-specific inlet boundary condition data basis for subsequent CFD simulations.

[0028] Specifically, the synthetic training data generation module 22 is used to construct a patient-specific CFD simulation database to generate paired low-resolution / high-resolution synthetic 4D Flow MRI training data. A three-dimensional geometric model of the ascending aortic inlet region is reconstructed based on the patient's medical images (CT or high-resolution MRI), and surface smoothing and mesh generation are performed. Physiologically realistic inlet boundary conditions are set, and transient blood flow simulation is performed using a Navier-Stokes equation solver (such as OpenFOAM, SimVascular, or ANSYS Fluent) to obtain a high-resolution three-dimensional velocity field vCFD(x, y, z, t). The spatial resolution is preferably 0.5-1.0 mm, and the temporal resolution covers the entire cardiac cycle (no less than 20 phases). The high-resolution CFD velocity field is physically sensed downsampled: the CFD velocity field is mapped to a low-resolution Cartesian mesh, a complex signal is constructed (with the velocity component as the phase, simulating the amplitude image), a fast Fourier transform is performed to k-space, high-frequency components are truncated and complex Gaussian noise is added, and then a low-resolution synthetic 4D Flow MRI data is obtained through inverse fast Fourier transform. Optionally, some volume effect simulation is introduced. Rigid transformation enhancement and elastic deformation enhancement are performed on paired low-resolution / high-resolution data.

[0029] The synthetic training data adjustment module 24 employs a multi-stage training strategy, including pre-training and fine-tuning. The first stage involves pre-training with CFD synthetic data, dividing paired low-resolution / high-resolution synthetic data into training, validation, and test sets (7:1:2), using a voxel-based mean squared error (MSE) loss function as the base loss; preferably, a physical constraint loss term, divergence constraint loss, is introduced. (Encouraging compliance with mass conservation) and loss of velocity amplitude consistency The total loss function is: L total = λ 1 ・L MSE + λ 2 ・ L div + λ 3 ・L mag (1) In the formula, L MSE For voxel-level MSE, preferred λ 1 = 1 . 0, λ 2 = 0 . 1, λ 3 = 0 . 5. The Adam optimizer is used with an initial learning rate of 1×10⁻⁶. −4The learning rate is decayed every 14-20 epochs at a rate of 0.5-0.8, with a batch size of 16-32, and training lasts for 200-300 epochs. A patch training strategy is used, selecting only patches that account for at least 20% of the fluid region for training.

[0030] Phase Two: Clinical Data Fine-tuning. A small amount of paired real-world low-resolution 4D flow MRI and high-resolution reference data were collected, and the pre-trained weights from Phase One were loaded with a low learning rate (1×10⁻⁶). −5 Up to 5 × 10 −5 Fine-tuning. Preferably, a distributed learning strategy is introduced, in which random Gaussian noise (pre-added noise model) is added to the input data during the fine-tuning stage, and a distributed loss function based on energy statistics is used to replace the traditional MSE. Fine-tuning is performed for 50-100 epochs and an early stopping mechanism is adopted.

[0031] The reconstruction quality assessment module 26 is used for reconstruction quality assessment and uncertainty quantification, including: calculating the root mean square error, structural similarity index, and relative velocity error between the reconstructed velocity field and the reference high-resolution velocity field; calculating the divergence-free error, vorticity field distribution, and wall shear stress distribution of the reconstructed velocity field and comparing them with the reference; quantifying reconstruction uncertainty using the Monte Carlo dropout method or ensemble learning strategy; and setting assessment thresholds (e.g., RMSE < 5% peak velocity, SSIM > 0.85), and only when the reconstruction results meet the thresholds can the subsequent CFD simulation process proceed.

[0032] The standard input data generation module 21 includes: a phase unwrapping unit, used to unwrap the original 4DFlow MRI data using a spatial continuity constraint-based unwrapping algorithm or a deep learning unwrapping algorithm to obtain the true velocity encoding value; a phase image conversion unit, used to convert the phase image into physical velocity values ​​according to the velocity encoding value to generate a three-dimensional velocity component map; an entrance region extraction unit, used to extract the three-dimensional geometric mask of the ascending aorta based on the amplitude image or phase-contrast magnetic resonance angiography image using threshold segmentation, random walk segmentation, or a deep learning segmentation network, locate the entrance interface of the ascending aorta according to anatomical landmarks, extract the volume of interest in the entrance region, and the volume of interest covers at least 1.5 times the pipe diameter upstream and at least 2 times the pipe diameter downstream of the entrance section; and a normalization processing unit, used to normalize the velocity components and amplitude image in the entrance region to generate standardized input data.

[0033] Specifically, the standard input data generation module 21 acquires the patient's low-resolution 4D Flow MRI raw data, including three-dimensional velocity-coded phase images ( v x , v y, v z The original data (including the velocity components) and their corresponding amplitude images are subjected to phase unwrapping by a phase unwrapping unit. The real velocity encoding values ​​are recovered by an unwrapping algorithm based on spatial continuity constraints or a deep learning unwrapping network (such as DAF-FlowNet). The phase image conversion unit converts the phase image into physical velocity values ​​based on the velocity encoding values, generating a three-dimensional velocity component map.

[0034] The inlet region extraction unit extracts the three-dimensional geometric mask of the ascending aorta based on amplitude images or phase-contrast magnetic resonance angiography (PC-MRA) images using threshold segmentation, random walk segmentation, or deep learning segmentation networks (such as U-Net). It automatically locates the inlet section of the ascending aorta based on anatomical landmarks (such as the aortic valve annulus plane and the sinus duct junction plane), and extracts the volume of interest (VOI) in the inlet region. The VOI covers at least 1.5 times the vessel diameter upstream and at least 2 times the vessel diameter downstream of the inlet section. Finally, a normalization processing unit normalizes the velocity components and amplitude images within the inlet region to generate standardized input data.

[0035] The deep residual super-resolution network comprises an input layer, a low-resolution feature encoder, an upsampling layer, a high-resolution feature encoder, and an output layer. The input layer uses a dual-path parallel input structure, including an anatomical path and a velocity path. The anatomical path inputs PC-MRA, amplitude images, and their product image to provide vascular anatomy information. The velocity path inputs a low-resolution 3D velocity component map, carrying hemodynamic information to be reconstructed. The low-resolution feature encoder consists of multiple stacked residual blocks in a low-resolution space. Each residual block contains two 3×3×3 3D convolutional layers and skip connections. The upsampling layer is located in the middle of the network and uses sub-pixel convolution or transposed convolution to improve spatial resolution. The high-resolution feature decoder consists of multiple stacked residual blocks in a high-resolution space for refined reconstruction. The output layer outputs a 3D high-resolution velocity component map and includes global skip connections to add the result of upsampling the input low-resolution velocity field through trilinear interpolation to the network residual output, obtaining the super-resolution velocity field dataset.

[0036] Specifically, when constructing a deep residual super-resolution network, a dual-path input, encoder-decoder structure is adopted.

[0037] The input layer employs a dual-path parallel input structure: the anatomical path inputs PC-MRA images, amplitude images, and their product image, providing vascular anatomy information; the velocity path inputs low-resolution three-dimensional velocity component maps (…). v x , v y, v z The anatomical path and velocity path carry the hemodynamic information to be reconstructed. Preferably, after the anatomical path and velocity path have their features extracted by independent convolutional layers in the initial layer, the anatomical-velocity feature coupling is achieved through a cross-path feature fusion mechanism.

[0038] The low-resolution feature encoder consists of multiple stacked residual blocks set in a low-resolution space. Each residual block contains two 3×3×3 three-dimensional convolutional layers (followed by batch normalization and ReLU activation function) and skip connections. Preferably, densely connected residual blocks or cross-stage partial connection structures are used instead of standard residual blocks.

[0039] The upsampling layer is located in the middle of the network and uses subpixel convolution or transposed convolution to improve spatial resolution.

[0040] The high-resolution feature decoder consists of multiple stacked residual blocks, set in a high-resolution space, for fine-grained reconstruction.

[0041] The output layer outputs a three-dimensional high-resolution velocity component map. v x ,SR, v y ,SR, v z (SR). Global residual learning sets up global skip connections at the end of the network, adding the result of upsampling the input low-resolution velocity field through trilinear interpolation to the network residual output to form the final super-resolution velocity field. The total number of parameters in the deep residual super-resolution network is controlled between 2 million and 3.5 million.

[0042] The boundary condition transformation module 25 includes: a patching inference unit, used to input the preprocessed patient 4DFlow MRI entry region data into the trained depth residual super-resolution network and perform patching inference; a divergence correction unit, used to perform divergence correction on the super-resolution velocity field dataset using the projection method or the divergence free radial basis function method, set the velocity values ​​of non-fluid regions to zero based on the PC-MRA mask, and denormalize to physical velocity units; and a boundary condition transformation unit, used to map the corrected super-resolution velocity field dataset to the CFD entry section grid nodes using trilinear or higher-order interpolation, map the time series to the CFD simulation time step, obtain the CFD simulation entry boundary conditions, and calculate the spatial average velocity, peak velocity, flow waveform, and Reynolds number time-varying curve for comparison and verification with clinical reference values. If the verification is successful, the data is output in a preset standard format.

[0043] Specifically, the boundary condition transformation module 25 is used to realize patient-specific ascending aortic inlet velocity field super-resolution reconstruction and CFD interface adaptation. Among them, the patching inference unit inputs the preprocessed patient low-resolution 4D Flow MRI inlet region data into the trained depth residual super-resolution network and performs patching inference: dividing the inlet region into overlapping spatial patches (step size is the patch size minus the boundary overlap width), performing weighted average stitching after super-resolution of each patch, and eliminating patch boundary artifacts.

[0044] The reconstructed velocity field is corrected by using divergence correction elements (the algorithm can be either the projection method or the divergence free radial basis function method). Based on the PC-MRA mask, the velocity values ​​in the non-fluid region are set to zero and then denormalized to physical velocity units (cm / s).

[0045] The boundary condition transformation unit converts the reconstructed high-resolution inlet velocity field dataset into CFD simulation inlet boundary conditions. This includes: mapping the velocity field to CFD inlet section grid nodes using trilinear or higher-order interpolation; mapping the time series to CFD simulation time steps; calculating the spatial average velocity, peak velocity, flow waveform, and Reynolds number time-varying curves for comparison and verification with clinical reference values; and outputting the data in a standard format (e.g., VTK, VTI, CSV, or OpenFOAM boundary condition format) for direct use by subsequent CFD simulation modules.

[0046] like Figure 3 As shown, the second boundary condition generation subsystem 30 includes: an ascending aorta geometry model reconstruction module 31, used to acquire the patient's static CT / MRI data and reconstruct the ascending aorta geometry model based on the static CT / MRI data; an anatomical landmark alignment module 32, used to perform patient-specific inlet and outlet plane localization and anatomical landmark alignment based on the reconstructed ascending aorta geometry model; a velocity profile extraction module 33, used to construct a large-scale clinical cohort 4D Flow MRI database, extract 4D Flow MRI velocity profiles from the clinical cohort, and perform spatiotemporal alignment; and a statistical model construction module 34, used to construct a statistical model based on the spatiotemporally aligned 4D Flow MRI profiles. The system includes: an MRI velocity profile module 35, which constructs a 4D velocity profile statistical shape model based on principal component analysis; a velocity profile set construction module 36, which generates synthetic velocity profiles and filters acceptance criteria to construct a synthetic 4D velocity profile dataset that meets a preset size; a velocity profile generation module 37, which generates patient-specific velocity profiles and associates them with anatomical morphology based on the synthetic 4D velocity profile dataset; an entry boundary condition generation module 38, which maps the patient-specific boundary condition space and adapts it to the CFD interface to generate CFD simulation entry boundary conditions; and an uncertainty quantification module 39, which quantifies the uncertainty of the CFD simulation entry boundary conditions and performs sensitivity analysis.

[0047] Specifically, the second boundary condition generation subsystem 30 is used to generate accurate and personalized hemodynamic boundary conditions based on the patient's static anatomical morphological characteristics and combined with the statistical laws of a large population, providing reliable entry and exit boundary conditions for CFD simulation. This subsystem is designed for patients who only have static CT / MRI and no 4D Flow data.

[0048] The ascending aorta geometry reconstruction module 31 is used for acquiring static anatomical images of the patient and reconstructing the geometry of the ascending aorta. It acquires static anatomical image data of the patient (e.g., CT angiography, enhanced CT, or high-resolution MRA), and uses a semi-automatic segmentation tool (such as ITK-SNAP) or a deep learning segmentation network (such as U-Net, nnU-Net) to perform three-dimensional segmentation of the ascending aorta and adjacent key anatomical structures, generating a binary vascular mask. Based on the mask, the Marching Cubes algorithm is used to generate a three-dimensional surface mesh of the aorta and performs Taubin smoothing. The centerline of the ascending aorta is extracted using a vascular modeling toolkit. Key anatomical parameters are measured in the ascending aorta inlet region, including the aortic diameter, luminal cross-sectional area, major axis / minor axis ratio, aortic arch curvature radius, and ascending aorta length at the inlet plane, establishing a static anatomical morphological feature vector of the patient.

[0049] The velocity profile extraction module 33 is used to extract and spatiotemporally align the velocity profiles of 4D Flow MRI in a clinical cohort. A large-scale clinical cohort 4D Flow MRI database is constructed, retrospectively collecting 4D Flow MRI data from patients with thoracic aortic disease (preferably ascending aortic aneurysm). The spatial resolution is preferably 1.4-2.0 mm, the flow velocity encoding is preferably 150-200 cm / s, and the temporal resolution is preferably 20-28 frames / cardiac cycle. Phase unwinding and velocity field calculation are performed on the 4D Flow MRI data to obtain the three-dimensional time-varying velocity field for each patient. Linear radial basis function interpolation is used to resample the velocity profiles to a fixed set of spatial nodes on a unit radius disk (preferably...). N =1071 uniformly distributed nodes), and zero velocity vectors are assigned to reference nodes that extend beyond the original plane in non-circular cross-sections; the time series of each patient's velocity profile are normalized to the standard cardiac cycle interval. t∈ [0 , 1], using cubic polynomial interpolation to unify to a fixed number of frames (preferably) T = 20 frames); will be spatiotemporally aligned J For example, patient velocity profile data are reorganized into a data matrix V = [V(1)] using column vectors. , . . . , V( J [The total dimension of the velocity profile of a single patient] P = 3 ×N×T .

[0050] The statistical model building module 34 constructs a statistical shape model based on the 4D velocity profile of principal component analysis. It calculates the mean 4D velocity profile and covariance matrix of the clinical cohort, performs eigenvalue decomposition on the covariance matrix, sorts it in descending order, and selects the top M principal component modes (preferably, the top M eigenvectors with cumulative explained variance ≥ 90%). Each mode corresponds to an eigenvalue, establishing a statistical shape model of the 4D velocity profile, obtaining mode weight coefficients, and quantitatively analyzing the physiological significance of each principal component mode. By fixing a single mode and uniformly varying its weight coefficients within the interval [−3, 3], a series of velocity profiles are generated, and flow morphology descriptors are calculated. The correlation between modes and specific blood flow characteristics (such as peak velocity, jet direction, secondary flow intensity, etc.) is analyzed, and a mode-physiological characteristic mapping table is established.

[0051] The velocity profile set construction module 35 realizes the generation of synthetic velocity profiles and the screening of acceptance criteria. The weight coefficients of each principal component mode are uniformly and randomly sampled (preferably within the range of [−1.5, 1.5] or [−3, 3]). A large number of candidate synthetic velocity profiles are generated by combining the sampling weights of M modes. The following flow morphology descriptors are calculated for the clinical cohort and the synthetic profiles: peak velocity, flow dispersion index (FDI, defined as the ratio of the area of ​​the top 15% of the peak velocity region to the total inlet area), flow jet angle (FJA, the angle between the average velocity direction and the normal to the inlet plane), secondary flow degree, and recirculation index (RFI, the ratio of the absolute value of the reverse flow integral to the absolute value of the total flow integral during the cardiac cycle). The acceptance interval is set based on the statistical distribution of each descriptor in the clinical cohort. Only when all descriptors of the synthetic profile fall into the corresponding acceptance interval are they included in the final synthetic dataset. This process is repeated until a synthetic 4D velocity profile dataset that meets the preset size is generated (preferably, 437 synthetic profiles that meet the acceptance criteria are selected and retained from 500 candidate profiles).

[0052] The velocity profile generation module 36 enables personalized generation of patient-specific velocity profiles and their association with anatomical morphology. It analyzes the statistical association between static anatomical parameters (inlet diameter, lumen area, long-axis / short-axis ratio, aortic arch curvature, etc.) and PCA modal weights in the clinical cohort. Pearson correlation analysis, multiple linear regression, or random forest regression are used to establish a mapping model between anatomical morphological feature vectors and modal weight vectors. The static anatomical morphological feature vector of the target patient is input into the mapping model to predict personalized PCA modal weight vectors. The predicted modal weights are substituted into the statistical shape model to generate patient-specific synthetic velocity profiles. If the prediction accuracy of the association model is insufficient, a nearest neighbor matching strategy is used to calculate the Euclidean or Mahalanobis distance between the target patient's anatomical feature vector and the corresponding anatomical feature vectors of each profile in the synthetic dataset. K nearest neighbor synthetic profiles (preferably K = 5 or K = 10) are selected, and a patient-specific velocity profile is generated by weighted averaging.

[0053] Uncertainty quantification module 38 is used to quantify boundary condition uncertainties and perform sensitivity analysis. Based on the predicted modal weights, random perturbations conforming to the population statistical distribution (such as Gaussian perturbations within ±1 standard deviation) are applied to generate Nperturb perturbation velocity profiles (preferably Nperturb = 100), forming a boundary condition uncertainty set. Each boundary condition in the uncertainty set is input into the CFD solver to perform parallel simulation and obtain a hemodynamic response set. The mean, standard deviation, and 95% confidence interval are calculated for the CFD response set to identify the hemodynamic parameters and anatomical regions most sensitive to boundary condition variations. Sensitivity analysis is used to identify the PCA master mode that has the greatest impact on the CFD results, prioritizing the accuracy of the weight prediction for key modes.

[0054] The anatomical landmark alignment module 32 includes: an inlet positioning unit, used to locate the inlet and inlet plane required for CFD simulation based on the reconstructed ascending aorta geometric model using an anatomical landmark-driven method; an outlet plane setting unit, used to set the outlet plane at the distal end of the descending aorta or at the branch vessels of the aortic arch according to the CFD simulation domain definition; a three-dimensional coordinate recording unit, used to perform triangulation mesh discretization on the inlet and outlet planes and record the three-dimensional coordinates of each node in the patient coordinate system; and a rigid transformation unit, used to perform a rigid transformation on the inlet plane so that the inlet plane meets the set standard spatial orientation.

[0055] Specifically, the anatomical landmark alignment module 32 can achieve patient-specific inlet and outlet plane positioning and alignment with anatomical landmarks.

[0056] The inlet positioning unit is based on the reconstructed ascending aorta geometric model and uses an anatomical landmark-driven method to locate the inlet and outlet planes required for CFD simulation: using the bifurcation point of the pulmonary artery as a reference landmark, the bifurcation point is identified in PC-MRA or CTA images; a corresponding point with the same axial (head-to-foot) coordinates as the bifurcation point is located on the aortic centerline; the inlet point is determined by tracing a preset arc length (preferably 30 mm) upstream along the aortic centerline, and the plane passing through this point and locally perpendicular to the centerline is defined as the inlet plane.

[0057] The exit plane setting unit sets the exit plane at the distal end of the descending aorta or at the branch vessels of the aortic arch, based on the CFD simulation domain definition.

[0058] The inlet and outlet planes are discretized by triangulation using a three-dimensional coordinate recording unit, and the three-dimensional coordinates of each node in the patient coordinate system are recorded.

[0059] A rigid transformation unit is used to perform a rigid transformation on the entrance plane to ensure that it meets the standard spatial orientation: first, the center of the entrance plane is moved to the origin of the coordinate system through a translation transformation, then a first rigid rotation is performed to align the normal of the entrance plane with the global z-axis, and a second rigid rotation is performed to align the x-axis in the plane with the left and right directions of the patient; the same coordinate system alignment process is performed on the exit plane to ensure that the entrance and exit planes of all patients are comparable in the standard coordinate system.

[0060] The inlet boundary condition generation module 37 includes: an inverse rigid transformation unit, used to perform the inverse operation of rigid transformation on the patient-specific velocity profile, restoring the velocity vector and node coordinates from the standard coordinate system to the patient's original coordinate system; a velocity vector mapping unit, used to map the fixed node velocity vector on the unit disk to the original triangulated mesh nodes of the patient's inlet plane through inverse radial basis function interpolation, keeping the velocity of nodes outside the vessel lumen zero for non-circular sections; a velocity profile mapping unit, used to map the velocity profile of a set frame within the standard cardiac cycle t∈[0, 1] to the physical time axis according to the patient's actual heart rate; a velocity field dataset output unit, used to output the time-varying three-dimensional velocity field dataset of the inlet plane in a set boundary condition format to obtain the CFD simulation inlet boundary conditions; and an outlet boundary condition generation unit, based on the inlet velocity profile and the principle of mass conservation, combined with the patient-specific aortic geometry model, derives the pressure-flow relationship at the distal outlet of the descending aorta through a one-dimensional hemodynamic model or empirical proportional relationship, using Windkessel... The model or impedance boundary condition model generates the outlet boundary conditions; the inlet flow waveform verification unit is used to verify the consistency between the inlet flow waveform and the patient's clinical reference value. If the deviation exceeds the preset threshold, the feedback correction mechanism is triggered to adjust the modal weights and regenerate the profile.

[0061] Specifically, the inlet boundary condition generation module 37 is used for patient-specific boundary condition spatial mapping and CFD interface adaptation. Specifically, an inverse rigid transformation unit performs inverse operations (inverse rotation and inverse translation) on the synthesized velocity profile, restoring the velocity vector and node coordinates from the standard coordinate system to the patient's original coordinate system. The velocity vector mapping unit maps the fixed node velocity vectors on the unit disk to the original triangulated mesh nodes of the patient's inlet plane using inverse radial basis function interpolation, maintaining zero velocity at nodes outside the vessel lumen for non-circular sections. The velocity profile mapping unit maps 20 frames of velocity profiles within the standard cardiac cycle t∈[0, 1] to the physical time axis according to the patient's actual heart rate. The time-varying three-dimensional velocity field dataset of the inlet plane is output in a preset standard format. The outlet boundary condition generation unit, based on the inlet velocity profile and the principle of mass conservation, combined with a patient-specific aortic geometry model, derives the pressure-flow relationship at the distal outlet of the descending aorta using a one-dimensional hemodynamic model or empirical proportional relationships, employing Windkessel... The model or impedance boundary condition model generates the outlet boundary conditions; the inlet flow waveform verification unit is used to verify the consistency between the inlet flow waveform and the patient's clinical reference value. If the deviation exceeds the preset threshold (e.g., flow deviation > 10%), the feedback correction mechanism is triggered to adjust the modal weights and regenerate the profile.

[0062] like Figure 4 As shown, the hemodynamic analysis subsystem 40 includes: a vessel wall geometry model construction module 41, used to acquire preoperative or postoperative image data of the patient and construct a vessel wall geometry model; a virtual surgical plan generation module 42, used to generate a virtual surgical plan for a covered stent micropore fenestration procedure based on the vessel wall geometry model; a coupling geometry module generation module 43, used to establish a parametric geometric model including the stent body skeleton, the covering layer, and the micropore array in the fenestration area according to the virtual surgical plan, and to perform registration and Boolean fusion based on anatomical landmarks and the vessel wall geometry model to generate a stent-vessel-micropore coupling geometric model; and a transmembrane flow coupling module 44, used to combine the inlet boundary conditions of the CFD simulation with the patient's systolic or diastolic blood pressure. The Windkessel three-element model or lumped parameter outflow boundary conditions with tension calibration are used to establish hemodynamic control equations based on micropore transmembrane flow, and transmembrane flow boundary conditions are obtained; the mesh generation module 45 is used to perform unstructured tetrahedral and hexahedral hybrid mesh generation on the stent-vessel-micropore coupled geometric model, and adaptive mesh refinement is performed; the surgical effect evaluation module 46 is used to extract postoperative hemodynamic parameters and evaluate the surgical effect, and obtain evaluation results. Hemodynamic parameters include at least one of the following: micropore outlet Reynolds number, branch artery laminar flow maintenance index, transmembrane pressure difference, micropore transmural flow, branch artery perfusion guarantee rate, branch artery pressure reduction rate, and false lumen / aneurysm pressure reduction rate.

[0063] Specifically, the hemodynamic analysis subsystem 40 is used for preoperative virtual surgical planning, stent-vessel coupling modeling, and postoperative quantitative hemodynamic assessment for covered stent micropore fenestration procedures.

[0064] The vessel wall geometry model construction module 41 is used to realize patient-specific anatomy and blood flow data reconstruction. It acquires preoperative or postoperative imaging data (CTA or 4D Flow MRI) of the patient, and extracts the vessel wall geometry model of the aortic arch, the three branches of the aortic arch, and the lesion area through image segmentation and three-dimensional reconstruction. If postoperative 4D Flow MRI data is acquired, a three-dimensional temporal-resolved blood flow velocity field is reconstructed after phase wrapping correction, background phase correction, and noise suppression.

[0065] The virtual surgical planning generation module 42 is used to generate a virtual surgical plan for a stent graft micropore fenestration procedure. Based on the vascular wall geometry model, the procedure is planned in the virtual surgical planning submodule: the diameter and length of the stent body are selected; the location of the fenestration area on the stent graft is determined, wherein the angle of the fenestration area in the circumferential direction of the stent graft is 45°-360°, and the length in the axial direction is no more than 12cm; micropore array parameters are configured in the fenestration area, wherein the micropore diameter is 1.5-1.8mm, the density is 25-36 per 2 square centimeters, and the arrangement is either an interlaced array or a rectangular array; based on the three-dimensional curvature of the patient's aortic arch, the finite element deformation algorithm is used to simulate the stent's adherence to the wall after release, the spatial deviation between the center of the fenestration area and the center of the target branch artery opening is calculated, and the position of the fenestration area is adjusted according to the deviation.

[0066] The coupling geometry module generation module 43 is used to realize parametric geometric modeling of the micropores of the covered stent. Based on the virtual surgical plan parameters, a parametric geometric model is established, including the stent body skeleton, the covering layer, and the array of micropores in the opening area. The micropores are generated into a continuous topological structure on the covering layer through cylindrical Boolean difference operations, and the micropore axis is perpendicular to the tangent plane of the local covering surface. The parametric geometric model and the vessel wall geometric model are registered and fused with Boolean based on anatomical landmarks to generate a stent-vessel-micropore coupled geometric model.

[0067] The mesh generation module 45 is used to implement adaptive mesh generation and transient solution. Unstructured tetrahedral / hexahedral hybrid mesh generation is performed on the stent-vessel-micropore coupled geometric model, employing an adaptive mesh refinement strategy: local refinement is performed in the covered stent and micropore opening regions, with the mesh scale in the micropore region not exceeding 1 / 5 of the minimum micropore diameter; an expanded boundary layer mesh is generated near the stent wire diameter and on the vessel wall, with the first layer mesh height satisfying y... + ≤1; Dynamically densify the high-shear region of the branch artery opening and micropore outlet based on the velocity gradient criterion; Set the time step to meet the CFL condition and solve for at least 3 cardiac cycles until the flow field is periodically stable.

[0068] The surgical outcome assessment module 46 is used to extract postoperative hemodynamic parameters and assess surgical outcomes. Based on the solution results, specific postoperative hemodynamic assessment parameters for the covered stent micropore fenestration procedure are extracted, including: spatiotemporal distribution of wall shear stress and oscillatory shear index; micropore outlet Reynolds number and branch artery laminar flow maintenance index, where the laminar flow maintenance index is defined as the ratio of non-vortex volume in the branch artery to the total volume of the branch artery, and vortices are identified using the Q criterion or λ² criterion; transmembrane pressure difference (pressure difference between the stent lumen and the branch artery / lesion lumen) and micropore transwall flow rate (obtained by integrating the normal velocity component of the micropore surface); branch artery perfusion guarantee rate (postoperative branch flow rate / preoperative branch flow rate × 100%) and branch artery pressure reduction rate; false lumen / aneurysm pressure reduction rate and false lumen thrombosis index; vortex structure and energy loss coefficient.

[0069] The transmembrane flow coupling module 44 includes: a dynamic control equation establishment unit, used to obtain the inlet boundary conditions of CFD simulation, and combine the Windkessel three-element model or lumped parameter outflow boundary conditions based on the patient's systolic / diastolic blood pressure calibration to establish hemodynamic control equations considering microporous transmembrane flow. The control equations introduce transmembrane flow boundary conditions or equivalent porous medium source terms in the micropore region. The transmembrane flow boundary conditions calculate the local flow resistance based on the micropore diameter and micropore density, so that when blood flows from the stent lumen to the outside through the micropores, it satisfies the laminar flow constraint; and a micropore array equivalent unit, used to adopt an equivalent porous wall model when the micropore scale is smaller than a set multiple of the grid scale, and to equate the micropore array to a thin film boundary with specific permeability and inertial resistance.

[0070] Specifically, the transmembrane flow coupling module 44 considers the CFD coupling boundary conditions and physical model establishment for microporous transmembrane flow. The kinetic control equation establishment unit uses the high-resolution velocity field output by the first boundary condition generation subsystem or the inlet velocity waveform predicted by the second boundary condition generation subsystem as the inlet boundary condition. Combined with the Windkessel three-element model or lumped parameter outflow boundary condition based on the patient's systolic / diastolic blood pressure calibration, the hemodynamic control equation considering microporous transmembrane flow is established. The control equation introduces transmembrane flow boundary conditions or equivalent porous medium source terms in the micropore region. The transmembrane flow boundary conditions calculate the local flow resistance based on the micropore diameter and micropore density, so that when blood flows from the stent lumen through the micropores to the outside, it satisfies the laminar flow constraint (micropore outlet Reynolds number Re < 2300). When the micropore scale is much smaller than the grid scale, the micropore array equivalent unit adopts an equivalent porous wall model, which equates the micropore array to a thin film boundary with specific permeability and inertial resistance.

[0071] In summary, the present invention has the following advantages compared with the prior art: First, a complete technological closed loop is constructed: the three subsystems form a complete software system from low-quality image data enhancement and personalized boundary condition generation to specialized surgical procedure evaluation, filling the gap in the technological chain from raw data to clinical decision support. The first boundary condition generation subsystem is for patients with existing 4D Flow MRI data to perform super-resolution reconstruction; the second boundary condition generation subsystem is for patients with only static CT / MRI and no 4D Flow data, generating personalized boundary conditions based on population statistical patterns; the two modules complement each other to cover different data availability scenarios, and both provide entry boundary conditions to the hemodynamic analysis subsystem.

[0072] Second, it breaks through the data resolution bottleneck: The first boundary condition generation subsystem uses super-resolution reconstruction technology based on HemoRes-Net deep residual network, employing dual-path input (anatomical path + velocity path), physical perception data synthesis (k-space truncation, noise, partial volume effect simulation), multi-stage training (pre-training + fine-tuning + distribution learning), and physical constraint loss (divergence constraint, amplitude consistency) to upgrade low-resolution 4D Flow MRI data to a usable high-resolution patient-specific velocity field, breaking through the limitations of traditional imaging hardware and enabling full utilization of existing clinical data resources.

[0073] Third, it enables non-invasive, personalized boundary condition generation: The second boundary condition generation subsystem establishes a quantitative correlation between patient anatomical morphology and blood flow function through statistical shape modeling and principal component analysis. It uses flow morphology descriptors and acceptance criteria to ensure the physiological rationality of the synthesized profile. Dynamic personalized boundary conditions can be generated from conventional CTA / MRA images without additional flow velocity measurement examinations, significantly improving the patient specificity and clinical feasibility of CFD simulation.

[0074] Fourth, procedure-specific virtual planning and assessment: The hemodynamic analysis subsystem has for the first time established a dedicated toolchain for preoperative virtual surgical planning and postoperative hemodynamic assessment of the covered stent micropore fenestration procedure, supporting the range of circumferential angle (45°–360°), axial length (≤12cm), micropore diameter (1.5-1.8mm), and density (25-36 per 2cm) of the fenestration area. 2 The parameterized configuration and preoperative simulation of the stent enable surgeons to optimize stent selection and micropore layout before surgery.

[0075] Fifth, quantitative verification of micropore laminar flow characteristics: The hemodynamic analysis subsystem, through local grid refinement and transmembrane flow boundary conditions at the micropore scale, has for the first time achieved quantitative CFD verification of the core mechanism of this procedure, namely, the maintenance of laminar flow in branch arteries by micropores. It can directly output the micropore outlet Reynolds number and laminar flow maintenance index, providing hemodynamic basis for the safety assessment of the procedure.

[0076] Sixth, direct quantification of surgical efficacy: The hemodynamic analysis subsystem establishes assessment indicators that directly correspond to clinical efficacy, such as branch artery perfusion guarantee rate, pressure reduction rate, and false lumen pressure reduction rate, filling the gap in the postoperative hemodynamic evaluation system for this procedure and providing decision support for clinical follow-up and surgical optimization.

[0077] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered within the scope of protection of the present invention.

Claims

1. A hemodynamic analysis and assessment system specific to patients with aortic arch lesions, characterized in that, include: The system includes a patient data acquisition module, a first boundary condition generation subsystem, a second boundary condition generation subsystem, and a hemodynamic analysis subsystem, wherein the first boundary condition generation subsystem and the second boundary condition generation subsystem are both connected to the patient data acquisition module and the hemodynamic analysis subsystem. The patient data acquisition module is used to acquire the patient's 4DFlow MRI raw data or static CT / MRI data; The first boundary condition generation subsystem is used to construct a patient-specific super-resolution velocity field dataset of the aortic inlet based on the patient's original 4DFlow MRI data, and convert it to obtain the CFD simulation inlet boundary conditions; The second boundary condition generation subsystem is used to generate CFD simulation entry boundary conditions based on the patient's static CT / MRI data and combined with the statistical laws of a large-scale population. The hemodynamic analysis subsystem is used to perform preoperative virtual planning, parametric modeling, and postoperative evaluation of covered stents based on the CFD simulation inlet boundary conditions, and to obtain evaluation results.

2. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 1, characterized in that, The first boundary condition generation subsystem includes: The standard input data generation module is used to preprocess the raw 4DFlow MRI data and extract the ascending aortic inlet region, and normalize it to obtain standard input data. The synthetic training data generation module is used to generate synthetic training data based on CFD simulation of physical perception. The super-resolution network construction module is used to construct a deep residual super-resolution network and input the standard input data to obtain a super-resolution velocity field dataset. A synthetic training data adjustment module is used to pre-train the synthetic training data and fine-tune it with clinical data. The boundary condition transformation module is used to convert the super-resolution velocity field dataset of the patient-specific ascending aortic inlet into the inlet boundary conditions for CFD simulation. The reconstruction quality assessment module is used to assess the reconstruction quality and set reconstruction pass conditions. When the reconstruction pass conditions are met, the CFD simulation entry boundary conditions are transmitted to the hemodynamic learning module.

3. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 2, characterized in that, The standard input data generation module includes: The phase unwrapping unit is used to unwrap the original 4DFlow MRI data using an unwrapping algorithm based on spatial continuity constraints or a deep learning unwrapping algorithm to obtain the true velocity encoding value; A phase image conversion unit is used to convert a phase image into a physical velocity value based on the velocity encoding value, and generate a three-dimensional velocity component map. The inlet region extraction unit is used to extract the three-dimensional geometric mask of the ascending aorta based on the amplitude image or phase contrast magnetic resonance angiography image, using threshold segmentation, random walk segmentation or deep learning segmentation network, locate the inlet interface of the ascending aorta according to anatomical landmarks, and extract the volume of interest in the inlet region. The volume of interest covers at least 1.5 times the pipe diameter upstream and at least 2 times the pipe diameter downstream of the inlet section. The normalization processing unit is used to normalize the velocity components and amplitude images within the entry region to generate standardized input data.

4. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 2, characterized in that, The deep residual super-resolution network includes: Input layer, low-resolution feature encoder, upsampling layer, high-resolution feature encoder, and output layer; The input layer adopts a dual-path parallel input structure, including an anatomical path and a velocity path. The anatomical path inputs PC-MRA and amplitude images, as well as their product image, to provide vascular anatomy information. The velocity path inputs low-resolution three-dimensional velocity component maps, carrying hemodynamic information to be reconstructed. The low-resolution feature encoder is composed of multiple stacked residual blocks, set in a low-resolution space. Each residual block contains two 3×3×3 three-dimensional convolutional layers and skip connections. The upsampling layer is located in the middle of the network and uses subpixel convolution or transposed convolution to improve spatial resolution. The high-resolution feature decoder is composed of multiple stacked residual blocks, set in a high-resolution space, for fine-grained reconstruction; The output layer outputs a three-dimensional high-resolution velocity component map, and a global skip connection is set in the input layer to add the result of the input low-resolution velocity field after trilinear interpolation upsampling to the network residual output to obtain a super-resolution velocity field dataset.

5. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 2, characterized in that, The boundary condition transformation module includes: The patched inference unit is used to input preprocessed patient 4D Flow MRI entry region data into the trained depth residual super-resolution network and perform patched inference. The divergence correction unit is used to perform divergence correction on the super-resolution velocity field dataset using the projection method or the divergence free radial basis function method, and to set the velocity values ​​in the non-fluid region to zero based on the PC-MRA mask and denormalize them to physical velocity units. The boundary condition transformation unit is used to map the corrected super-resolution velocity field dataset to the CFD inlet section grid nodes using trilinear or higher-order interpolation, map the time series to the CFD simulation time step, obtain the CFD simulation inlet boundary conditions, and calculate the spatial average velocity, peak velocity, flow waveform and Reynolds number time-varying curve for comparison and verification with clinical reference values. If the verification is successful, the data is output in a preset standard format.

6. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 1, characterized in that, The second boundary condition generation subsystem includes: The ascending aorta geometric model reconstruction module is used to acquire the patient's static CT / MRI data and reconstruct the ascending aorta geometric model based on the static CT / MRI data. The anatomical landmark alignment module is used to perform patient-specific inlet and outlet plane localization and anatomical landmark alignment based on the reconstructed ascending aorta geometry model. The velocity profile extraction module is used to construct a large-scale clinical cohort 4D Flow MRI database, based on the extracted 4D Flow MRI velocity profiles of the clinical cohorts and spatiotemporal alignment. The statistical model building module is used to construct a statistical shape model of the 4D velocity profile based on principal component analysis, based on the spatiotemporally aligned 4D Flow MRI velocity profile. The velocity profile set construction module is used to generate synthetic velocity profiles and select acceptance criteria to build a synthetic 4D velocity profile dataset that meets the preset size. The velocity profile generation module is used to generate patient-specific velocity profiles by personalizing the generation of patient-specific velocity profiles and associating them with anatomical morphology based on the synthetic 4D velocity profile dataset. The entry boundary condition generation module is used to map the patient-specific boundary condition space and adapt to the CFD interface to generate CFD simulation entry boundary conditions. The uncertainty quantification module is used to quantify the uncertainty of the CFD simulation entry boundary conditions and perform sensitivity analysis.

7. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 6, characterized in that, The anatomical landmark alignment module includes: The inlet localization unit is used to locate the inlet and inlet plane required for CFD simulation based on the reconstructed ascending aorta geometry model, using an anatomical landmark-driven method. The exit plane setting unit sets the exit plane at the distal end of the descending aorta or at the branch vessels of the aortic arch, according to the CFD simulation domain definition. A three-dimensional coordinate recording unit is used to perform triangulation mesh discretization on the inlet plane and the outlet plane, and record the three-dimensional coordinates of each node in the patient coordinate system; A rigid transformation unit is used to perform a rigid transformation on the entrance plane so that the entrance plane meets a set standard spatial orientation.

8. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 7, characterized in that, The inlet boundary condition generation module includes: The inverse rigid transformation unit is used to perform the inverse operation of the rigid transformation on the patient-specific velocity profile, restoring the velocity vector and nodal coordinates from the standard coordinate system to the patient's original coordinate system; The velocity vector mapping unit is used to map the fixed node velocity vectors on the unit disk to the original triangulated mesh nodes on the patient inlet plane through inverse radial basis function interpolation, keeping the extravascular node velocity zero for non-circular cross sections; The velocity profile mapping unit is used to map the velocity profile of a set frame within the standard cardiac cycle t∈[0, 1] to the physical time axis according to the patient's actual heart rate; The velocity field dataset output unit is used to output the time-varying three-dimensional velocity field dataset of the inlet plane in a set boundary condition format to obtain the inlet boundary conditions for CFD simulation. The outlet boundary condition generation unit, based on the inlet velocity profile and the principle of mass conservation, combined with the patient-specific aortic geometry model, derives the pressure-flow relationship at the distal outlet of the descending aorta through a one-dimensional hemodynamic model or empirical proportional relationship, and generates the outlet boundary conditions using the Windkessel model or impedance boundary condition model. The inlet flow waveform verification unit is used to verify the consistency between the inlet flow waveform and the patient's clinical reference value. If the deviation exceeds the preset threshold, the feedback correction mechanism is triggered to adjust the modal weights and regenerate the profile.

9. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 1, characterized in that, The hemodynamic analysis subsystem includes: The vessel wall geometry model construction module is used to acquire preoperative or postoperative imaging data of patients and construct a vessel wall geometry model. The virtual surgical plan generation module is used to generate a virtual surgical plan for the covered stent micropore fenestration procedure based on the aforementioned vascular wall geometry model. The coupling geometry module generation module is used to establish a parametric geometric model including the stent body skeleton, the covering layer and the micropore array in the opening area according to the virtual surgical plan, and to perform registration and Boolean fusion based on anatomical landmarks and the vessel wall geometric model to generate a stent-vessel-micropore coupling geometric model. The transmembrane flow coupling module is used to establish hemodynamic control equations based on microporous transmembrane flow, based on the CFD simulation inlet boundary conditions and combined with the Windkessel three-element model or lumped parameter outflow boundary conditions calibrated based on the patient's systolic or diastolic blood pressure, to obtain the transmembrane flow boundary conditions. The mesh generation module is used to perform unstructured tetrahedral and hexahedral hybrid mesh generation on the stent-vessel-micropore coupled geometric model, and to perform adaptive mesh refinement. The surgical outcome assessment module is used to extract postoperative hemodynamic parameters and assess the surgical outcome to obtain assessment results. The hemodynamic parameters include at least one of the following: microorifice outlet Reynolds number, branch artery laminar flow maintenance index, transmembrane pressure difference, microorifice transmural flow, branch artery perfusion guarantee rate, branch artery pressure reduction rate, and false lumen / aneurysm pressure reduction rate.

10. The patient-specific hemodynamic analysis and evaluation system for aortic arch lesions according to claim 9, characterized in that, The transmembrane flow coupling module includes: The dynamic control equation establishment unit is used to obtain the inlet boundary conditions of the CFD simulation, and combine the Windkessel three-element model or lumped parameter outflow boundary conditions based on the patient's systolic / diastolic blood pressure calibration to establish hemodynamic control equations considering transmembrane flow in micropores. The control equations introduce transmembrane flow boundary conditions or equivalent porous media source terms in the micropore region. The transmembrane flow boundary conditions calculate the local flow resistance based on the micropore diameter and micropore density, so that when blood flows from the stent lumen to the outside through the micropores, laminar flow constraints are satisfied. The equivalent unit of the micropore array is used to represent the micropore array as a thin film boundary with specific permeability and inertial resistance when the micropore size is smaller than a set multiple of the grid size, using an equivalent porous wall model.

Images