Multimodal digital twin heart blood flow real-time simulation method and system

Abstract

The application provides a multi-modal digital twin heart blood flow real-time simulation method and system. The method comprises: acquiring multi-modal heart image data and heart synchronization data of a target object, and generating a unified heart cycle index; based on the unified heart cycle index, generating a heart structure motion fusion representation with modal credibility; constructing a heart anatomical topology graph and generating an individualized dynamic boundary condition set; constructing a digital twin heart blood flow dynamics model, inputting the digital twin heart blood flow dynamics model into a hybrid blood flow dynamics solver, and using local high-fidelity solving, graph structure reduced-order solving and physical constraint neural operator to cooperatively deduce a blood flow state field; updating the individualized dynamic boundary condition set, the modal credibility and a solving residual correction parameter, and outputting a real-time blood flow dynamics simulation result. The application can improve the accuracy of dynamic boundary determination, improve the real-time simulation efficiency, and enhance the model closed-loop correction capability.

Description

Technical Field

[0001] This application relates to the fields of medical image processing and digital twin simulation technology, and in particular to a multimodal digital twin real-time simulation method and system for cardiac blood flow. Background Technology

[0002] Digital twin cardiac hemodynamic simulation is an important direction for the integrated development of medical image processing, computational fluid dynamics, and intelligent modeling. It can be used to perform individualized simulations of cardiac structural motion and blood flow status. Existing technologies typically rely on cardiac CT, MRI, and ultrasound image data to segment and 3D reconstruct the cardiac structure, and then combine this with boundary conditions such as inlet flow rate and outlet pressure to perform blood flow simulation calculations. Some schemes also introduce reduced-order models or machine learning models to improve computational efficiency.

[0003] Because different modalities of imagery differ in acquisition time, spatial coordinates, cardiac phase, and imaging quality, existing multimodal fusion methods mostly remain at the level of static registration or structural overlay, making it difficult to form dynamic boundary conditions synchronized with the cardiac cycle. Meanwhile, traditional hemodynamic simulations rely on complex meshes and high-precision iterative solutions, resulting in long computation times and failing to meet real-time simulation requirements. On the other hand, purely data-driven models lack physical constraints and individualized correction mechanisms, easily producing simulation results that do not conform to blood flow continuity and boundary consistency. Furthermore, existing digital twin heart models typically lack mechanisms for bias identification, reliability adjustment, and parameter closed-loop updates based on real-time cardiac observation data, resulting in insufficient ability to continuously correct the model according to individual state changes. Summary of the Invention

[0004] In view of this, embodiments of this application provide a multimodal digital twin cardiac blood flow real-time simulation method and system to solve the problems of difficulty in determining multimodal dynamic boundaries, low real-time simulation efficiency, and insufficient model closed-loop correction in the prior art.

[0005] A first aspect of this application provides a method for real-time simulation of cardiac blood flow using a multimodal digital twin, comprising: acquiring multimodal cardiac image data and cardiac synchronization data of a target object; performing quality evaluation, coordinate registration, and cardiac phase normalization on the multimodal cardiac image data to generate a unified cardiac cycle index; based on the unified cardiac cycle index, performing anatomical structure identification, motion boundary extraction, and blood flow observation mapping on the multimodal cardiac image data to generate a cardiac structure motion fusion representation with modal credibility; constructing a cardiac anatomical topology map based on the cardiac structure motion fusion representation, and performing entry state analysis based on the cardiac anatomical topology map. The parameters of the outlet state, valve state, and wall motion state are inverted to generate an individualized dynamic boundary condition set. Based on the cardiac anatomical topology and the individualized dynamic boundary condition set, a digital twin cardiac hemodynamic model is constructed. The digital twin cardiac hemodynamic model is input into a hybrid hemodynamic solver, and the blood flow state field is deduced by using local high-fidelity solving, graph structure order reduction solving, and physical constraint neural operators. According to the simulation deviation between the current cardiac observation data and the blood flow state field, the individualized dynamic boundary condition set, modal credibility, and solution residual correction parameters are updated, and the real-time hemodynamic simulation results are output.

[0006] A second aspect of this application provides a multimodal digital twin cardiac blood flow real-time simulation system, comprising: an acquisition module for acquiring multimodal cardiac image data and cardiac synchronization data of a target object, performing quality evaluation, coordinate registration, and cardiac phase normalization on the multimodal cardiac image data, and generating a unified cardiac cycle index; a generation module for performing anatomical structure recognition, motion boundary extraction, and blood flow observation mapping on the multimodal cardiac image data based on the unified cardiac cycle index, and generating a cardiac structure motion fusion representation with modal credibility; and an inversion module for constructing a cardiac anatomical topology map based on the cardiac structure motion fusion representation, and performing entry state analysis based on the cardiac anatomical topology map. The system performs parameter inversion on the exit state, valve state, and wall motion state to generate an individualized dynamic boundary condition set; a construction module is used to construct a digital twin cardiac hemodynamic model based on the cardiac anatomical topology map and the individualized dynamic boundary condition set; a derivation module is used to input the digital twin cardiac hemodynamic model into a hybrid hemodynamic solver, and use local high-fidelity solution, graph structure order reduction solution, and physically constrained neural operators to jointly deduce the blood flow state field; an output module is used to update the individualized dynamic boundary condition set, modal confidence, and solution residual correction parameters according to the simulation deviation between the current cardiac observation data and the blood flow state field, and output the real-time hemodynamic simulation results.

[0007] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By acquiring multimodal cardiac imaging data and cardiac synchronization data of the target object, the multimodal cardiac imaging data is subjected to quality evaluation, coordinate registration, and cardiac phase normalization to generate a unified cardiac cycle index. Based on the unified cardiac cycle index, anatomical structure identification, motion boundary extraction, and blood flow observation mapping are performed on the multimodal cardiac imaging data to generate a cardiac structure motion fusion representation with modal credibility. A cardiac anatomical topology map is constructed based on the cardiac structure motion fusion representation, and parameter inversion is performed on the inlet state, outlet state, valve state, and wall motion state based on the cardiac anatomical topology map to generate an individualized dynamic boundary condition set. A digital twin cardiac hemodynamic model is constructed based on the cardiac anatomical topology map and the individualized dynamic boundary condition set. The digital twin cardiac hemodynamic model is input into a hybrid hemodynamic solver, and the blood flow state field is deduced by using local high-fidelity solving, graph structure order reduction solving, and physically constrained neural operators. Based on the simulation deviation between the current cardiac observation data and the blood flow state field, the individualized dynamic boundary condition set, modal credibility, and solution residual correction parameters are updated, and real-time hemodynamic simulation results are output. This application can improve the accuracy of dynamic boundary determination, enhance real-time simulation efficiency, and strengthen the model's closed-loop correction capability. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the multimodal digital twin real-time simulation method for cardiac blood flow provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of the multimodal digital twin real-time simulation system for cardiac blood flow provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] In existing technologies, digital twin cardiac hemodynamic simulation typically begins with structural segmentation and 3D reconstruction based on cardiac CT, MRI, and ultrasound imaging data, followed by blood flow simulation calculations using preset boundary conditions such as inlet flow rate and outlet pressure. Some solutions also shorten computation time through reduced-order models or machine learning models to meet the needs of personalized cardiac modeling and blood flow state analysis.

[0012] However, different modalities of cardiac images differ in acquisition time, spatial coordinates, cardiac phase, and imaging quality. Existing multimodal image fusion methods mostly focus on static registration or structural overlay, making it difficult to generate dynamic boundary conditions that change synchronously with the cardiac cycle. Meanwhile, traditional hemodynamic simulations rely on complex meshes and high-precision iterative solutions, resulting in insufficient real-time performance; purely data-driven models lack physical constraints and individualized correction mechanisms, easily leading to simulation results that do not match blood flow continuity and boundary consistency. Therefore, existing technologies suffer from difficulties in determining multimodal dynamic boundaries, low real-time simulation efficiency, and insufficient model closed-loop correction.

[0013] To address the aforementioned issues, this application provides a multimodal digital twin method for real-time simulation of cardiac blood flow. This method acquires multimodal cardiac imaging data and cardiac synchronization data of the target object, and generates a unified cardiac cycle index through quality assessment, coordinate registration, and cardiac phase normalization. Based on the unified cardiac cycle index, anatomical structure identification, motion boundary extraction, and blood flow observation mapping are performed to generate a cardiac structure motion fusion representation with modal credibility. Furthermore, a cardiac anatomical topology map is constructed, and based on this map, parameter inversion is performed on the blood flow boundary state and motion constraint state to generate an individualized dynamic boundary condition set.

[0014] Building upon this foundation, this application constructs a digital twin cardiac hemodynamic model based on a cardiac anatomical topology map and a set of individualized dynamic boundary conditions. This model is then input into a hybrid hemodynamic solver, which collaboratively infers the blood flow state field through local high-fidelity solving, graph-structured order reduction solving, and physically constrained neural operators. Furthermore, based on the simulation deviation between current cardiac observation data and the blood flow state field, this application updates the set of individualized dynamic boundary conditions, modal reliability, and solution residual correction parameters, outputting real-time hemodynamic simulation results.

[0015] Therefore, this application can transform the results of multimodal image fusion into dynamic boundary conditions that can be used for real-time simulation, and achieve continuous updating of the digital twin heart model through hybrid solution under physical constraints and deviation feedback correction, thereby improving the accuracy of dynamic boundary determination, enhancing real-time simulation efficiency, and strengthening the model's closed-loop correction capability.

[0016] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1 This is a flowchart illustrating the multimodal digital twin real-time simulation method for cardiac blood flow provided in this application embodiment. Figure 1 As shown, the method may specifically include: S101: Acquire multimodal cardiac imaging data and cardiac synchronization data of the target object, perform quality evaluation, coordinate registration and cardiac phase normalization on the multimodal cardiac imaging data, and generate a unified cardiac cycle index. S102, based on the unified cardiac cycle index, performs anatomical structure identification, motion boundary extraction and blood flow observation mapping on multimodal cardiac imaging data, and generates a cardiac structure motion fusion representation with modal credibility; S103, construct a cardiac anatomical topology map based on the cardiac structure motion fusion representation, and perform parameter inversion on the inlet state, outlet state, valve state and wall motion state based on the cardiac anatomical topology map to generate an individualized dynamic boundary condition set; S104, based on the cardiac anatomical topology map and individualized dynamic boundary condition set, constructs a digital twin cardiac hemodynamic model; S105: Input the digital twin cardiac hemodynamic model into the hybrid hemodynamic solver, and use local high-fidelity solution, graph structure order reduction solution and physical constraint neural operator to jointly infer the blood flow state field; S106 updates the individualized dynamic boundary condition set, modal reliability, and solution residual correction parameters based on the simulation deviation between the current cardiac observation data and the blood flow state field, and outputs the real-time hemodynamic simulation results.

[0018] In some embodiments, multimodal cardiac imaging data and cardiac synchronization data of the target object are acquired, and the multimodal cardiac imaging data are subjected to quality assessment, coordinate registration, and cardiac phase normalization to generate a unified cardiac cycle index, including: According to the preset acquisition protocol, multimodal cardiac imaging data and cardiac synchronization data are received, and an image data package with acquisition time, modality type and cardiac event identifier is generated; The quality of image data packets is evaluated based on image integrity, motion artifacts, and cardiac synchronization consistency, and modal quality weights are generated. Cross-modal coordinate registration of image data packets is performed based on modal quality weights, and the registered image data packets are mapped to the standard cardiac phase axis based on cardiac event identifiers to generate a unified cardiac cycle index.

[0019] Specifically, in a real-time simulation scenario of a digital twin of cardiac blood flow for a target object, the acquisition terminal receives multimodal cardiac imaging data and cardiac synchronization data according to a preset acquisition protocol. The multimodal cardiac imaging data may include cardiac computed tomography (CT) data, cardiac magnetic resonance cinnamon sequences, transthoracic echocardiography sequences, color Doppler flow velocity images, and four-dimensional phase-contrast magnetic resonance velocity field data. The cardiac synchronization data may include electrocardiogram gating signals, heart sound trigger signals, ultrasound valve opening and closing recognition results, and ventricular volume change curves.

[0020] For example, taking left ventricular blood flow simulation as an example, after receiving data from each modality, the system first encapsulates the data according to the acquisition device, acquisition time, spatial coordinate system, temporal resolution, heart rate status, and target cardiac cycle number to generate image data packets. Each frame of image data is associated with a modality type identifier and a cardiac event identifier. The cardiac event identifier can be determined by the main ECG peak, end-systole, end-diastole, aortic valve opening time, or mitral valve opening time, enabling subsequent processing to clearly define the temporal position of different image frames within the same cardiac cycle.

[0021] After generating the image data package, the system performs a quality evaluation on the image data package. Specifically, it can first detect image integrity to determine whether the target cardiac chamber, outflow tract, and valve region are completely covered; then detect motion artifacts by generating artifact evaluation values ​​using the jitter amplitude of adjacent frame boundaries, the sharpness of blood vessel edges, and local grayscale stability; it can also detect cardiac synchronization consistency by comparing the timestamps corresponding to the image frames with cardiac event identifiers to determine whether phase drift exists.

[0022] For example, when a cardiac magnetic resonance cine sequence covers the entire cardiac cycle and has stable boundaries, its modal quality weight can be set to 0.88; when a transthoracic ultrasound image is affected by acoustic window obstruction in part of the phase, its modal quality weight can be set to 0.63; and when the temporal resolution of the four-dimensional phase-contrast magnetic resonance velocity field is low but the velocity direction is stable, its modal quality weight can be set to 0.72. The modal quality weight is used not only to determine reference data during subsequent registration but also to limit the excessive influence of low-quality modes on the cardiac phase mapping results.

[0023] Furthermore, the system determines the reference mode and the mode to be registered for cross-modal coordinate registration based on modal quality weights. For example, cardiac computed tomography data with high geometric resolution and good image integrity can be selected as the spatial registration reference, registering cardiac magnetic resonance cine sequences and echocardiogram sequences to the same cardiac coordinate system. During registration, the system extracts anatomically consistent features such as the apex, annular plane, outflow tract centerline, and cardiac chamber boundaries, and generates cross-modal spatial transformation relationships by combining rigid transformation, non-rigid deformation field, and quality weight constraints. When the quality weight of a local region in a certain modality is low, the system reduces the constraint strength of that region in the deformation field solution to avoid registration offset caused by local artifacts. After registration, the image data of each modality has a unified coordinate reference, and the registration reliability is recorded for each frame of data.

[0024] After coordinate registration, the system maps the registered image data packets to the standard cardiac phase axis based on cardiac event identifiers. The standard cardiac phase axis can represent a complete cardiac cycle from 0% to 100%, with end-diastole corresponding to the initial phase, end-systole corresponding to the intermediate phase, and the next end-diastole corresponding to the final phase. For phase gaps caused by different sampling frequencies in different modalities, the system performs temporal compensation by combining registration reliability, boundary deformation trends of adjacent cardiac phases, and ventricular volume change curves.

[0025] For example, when a clear image is missing in 30% to 40% of the phase segment of an ultrasound sequence, the system can interpolate and generate a mapping relationship for that phase segment using the ventricular wall motion trajectory in the magnetic resonance cine sequence and the valve state changes in the preceding and following ultrasound frames. Ultimately, the system establishes corresponding available modal data, spatial transformation parameters, mass weights, and cardiac event sources for each standard cardiac phase, forming a unified cardiac cycle index.

[0026] Through the above processing, multimodal cardiac image data with inconsistent acquisition time, spatial coordinates, and cardiac phase can be unified into the same cardiac cycle framework, providing a reliable data foundation for subsequent cardiac structure motion fusion characterization, dynamic boundary condition inversion, and real-time hemodynamic simulation, thereby improving the consistency of multimodal image fusion and the accuracy of dynamic boundary determination.

[0027] In some embodiments, cross-modal coordinate registration of image data packets is performed according to modal quality weights, and the registered image data packets are mapped to standard cardiac phase axes based on cardiac event identifiers to generate a unified cardiac cycle index, including: The registration reference mode and the mode to be registered are determined based on the modal quality weights, and the anatomical consistency features between the registration reference mode and the mode to be registered are extracted. Based on anatomical consistency features, cross-modal spatial transformation relationships are established, coordinate unification of image data packets is performed, and registration reliability is generated. Based on cardiac event identifiers, the phase position of each registered image data packet on the standard cardiac phase axis is determined, and the missing phase is compensated for in conjunction with the registration reliability to generate a unified cardiac cycle index.

[0028] Specifically, in a real-time simulation scenario of digital twin cardiac blood flow oriented towards the left ventricle and aortic outflow tract, the system first reads the image data package corresponding to the target object. This image data package may include cardiac computed tomography (CT) data, cardiac magnetic resonance imaging (MRI) cine sequence, transthoracic echocardiography (TTE) sequence, and four-dimensional phase-contrast magnetic resonance (4D) velocity field data. The system obtains the modal quality weights based on the prior quality assessment results. For example, the CT data has clear structural boundaries and complete coverage, resulting in a modal quality weight of 0.91; the MRI cine sequence covers the entire cardiac cycle, resulting in a modal quality weight of 0.86; the TTE echocardiography has acoustic window obstruction in some phases, resulting in a modal quality weight of 0.62; and the four-dimensional phase-contrast magnetic resonance (4D) velocity field has low temporal resolution, resulting in a modal quality weight of 0.74. Based on these modal quality weights, the system selects the CT data as the registration baseline modality and uses the MRI cine sequence, TTE echocardiography sequence, and 4D phase-contrast magnetic resonance (4D) velocity field data as the modalities to be registered, ensuring that spatial registration is based on high-resolution anatomical structures.

[0029] After determining the registration reference mode and the mode to be registered, the system extracts anatomically consistent features from both modes. These features can include the apex position, mitral valve annulus plane, aortic valve annulus plane, left ventricular long axis, outflow tract centerline, and endocardial boundary. For cardiac magnetic resonance cine sequences, the system extracts the endocardial contour at end-diastole and end-systole and matches it with the corresponding structures in the registration reference mode. For transthoracic echocardiography sequences, the system determines the position of the two-dimensional ultrasound section in the three-dimensional cardiac coordinate system based on the annulus plane and apex direction. For four-dimensional phase-contrast magnetic resonance velocity field data, the system establishes a velocity observation coordinate mapping based on the spatial relationship between the velocity field sampling grid and the outflow tract centerline. During the anatomically consistent feature matching process, the system adjusts the matching contribution of different modal features according to modal quality weights, with lower-quality local regions only serving as auxiliary constraints in the spatial transformation solution.

[0030] Furthermore, the system establishes cross-modal spatial transformation relationships based on anatomically consistent features. For overall posture differences, rigid transformations are first used to complete translation, rotation, and scale unification; for local morphological differences in the myocardial wall, cardiac chamber boundaries, and outflow tract, non-rigid deformation fields are then used for local correction. Spatial smoothness constraints and anatomical topological consistency constraints are introduced during the deformation field solution process to ensure the continuity of cardiac chamber boundaries, valve annulus regions, and large vessel connection regions after registration. The system generates registration confidence based on the matching error of anatomically consistent features, deformation field smoothness, and modal quality weights involved in registration. For example, when the average distance error between the magnetic resonance imaging (MRI) cardiac chamber boundary and the computed tomography (CT) cardiac chamber boundary in a certain cardiac phase is less than 1.5 mm, the registration confidence of that phase can be set to high; when the annulus features are incomplete due to occlusion in the ultrasound section, the registration confidence of that section is reduced, and the contribution of this data is reduced in subsequent phase mapping.

[0031] After coordinate unification, the system determines the phase position of each registered image data packet on the standard cardiac phase axis based on cardiac event identifiers. The standard cardiac phase axis normalizes a complete cardiac cycle to 0% to 100%, with end-diastole as 0% phase, end-systole as approximately 35% to 45% phase, and the next end-diastole as 100% phase. The system maps the ECG main wave peak, valve opening time, valve closing time, and positive ventricular volume point to the standard cardiac phase axis and performs phase localization on the sampled frames of different modal data. For example, in a magnetic resonance cine sequence, frame 8 corresponds to end-systole, and the system maps this frame to 40% phase; in transthoracic echocardiography, the aortic valve opening frame corresponds to 18% phase, and the aortic valve closing frame corresponds to 43% phase; in a four-dimensional phase-contrast magnetic resonance velocity field, the peak jet frame corresponds to 32% phase.

[0032] For phase intervals on the standard cardiac phase axis lacking clear images or velocity observations, the system performs temporal compensation based on registration reliability. When the registration reliability of adjacent phases is high, the system interpolates based on boundary deformation trajectories and ventricular volume change curves; when the registration reliability of an adjacent phase is low, the system prioritizes the temporal information of the high-reliability mode and uses cardiac cycle continuity constraints to limit abrupt changes in the compensation results. For example, when 30% to 36% of the phase segment lacks clear ultrasound images, the system can compensate for the ultrasound phase gap using the left ventricular wall contraction trajectory of the magnetic resonance cine sequence and the trend of four-dimensional phase-contrast magnetic resonance velocity peak changes. Finally, the system establishes available modal data, spatial transformation parameters, registration reliability, cardiac event sources, and compensation status identifiers for each phase node in the standard cardiac phase axis, generating a unified cardiac cycle index.

[0033] Through the above embodiments, the system can achieve coordinate and phase unification by utilizing modal quality weights, anatomical consistency features, cross-modal spatial transformation, and temporal compensation mechanisms when there are inconsistencies in multimodal image quality, sampling frequencies, and cardiac phases. This improves the spatiotemporal consistency of multimodal cardiac image fusion and provides a more stable data foundation for subsequent dynamic motion boundary extraction, individualized dynamic boundary condition inversion, and real-time hemodynamic simulation.

[0034] In some embodiments, based on a unified cardiac cycle index, anatomical structure identification, motion boundary extraction, and blood flow observation mapping are performed on multimodal cardiac imaging data to generate a cardiac structure motion fusion representation with modal credibility, including: Cardiac structural features at each cardiac phase are extracted using a unified cardiac cycle index, and cardiac anatomical boundary representations are determined based on cross-modal consistency constraints. Based on the structural displacement relationship between consecutive cardiac phases, a dynamic motion boundary corresponding to the cardiac anatomical boundary representation is generated. The blood flow observation information in multimodal cardiac imaging data is mapped to dynamic motion boundaries and corresponding anatomical regions. Modal credibility is determined based on modal quality weights and mapping residuals, and a motion fusion representation of cardiac structure is generated.

[0035] Specifically, after generating a unified cardiac cycle index, the system retrieves corresponding cardiac computed tomography (CT) data, cardiac magnetic resonance cinnabar (MRI) sequences, transthoracic echocardiography (TECG) sequences, and four-dimensional phase-contrast magnetic resonance velocity field data phase-by-phase according to the standard cardiac phase axes from 0% to 100%. Taking the left ventricle, mitral valve, and aortic outflow tract regions of the target object as examples, the system extracts the end-diastolic endocardial boundary, valve annulus position, and outflow tract centerline at the 0% phase, and extracts the end-systolic endocardial boundary, valve opening and closing status, and ventricular wall thickness changes at the 40% phase, loading the structural features of each phase into the same cardiac coordinate system. For the same anatomical boundary obtained from different modalities, the system sets cross-modal consistency constraints, such as left ventricular long axis direction error, valve annulus plane angle error, average distance error of cardiac chamber boundaries, and topological connectivity consistency, all of which are used as boundary fusion evaluation factors. When cardiac computed tomography data provides clear boundaries and magnetic resonance cine sequence has continuous motion information in the corresponding phase, the system enhances the contribution of both to the characterization of cardiac anatomical boundaries; when the ultrasound section has local boundary loss due to acoustic window obstruction, it is only used as an auxiliary constraint on valve activity.

[0036] After defining the anatomical boundaries of the heart, the system generates dynamic motion boundaries based on the structural displacement relationships between consecutive cardiac phases. Specifically, a correspondence can be established between adjacent phases for endocardial boundary points, valve annulus feature points, and outflow tract centerline sampling points, and the boundary displacement trajectory can be calculated using a non-rigid deformation field. For example, within the 0% to 40% phase segment, the left endocardial boundary contracts towards the center of the heart chamber, with a displacement amplitude of approximately 6 mm in the apex region and approximately 4 mm along the long axis of the basal segment valve annulus plane. The system records this displacement trajectory as boundary motion constraints in phase order. For abnormal displacement points caused by missing images, the system combines the deformation trend of adjacent phases, ventricular volume curves, and motion smoothing constraints for correction, ensuring that the dynamic motion boundary remains continuous throughout the complete cardiac cycle.

[0037] Furthermore, the system maps blood flow observation information from multimodal cardiac imaging data to dynamic motion boundaries and corresponding anatomical regions. For color Doppler velocity images, the system maps the flow direction, peak velocity, and jet width near the mitral and aortic valve orifices to the corresponding valve regions based on the spatial position of the ultrasound section in the three-dimensional cardiac coordinate system. For four-dimensional phase-contrast magnetic resonance velocity fields, the system projects three-dimensional velocity sampling points to the left ventricular inflow tract, left ventricular cavity, and aortic outflow tract regions, and correlates them with the dynamic motion boundaries in the same cardiac phase.

[0038] Furthermore, after mapping, the system calculates modal reliability based on modal quality weights and mapping residuals. For example, if the difference between the four-dimensional phase-contrast magnetic resonance velocity field and the Doppler peak velocity is small at a certain phase, and the spatial projection residual is less than 2 mm, the modal reliability of the corresponding blood flow observation information is set to high; if there is significant aliasing or sectional shift in the ultrasound flow velocity image, the modal reliability of the corresponding region is reduced.

[0039] Finally, the system fuses cardiac anatomical boundary representations, dynamic motion boundaries, blood flow observation information, and modal reliability to generate a fused representation of cardiac structural motion. This fused representation not only records the structural boundaries and motion trajectories of each anatomical region under different cardiac phases, but also records the corresponding blood flow observation sources, observation intensities, and reliability status.

[0040] Through the above processing, the correspondence between cardiac structure, wall motion, and blood flow observation can be uniformly expressed within the same cardiac cycle framework. This provides highly consistent input for subsequent construction of cardiac anatomical topology maps, individualized dynamic boundary condition inversion, and real-time hemodynamic simulation, thereby improving the accuracy of dynamic motion boundary extraction and the reliability of blood flow observation mapping.

[0041] In some embodiments, a dynamic motion boundary corresponding to the cardiac anatomical boundary representation is generated based on the structural displacement relationship between consecutive cardiac phases, including: Based on the unified cardiac cycle index, the correspondence between the same anatomical boundary regions under continuous cardiac phases is determined, and the cardiac anatomical boundary representation is continuously registered in time to generate the boundary deformation field. The displacement trajectory of the anatomical boundary region during the cardiac cycle is extracted based on the boundary deformation field, and the displacement trajectory is corrected by combining motion smoothing constraints and topological consistency constraints. The corrected displacement trajectory is converted into boundary motion constraints in the simulation computation domain, generating a dynamic motion boundary corresponding to the cardiac anatomical boundary representation.

[0042] Specifically, after generating the cardiac anatomical boundary representation, the system can perform temporal continuity processing of the anatomical boundaries under each standard cardiac phase according to the unified cardiac cycle index. Taking the left ventricle, mitral valve annulus, and aortic outflow tract region of the target object as examples, the unified cardiac cycle index divides a complete cardiac cycle into phase nodes such as 0%, 10%, 20%, 30%, 40%, 60%, 80%, and 100%. Each phase node is associated with the corresponding endocardial boundary, valve annulus contour, outflow tract centerline, and registration confidence.

[0043] The system first determines the correspondence between identical anatomical boundary regions in adjacent phases. For example, it matches the apical, middle, and basal segments of the left ventricular endocardial boundary at 0% phase with their corresponding regions at 10% phase. It also binds the anterior leaflet, posterior leaflet, and commissural region markers on the mitral valve annulus to their corresponding locations. This correspondence can be determined based on anatomical markers, boundary curvature, central axis projection distance, and local surface normal vectors, avoiding regional mismatches during myocardial contraction caused by relying solely on pixel distance.

[0044] After determining the correspondence between consecutive cardiac phases, the system performs temporal continuous registration of the cardiac anatomical boundary representation to generate a boundary deformation field. Specifically, the cardiac anatomical boundary of the previous phase can be used as the source boundary, and the cardiac anatomical boundary of the subsequent phase as the target boundary. A non-rigid registration model with myocardial motion constraints is used to calculate the three-dimensional displacement vector. For example, in the 0% to 40% phase segment, the left ventricle gradually transitions from diastole to systole, the endocardial boundary moves towards the center of the heart chamber, the apex region displaces by approximately 5.8 mm, the mid-section circumferential systolic displacement is approximately 4.6 mm, and the basal segment displaces along the long axis by approximately 3.9 mm; in the 40% to 100% phase segment, the left ventricle gradually relaxes, and the endocardial boundary expands outward. The system organizes the three-dimensional displacement vectors corresponding to each region into a boundary deformation field according to the cardiac phase sequence, and associates each displacement vector with the source mode, registration confidence, and boundary region identifier.

[0045] Furthermore, the system extracts the displacement trajectory of the anatomical boundary region during the cardiac cycle based on the boundary deformation field. For the left ventricular endocardial boundary, radial displacement trajectory, axial displacement trajectory, and circumferential contraction trajectory can be generated for the apical, intermediate, and basal segments, respectively; for the valve annulus region, annular plane movement trajectory and valve orifice area change trajectory can be generated; for the aortic outflow tract region, outflow tract wall expansion trajectory and centerline offset trajectory can be generated. When a phase experiences abrupt shifts in displacement points due to low image quality, the system performs corrections by combining motion smoothing constraints and topological consistency constraints.

[0046] Motion smoothing constraints are used to limit abnormal changes in displacement velocity and acceleration between adjacent phases, while topological consistency constraints are used to maintain the connectivity between the cardiac chamber boundaries, valve annulus boundaries, and outflow tract boundaries. For example, if a certain ultrasound section at 30% phase causes an abnormal outward shift of the basal segment boundary, the system can perform regression correction on the abnormal displacement based on the continuous boundaries of the magnetic resonance cine sequence, the displacement trend of adjacent phases, and the ventricular volume change curve, so that the basal segment boundary corresponding to that phase returns to a reasonable range of motion.

[0047] After completing the displacement trajectory correction, the system converts the corrected displacement trajectory into boundary motion constraints of the simulation computational domain. During the conversion, the system maps surface points or mesh nodes in the cardiac anatomical boundary representation to boundary nodes in the blood flow simulation computational domain, and writes the displacement, velocity, and normal motion direction corresponding to each standard cardiac phase into the boundary nodes. For the left ventricular intracavitary blood flow simulation, the endocardial boundary nodes are set as moving wall boundaries that change with the cardiac phase; for the regions near the mitral and aortic valves, the valve annulus motion trajectory and the valve orifice opening and closing state together form local opening change constraints; for the aortic outflow tract, the wall motion trajectory is used to limit the change of the outflow tract cross-section over time. Thus, the system generates a dynamic motion boundary corresponding to the cardiac anatomical boundary representation and provides this dynamic motion boundary to the subsequent individualized dynamic boundary condition inversion and hybrid hemodynamic solver.

[0048] Through the above processing, the static anatomical boundary under continuous cardiac phase can be transformed into a dynamic motion boundary with temporal continuity, smooth motion, and topological consistency. This enables the blood flow simulation computational domain to realistically reflect the motion changes of the myocardial wall, valve annulus region, and outflow tract structure during the cardiac cycle, thereby improving the accuracy of dynamic boundary construction and the stability of subsequent real-time hemodynamic simulation.

[0049] In some embodiments, constructing a cardiac anatomical topology map based on cardiac structural motion fusion representation includes: Anatomical functional zoning analysis was performed on the fusion representation of cardiac structure and motion to identify the structural nodes corresponding to the cardiac blood flow pathway. Topological relationship edges are generated based on the spatial connectivity and blood flow transmission relationships between each structural node, and the dynamic motion boundary and blood flow observation information are associated with the corresponding structural node or topological relationship edge. Based on modal credibility, consistency checks are performed on the associated structural nodes and topological relationship edges to generate a cardiac anatomical topology map for boundary condition inversion.

[0050] Specifically, after the cardiac structure motion fusion representation is generated, the system performs anatomical functional partitioning analysis on the fusion representation. Taking the real-time simulation of blood flow in the left ventricular system of the target object as an example, the system reads the cardiac anatomical boundaries, dynamic motion boundaries, blood flow observation information and modal reliability at 0% to 100% phase according to the unified cardiac cycle index, and divides the left atrium, mitral valve orifice, left ventricle, aortic valve orifice, aortic root, left ventricular inflow tract and left ventricular outflow tract into different anatomical functional regions.

[0051] For each anatomical functional region, the system extracts the region's centerline, boundary contour, flow inlet, flow outlet, and phase motion attributes to generate structural nodes corresponding to the cardiac blood flow pathway. For example, the left atrial node represents the source of diastolic inflow, the mitral valve orifice node represents the opening constraint from the left atrium to the left ventricle, the left ventricle node represents the main blood flow calculation chamber, the aortic valve orifice node represents the systolic jet outlet, and the aortic root node represents the outflow tract continuation boundary.

[0052] After determining the structural nodes, the system generates topological relationship edges based on the spatial connectivity and blood flow transmission relationships between each structural node. Spatial connectivity can be determined by the boundary adjacency of adjacent anatomical functional areas, the connection relationship of the central axis, and the position of the valve annulus plane. Blood flow transmission relationships can be determined by the direction of flow velocity under cardiac phase, the opening and closing state of the valve, and the trend of volume change.

[0053] For example, during diastole, the system generates inflow topology edges between the left atrial node and the mitral valve orifice node, and diastolic filling edges between the mitral valve orifice node and the left ventricular node; during systole, the system generates jet topology edges between the left ventricular node and the aortic valve orifice node, and outflow transmission edges between the aortic valve orifice node and the aortic root node. For the valve closure phase, the system marks the corresponding topology edges as restricted transmission states, ensuring that subsequent boundary condition inversion does not incorrectly treat a closed valve as an open channel.

[0054] Furthermore, the system associates dynamic motion boundaries and blood flow observation information with corresponding structural nodes or topological relationships. For the left ventricular node, the system records the displacement trajectory of the endocardial boundary as a function of phase, wall velocity, and changes in cavity volume; for the mitral valve orifice and aortic valve orifice nodes, the system records changes in valve orifice area, valve annulus motion trajectory, and opening / closing status; for inflow and outflow topological relationships, the system records Doppler flow velocity, four-dimensional phase-contrast magnetic resonance velocity field projection results, and flow direction consistency information.

[0055] For example, exemplarily, in the 30% to 40% phase segment, the aortic valve orifice node is in an open state, and the topological relationship between the aortic valve orifice and the aortic root is associated with the peak jet velocity and jet direction; in the 70% to 90% phase segment, the mitral valve orifice node is in an open state, and the topological relationship between the left atrium and the left ventricle is associated with the diastolic inflow velocity and the valve orifice opening area.

[0056] Furthermore, after completing the association, the system performs consistency verification on structural nodes and topological relationship edges based on modal credibility. For the same structural node, the system compares whether the boundary positions, motion trajectories, and phase events provided by different modalities are consistent; for the same topological relationship edge, the system compares whether the blood flow direction, peak velocity, and valve opening and closing state conform to the anatomical connectivity rules. When the spatial deviation between the magnetic resonance motion boundary and the computed tomography boundary of the left ventricular node is small, and the ultrasound valve event is consistent with the cardiac phase index, the system increases the credibility state of that node; when the Doppler velocity direction of a certain topological relationship edge is inconsistent with the four-dimensional phase-contrast magnetic resonance velocity direction, the system reduces the constraint strength of that edge according to modal credibility and mapping residuals, and marks that edge as a relationship edge to be corrected. After consistency verification, the system generates a cardiac anatomical topology map for boundary condition inversion, in which each structural node and topological relationship edge carries anatomical attributes, motion attributes, blood flow attributes, and credibility state.

[0057] Through the above embodiments, the system can further transform the cardiac structural motion fusion representation into a cardiac anatomical topology map with blood flow path meaning, so that dynamic motion boundaries and blood flow observation information can be accurately bound to the corresponding structural nodes and topological relationship edges, providing a clear graph structure basis for parameter inversion of inlet state, outlet state, valve state and wall motion state, thereby improving the accuracy of boundary condition inversion and the stability of real-time blood flow simulation.

[0058] In some embodiments, parameter inversion is performed on the inlet state, outlet state, valve state, and wall motion state based on the cardiac anatomical topology map to generate an individualized set of dynamic boundary conditions, including: Based on the cardiac anatomical topology map, blood flow boundary nodes and associated topological relationship edges are determined, and dynamic motion boundaries, blood flow observation information, and modal confidence are written into the corresponding blood flow boundary nodes; Based on blood flow conservation constraints and cardiac cycle continuity constraints, a boundary parameter inversion model is constructed with the goal of minimizing observation residuals; By using the boundary parameter inversion model, the blood flow state parameters and motion constraint parameters of the blood flow boundary nodes are jointly estimated, generating an individualized set of dynamic boundary conditions with parameter confidence.

[0059] Specifically, after the cardiac anatomical topology map is generated, the system determines the blood flow boundary nodes that need to participate in the boundary condition inversion based on the structural nodes, topological relationship edges, and phase attributes in the map. Taking the target object, the left ventricular system, as an example, the system determines the left atrial inlet region, mitral valve orifice region, left ventricular wall region, aortic valve orifice region, and aortic root outlet region as blood flow boundary nodes, and determines the topological relationship edges between the left atrium and mitral valve orifice, between the mitral valve orifice and the left ventricle, between the left ventricle and the aortic valve orifice, and between the aortic valve orifice and the aortic root as associated topological relationship edges.

[0060] For each blood flow boundary node, the system writes the corresponding dynamic motion boundary, blood flow observation information, and modal reliability. For example, the left ventricular wall node writes the displacement trajectory of the endocardial boundary within 0% to 100% of the cardiac phase, the wall normal velocity, and the cavity volume change; the mitral valve orifice node writes the valve annulus motion trajectory, valve orifice opening area, and diastolic inflow velocity; the aortic valve orifice node writes the valve orifice opening and closing status, systolic jet velocity, and outflow direction; and the aortic root outlet node writes the outflow tract cross-sectional changes and velocity field projection results.

[0061] After the aforementioned data is written, the system constructs a boundary parameter inversion model based on blood flow conservation constraints and cardiac cycle continuity constraints. Blood flow conservation constraints are used to ensure consistency between the inflow and outflow of blood into the heart chambers and changes in chamber volume, avoiding contradictions between the inverted inflow and outflow rates and ventricular volume changes. Cardiac cycle continuity constraints are used to limit the magnitude of boundary parameter changes between adjacent cardiac phases, ensuring smooth changes in valve resistance, inflow velocity, outflow pressure, and wall motion constraints along the phase axis. The system can also introduce an observation residual minimization objective, using Doppler velocity, four-dimensional phase-contrast magnetic resonance velocity field, ventricular volume curve, and dynamic motion boundaries as observation constraints, allowing the inversion model to find the optimal combination of boundary parameters among multiple modal evidences.

[0062] Furthermore, the system utilizes a boundary parameter inversion model to jointly estimate the blood flow state parameters and motion constraint parameters of the blood flow boundary nodes. The blood flow state parameters may include the inlet flow rate curve, the outlet equivalent resistance, the valve orifice local resistance, the transvalvular flow rate, and the local velocity distribution. The motion constraint parameters may include the wall displacement amplitude, the wall motion velocity, the valve orifice area change curve, and the outflow tract cross-sectional change parameters.

[0063] For example, during the diastolic phase from 30% to 70%, the system estimates the mitral valve inflow and equivalent resistance based on the mitral valve orifice Doppler flow velocity and the left ventricular volume growth curve. During the systolic phase from 15% to 45%, the system estimates the aortic valve outlet flow, local resistance, and aortic root outlet status based on the aortic valve jet velocity, aortic root velocity field, and left ventricular volume decrease curve. If the confidence level of a certain mode is low, for example, due to aliasing of ultrasound flow velocities, the system reduces the weight of that observation in the inversion target and prioritizes the use of high-confidence magnetic resonance velocity field and ventricular volume change information.

[0064] During the inversion solution process, the system can update the parameter values ​​of each blood flow boundary node using an iterative optimization method. After each iteration, the system substitutes the current boundary parameters into a reduced-order blood flow propagation model or a fast physical constraint solution model to obtain the predicted flow velocity and predicted volume change, and then compares them with blood flow observation information to calculate the observation residuals. When the observation residuals gradually converge and the parameter change amplitude is less than a preset threshold, the system outputs the boundary parameter estimation results; when the local residuals are consistently high, the system marks the corresponding blood flow boundary node as a low-confidence node and restricts the node parameters from directly participating in subsequent real-time simulations. Finally, the system generates parameter confidence based on the iterative convergence state, observation residuals, modal confidence, and node topology consistency, and organizes the inlet state, outlet state, valve state, and wall motion state into an individualized dynamic boundary condition set.

[0065] Through the above embodiments, the system can transform dynamic motion boundaries and blood flow observation information into calculable and calibrable individualized dynamic boundary conditions based on the cardiac anatomical topology map. This enables the joint inversion of inlet state, outlet state, valve state, and wall motion state under the constraints of blood flow conservation and continuous cardiac cycle, thereby improving the accuracy of boundary condition determination and the reliability of subsequent real-time hemodynamic simulation.

[0066] In some embodiments, a boundary parameter inversion model is used to jointly estimate the blood flow state parameters and motion constraint parameters of the blood flow boundary nodes, generating an individualized set of dynamic boundary conditions with parameter confidence, including: The initial constraint range of the parameters to be inverted is determined based on the associated topological relationships and modal confidence of the blood flow boundary nodes. The blood flow observation information, dynamic motion boundary, and physical constraint are input into the boundary parameter inversion model, and the blood flow state parameters and motion constraint parameters are jointly iteratively solved. Based on the iterative convergence state and observation residuals, the parameter confidence level is determined, and an individualized dynamic boundary condition set with parameter confidence level is generated.

[0067] Specifically, after constructing the cardiac anatomical topology map, the system establishes a list of parameters to be inverted for each blood flow boundary node. Taking real-time simulation of left ventricular blood flow as an example, the cardiac anatomical topology map may include left atrial inlet nodes, mitral valve orifice nodes, left ventricular wall nodes, aortic valve orifice nodes, and aortic root outlet nodes. The system first reads the associated topological relationship edges of each node to determine the direction of blood flow transmission, open state, and constraint strength during the cardiac cycle.

[0068] For example, the mitral valve orifice node is connected to the left atrial inlet node and the left ventricular wall node during diastole, and the parameters to be inverted include diastolic inlet flow, valve orifice equivalent resistance, and valve orifice area change parameters; the aortic valve orifice node is connected to the left ventricular wall node and the aortic root outlet node during systole, and the parameters to be inverted include systolic outlet flow, transvalvular pressure gradient equivalent parameters, and local jet velocity distribution parameters.

[0069] The system determines the initial constraint range by combining the modal reliability of each node. When the reliability of Doppler velocity and four-dimensional phase-contrast magnetic resonance velocity field is high, the initial flow range is narrowed. When there is occlusion or velocity aliasing in the local ultrasound image, the initial constraint range is appropriately widened and the constraint weight of the observation on the inversion result is reduced.

[0070] After determining the initial constraint range, the system inputs blood flow observation information, dynamic motion boundaries, and physical constraints into the boundary parameter inversion model. Blood flow observation information may include peak valve velocity, local velocity direction, ventricular volume change curve, and velocity field projection results; dynamic motion boundaries may include left ventricular endocardial boundary displacement, wall motion velocity, valve annulus movement trajectory, and outflow tract cross-sectional changes; physical constraints may include mass conservation constraints, cardiac cycle continuity constraints, boundary motion continuity constraints, and topological connectivity constraints.

[0071] For example, exemplarily, in the 30% to 45% cardiac phase segment, the system calculates the outflow observation residual based on the aortic valve jet velocity and the decrease in left ventricular volume, and corrects the outlet resistance parameters based on the aortic root velocity field projection results; in the 60% to 90% cardiac phase segment, the system corrects the inlet flow curve and the equivalent mitral valve resistance based on the diastolic flow velocity and the increase in left ventricular volume. In each iteration, the current blood flow state parameters and motion constraint parameters are substituted into the rapid order reduction solution model to obtain the predicted velocity, predicted flow rate, and predicted volume change, which are then compared with the actual observation data to form the iterative residual.

[0072] During the joint iterative solution process, the system synchronously updates blood flow state parameters and motion constraint parameters, rather than estimating the inlet, outlet, or wall boundaries separately. For example, when the left ventricular wall contraction amplitude increases but the aortic valve orifice velocity observation does not increase synchronously, the system does not directly increase the outlet flow rate. Instead, it simultaneously adjusts the local resistance of the aortic valve orifice, the aortic root outlet state, and the wall motion constraint parameters to ensure that the inversion results meet the requirements of blood flow conservation and phase continuity. When the observation residual of a certain blood flow boundary node exceeds a threshold in multiple consecutive iterations, the system marks the node as a low-confidence node and reduces the propagation influence of the node parameters on adjacent topological relationships. When the residual decreases steadily with iteration and the changes in adjacent phase parameters are smooth, the system marks the corresponding parameters as high-confidence parameters.

[0073] Finally, the system determines the parameter confidence level based on the iterative convergence state and the observed residuals. The parameter confidence level can be jointly determined by the residual convergence amplitude, the final residual size, modal confidence, and topological consistency. For example, when the aortic valve outlet flow rate residual is below a preset threshold after 8 iterations, and the velocity direction is consistent with the four-dimensional phase-contrast magnetic resonance velocity field, the system assigns a higher parameter confidence level; when the mitral valve local resistance causes large residual fluctuations due to ultrasonic flow velocity aliasing, the system assigns a lower parameter confidence level. The system organizes the inlet flow rate curve, outlet resistance parameter, valve state parameter, and wall motion constraint parameter with parameter confidence levels to generate an individualized dynamic boundary condition set.

[0074] Through the above embodiments, the system can determine a reasonable inversion range by utilizing the topological relationship of blood flow boundary nodes and multimodal reliable information, and jointly estimate blood flow state parameters and motion constraint parameters under the combined effect of blood flow observation, dynamic motion boundary and physical constraints, thereby improving the accuracy of individualized dynamic boundary condition determination and reducing the impact of single modal error on real-time hemodynamic simulation.

[0075] In some embodiments, a digital twin cardiac hemodynamic model is constructed based on a cardiac anatomical topology map and a personalized set of dynamic boundary conditions, including: A computational domain corresponding to the cardiac blood flow path is established based on the cardiac anatomical topology, and an individualized set of dynamic boundary conditions is loaded onto the boundary nodes of the computational domain. Based on the computational domain, boundary nodes, and topological relationships, a multi-layer model structure containing geometric state, motion state, and blood flow state is constructed. Modal confidence and parameter confidence are written into the correction layer of the multi-layer model structure to generate a digital twin cardiac hemodynamic model for real-time simulation.

[0076] Specifically, after generating the individualized dynamic boundary condition set, the system establishes a simulation computational domain corresponding to the cardiac blood flow path based on the cardiac anatomical topology map. Taking the real-time simulation of the target object, the left ventricular system, as an example, the system determines the computational domain consisting of the left atrial inflow region, mitral valve passage, left ventricular intracavitary flow region, aortic valve jet region, and aortic root outflow region based on the topological relationships between the left atrial inlet node, mitral valve orifice node, left ventricular cavity node, aortic valve orifice node, and aortic root outlet node in the cardiac anatomical topology map.

[0077] The computational domain can be represented using voxelized meshes, surface meshes, and graph-structured nodes. Voxelized meshes describe the blood flow field within the cardiac chambers, surface meshes describe the endocardial boundaries and valve orifice regions, and graph-structured nodes describe the blood flow transmission paths and the locations where boundary parameters are applied. The system loads individualized dynamic boundary condition sets onto the boundary nodes of the computational domain. For example, the diastolic inlet flow curve is loaded onto the mitral valve inlet boundary node, the systolic outlet resistance parameter is loaded onto the aortic root outlet boundary node, the valve orifice area change curve is loaded onto the valve boundary node, and wall displacement and wall velocity are loaded onto the left ventricular endocardial moving wall node.

[0078] After the boundary conditions are loaded, the system constructs a multi-layer model structure based on the computational domain, boundary nodes, and their topological relationships. This multi-layer model structure can include a geometric state layer, a motion state layer, and a blood flow state layer. The geometric state layer records the cardiac chamber contours, valve annulus positions, outflow tract centerlines, and computational domain grid relationships at each cardiac phase, used to determine the blood flow solution range. The motion state layer records the wall displacement trajectory, valve opening and closing states, valve annulus motion trajectory, and outflow tract cross-sectional changes, used to drive the computational domain to change with the cardiac cycle. The blood flow state layer records inlet flow rate, outlet resistance, local velocity, pressure transmission relationships, and nodal flow conservation relationships, used to provide computable input for the hybrid hemodynamics solver.

[0079] For example, exemplarily, in the 30% to 45% cardiac phase segment, the motion state layer indicates that the aortic valve orifice is open, while the blood flow state layer simultaneously loads the aortic valve orifice outlet flow and aortic root resistance parameters; in the 60% to 90% cardiac phase segment, the motion state layer indicates that the mitral valve orifice is open, while the blood flow state layer loads the mitral valve orifice inlet flow and left ventricular wall diastolic motion constraint.

[0080] Furthermore, the system incorporates modal confidence and parameter confidence into the calibration layer of the multi-layer model structure. The calibration layer records the contribution intensity of different modal data to the geometric, motion, and blood flow states, and records the confidence status of each boundary parameter in the individualized dynamic boundary condition set. If the confidence of ultrasound valve orifice observation is low in a certain phase, the calibration layer reduces the update weight of valve orifice motion constraints in that phase; if the residual of the aortic valve outlet flow parameter is small after multiple rounds of inversion, the calibration layer marks this parameter as a high-confidence parameter and allows it to participate in subsequent real-time simulations.

[0081] In some examples, for low-confidence boundary nodes, the system can set residual monitoring flags in the model, enabling the hybrid hemodynamics solver to prioritize adjacent high-confidence nodes and physical constraints to correct the corresponding regions during computation. Ultimately, the geometric state layer, motion state layer, blood flow state layer, and correction layer together form a digital twin cardiac hemodynamic model for real-time simulation.

[0082] Through the above embodiments, the cardiac anatomical topology map and individualized dynamic boundary condition set can be transformed into a digital twin cardiac hemodynamic model with computational domain, boundary nodes, topological relationships and reliable correction capabilities. This enables the model to simultaneously express cardiac structural morphology, cardiac cycle motion and blood flow state changes, providing a stable foundation for subsequent hybrid hemodynamic solutions and simulation bias closed-loop correction, thereby improving the accuracy of real-time simulation model construction and operational reliability.

[0083] In some embodiments, a digital twin cardiac hemodynamic model is input into a hybrid hemodynamic solver, and the blood flow state field is co-evolved using local high-fidelity solving, graph-structured order reduction solving, and physically constrained neural operators, including: Based on the digital twin cardiac hemodynamics model, the global blood flow path and local fine calculation regions are divided to generate a hierarchical solution task; The global blood flow state is rapidly propagated by using graph structure-based order reduction solution, and the blood flow state of the corresponding local fine calculation region is corrected by calling local high-fidelity solution when the local residual meets the triggering condition; The hierarchical solution results and the set of individualized dynamic boundary conditions are input into the physical constraint neural operator, and the field is completed based on the physical residual constraints to generate the blood flow state field.

[0084] Specifically, after the digital twin cardiac hemodynamic model is constructed, the system inputs the model into a hybrid hemodynamic solver and generates layered solution tasks according to the blood flow paths and local complex flow regions in the cardiac anatomical topology diagram. Taking the real-time simulation of the target object from the left ventricle to the aortic root as an example, the system divides the main blood flow path formed by the left atrial inlet, mitral valve orifice, left ventricular cavity, aortic valve orifice, and aortic root outlet into global blood flow paths, and divides the diastolic jet region near the mitral valve orifice, the vortex region in the left ventricular cavity, the high-speed jet region at the aortic valve orifice, and the aortic root refluxing region into local fine calculation regions.

[0085] The global blood flow pathway is used to rapidly propagate flow, pressure, and resistance states, while the local fine-grained computation region is used to handle areas with large velocity gradients, rapid boundary changes, and complex flow directions. The system establishes a solution task queue for each phase based on a unified cardiac cycle index. For example, the aortic valve jet is mainly calculated in the 20% to 45% phase segment, and the mitral valve inflow is mainly calculated in the 60% to 90% phase segment.

[0086] During the global solution phase, the system utilizes graph-based order reduction to rapidly propagate the global blood flow state. Specifically, the left atrial inlet node, mitral valve orifice node, left ventricular node, aortic valve orifice node, and aortic root outlet node are used as nodes in the reduced-order blood flow graph. The inflow, outflow, and valve constraint relationships between nodes are treated as edge states, and the inflow flow, outlet resistance, equivalent valve resistance, and wall motion parameters from the individualized dynamic boundary condition set are loaded onto the corresponding nodes or edges. The system iteratively propagates the node pressure, node flow, and local resistance states in phase order, and uses blood flow conservation constraints to ensure consistency between the left ventricular volume change and the inflow-outflow difference. If the local residual between the global order reduction solution result and the blood flow observation information does not exceed the trigger threshold, the system directly uses the reduced-order result as the preliminary solution result for that region.

[0087] When the local residuals meet the triggering conditions, the system invokes a local high-fidelity solution to correct the blood flow state in the corresponding locally refined computational region. For example, at 35% phase, if the residual between the reduced-order jet velocity from the aortic valve orifice to the aortic root and the projection result of the four-dimensional phase-contrast magnetic resonance velocity field exceeds a preset threshold, the system uses the high-velocity jet region at the aortic valve orifice as the trigger region and constructs a locally refined computational grid based on the dynamic motion boundary, valve area, and outlet state of this region. The local high-fidelity solution can solve for the velocity distribution, pressure gradient, and wall shear stress within a relatively small computational domain, and then writes the corrected local velocity field and pressure boundary back to the reduced-order blood flow map. For the vortex region within the left ventricle, the system can invoke a local high-fidelity solution when the vortex intensity residual is high to correct for swirling and stagnant states that are difficult to express by the reduced-order model.

[0088] Furthermore, the system inputs the hierarchical solution results and individualized dynamic boundary condition sets into the physical constraint neural operator for full-field completion. The physical constraint neural operator receives the computational domain geometry, dynamic motion boundaries, nodal flow states, local high-fidelity correction results, and phase identifiers, and outputs the velocity field, pressure field, wall shear stress field, and vortex characterization corresponding to the current cardiac phase. To avoid deviations from physical laws in the field completion, the system introduces mass conservation residuals, momentum conservation residuals, boundary condition residuals, and cardiac cycle continuity residuals during the neural operator derivation process. When the neural operator output in a certain local region is inconsistent with the boundary conditions, the system uses high-fidelity correction results and reduced-order propagation results to perform secondary correction on that region until the residuals meet the requirements of real-time simulation. Finally, the system generates a blood flow state field covering the global blood flow path and local complex flow regions.

[0089] Through the above embodiments, the system can synergistically combine global rapid propagation, local fine correction, and physical constraint field completion, which avoids the excessive computational load of traditional global high-fidelity solutions and overcomes the problem that a single reduced-order model is difficult to express complex local flows, thereby improving the efficiency of blood flow state field generation, local simulation accuracy, and real-time simulation stability.

[0090] In some embodiments, a graph-structured reduced-order solution is used to rapidly propagate the global blood flow state, and when the local residuals meet the triggering conditions, a local high-fidelity solution is invoked to correct the blood flow state of the corresponding local fine-computation region, including: The computational domain in the digital twin cardiac hemodynamics model is mapped to a reduced-order blood flow graph, and the node and edge states of the reduced-order blood flow graph are initialized according to the individualized dynamic boundary condition set. The global blood flow state is iteratively propagated based on the topological relationships and physical constraints of the reduced-order blood flow graph to generate a global reduced-order solution result; The trigger region is determined based on the local residual between the global reduced-order solution and the blood flow observation information. The local high-fidelity solution is then called to refine the trigger region, and the correction result is written back to the reduced-order blood flow map.

[0091] Specifically, in the process of solving the hybrid hemodynamics problem, the system first converts the computational domain in the digital twin cardiac hemodynamics model into a reduced-order blood flow graph. Taking the real-time simulation from the left ventricle to the aortic root as an example, the system maps the left atrial inlet, mitral valve orifice, left ventricular cavity, aortic valve orifice, and aortic root outlet as nodes in the reduced-order blood flow graph, and maps the blood flow paths from the left atrium to the mitral valve orifice, from the mitral valve orifice to the left ventricle, from the left ventricle to the aortic valve orifice, and from the aortic valve orifice to the aortic root as edges. Each node corresponds to an updatable blood flow state unit, and each edge corresponds to a blood flow transmission unit.

[0092] The system initializes the node and edge states of the reduced-order blood flow map based on an individualized set of dynamic boundary conditions. For example, the mitral valve orifice node is initialized with diastolic inlet flow, valve opening area, and valve equivalent resistance; the left ventricular cavity node is initialized with phase volume, wall motion velocity, and initial intracavitary pressure; the aortic valve orifice node is initialized with systolic outlet flow, transvalvular resistance, and jet direction; and the aortic root outlet node is initialized with outlet resistance and compliance parameters.

[0093] After initialization, the system iteratively propagates the global blood flow state based on the topological relationships and physical constraints of the reduced-order blood flow graph. Specifically, it calculates node flow, node pressure, and edge resistance states phase by phase according to a unified cardiac cycle index, and propagates the flow changes of upstream nodes and the pressure feedback of downstream nodes in each iteration based on the blood flow path direction. Taking the 30% to 45% phase segment as an example, the aortic valve orifice is in an open state, and the system propagates systolic flow from the left ventricular cavity node to the aortic valve orifice node, and then propagates the outlet pressure and velocity states from the aortic valve orifice node to the aortic root outlet node; taking the 60% to 90% phase segment as an example, the mitral valve orifice is in an open state, and the system propagates filling flow from the left atrial inlet node to the mitral valve orifice node, and then propagates the diastolic volume increment to the left ventricular cavity node. During the iterative propagation process, the system introduces mass conservation constraints, pressure continuity constraints, valve unidirectional transmission constraints, and cardiac cycle continuity constraints to keep the inflow, outflow, and volume changes of each node consistent, generating a global reduced-order solution covering the entire blood flow path.

[0094] Furthermore, the system compares the global reduced-order solution results with the blood flow observation information using local residuals to determine whether a local high-fidelity solution should be triggered. Blood flow observation information can come from Doppler velocity images, four-dimensional phase-contrast magnetic resonance velocity fields, and preceding cardiac observation data. For example, at 35% phase, the reduced-order blood flow map calculates an aortic valve orifice jet velocity of 1.2 m / s, while the four-dimensional phase-contrast magnetic resonance velocity field maps the velocity in the same region to 1.7 m / s. Since both the velocity direction residual and the flow residual exceed a preset trigger threshold, the system identifies the local region from the aortic valve orifice to the aortic root as the trigger region. Similarly, at 75% phase, the residual between the mitral valve orifice inflow velocity and the Doppler observation is large, so the system identifies the region adjacent to the mitral valve orifice as the trigger region.

[0095] After determining the trigger region, the system uses a local high-fidelity solver to refine the region. For the aortic valve trigger region, the system constructs a local fine-grained computational grid based on the valve area, dynamic motion boundary, outlet resistance, and velocity observations at the corresponding phase, and calculates the local velocity distribution, pressure gradient, and wall shear stress. For the mitral valve trigger region, the system calculates the local inflow jet and vortex state based on the valve annulus trajectory, diastolic inlet flow, and left ventricular cavity volume changes. After the local high-fidelity solver results are generated, the system writes the corrected local velocity, pressure, and equivalent resistance back to the corresponding nodes and edges in the reduced-order flow graph, and updates the node state, edge state, and local residual identifiers, so that subsequent phase propagation can inherit the correction results.

[0096] Through the above embodiments, the system can quickly calculate the global blood flow state using reduced-order blood flow maps, and perform high-fidelity correction only on the triggering region when local residuals are abnormal, avoiding the computational burden caused by fine-grained global solutions, while improving the simulation accuracy of complex flow regions, thereby enhancing the efficiency, stability and local reliability of real-time hemodynamic simulation.

[0097] In some embodiments, based on the simulation deviation between the current cardiac observation data and the blood flow state field, the individualized dynamic boundary condition set, modal confidence, and solution residual correction parameters are updated, and real-time hemodynamic simulation results are output, including: The current cardiac observation data is phase-aligned with the blood flow state field according to the unified cardiac cycle index to generate a simulation bias characterization; Based on the simulation deviation characterization, identify the sources of deviation, and adjust the individualized dynamic boundary condition set, modal confidence, and solution residual correction parameters according to the sources of deviation; The blood flow state field was re-inferred based on the adjusted digital twin cardiac hemodynamic model, and real-time hemodynamic simulation results were generated by combining the deviation correction record.

[0098] Specifically, after the hybrid hemodynamics solver generates the blood flow state field of the current cardiac phase, the system acquires the current cardiac observation data generated by the target object within the same time window. The current cardiac observation data may include valve opening and closing status, color Doppler flow velocity, ventricular wall motion trajectory, ECG gating signals, and cardiac event information output by bedside acquisition equipment from real-time echocardiography.

[0099] The system maps the current cardiac observation data to the standard cardiac phase axis according to a unified cardiac cycle index and aligns it with the corresponding phase results in the blood flow state field. For example, in the current cardiac observation data, the aortic valve opening event is located at the 18% phase, the aortic valve peak jet is located at the 35% phase, and the rapid diastolic filling of the left ventricle is located at the 70% phase. The system extracts the valve orifice velocity, local pressure gradient, wall motion state, and flow distribution at the same phase in the blood flow state field, respectively, and generates phase-by-phase comparison objects.

[0100] Furthermore, after phase alignment, the system generates a simulation deviation characterization based on the difference between the current cardiac observation data and the blood flow state field. This simulation deviation characterization can include velocity deviation, flow direction deviation, wall motion deviation, valve state deviation, volume change deviation, and local residual distribution. For example, at 35% phase, real-time Doppler observation shows a peak aortic valve orifice velocity of 1.8 m / s, while the predicted velocity in the corresponding region of the blood flow state field is 1.3 m / s. The system records a velocity deviation of 0.5 m / s for this region and combines this deviation with the velocity direction difference and jet width difference to form an aortic valve orifice jet deviation characterization. At 70% phase, ultrasound observation shows the mitral valve orifice is open, but the mitral valve orifice inflow in the blood flow state field is low. The system combines this with the left ventricular volume growth curve to generate a diastolic insufficiency deviation characterization.

[0101] Furthermore, the system identifies the source of deviation based on the simulation deviation characterization. If the velocity deviation is concentrated at the inlet or outlet boundary, and the pressure propagation relationship between adjacent nodes remains stable, the system judges that the deviation mainly originates from the inlet flow rate, outlet resistance, or valve equivalent resistance in the individualized dynamic boundary condition set; if the wall motion deviation is inconsistent with the echocardiographic ventricular wall motion trajectory, the system judges that the deviation originates from the dynamic motion boundary or wall motion constraint parameters; if a certain mode corresponds to a region where observation jumps or residual anomalies continuously occur, the system judges that the deviation originates from insufficient mode reliability; if the global state is stable but the residuals in local vortices, jets, or recirculation regions are high, the system judges that the deviation originates from insufficient residual correction parameters.

[0102] Based on the source of the bias, the system adjusts the individualized dynamic boundary condition set, modal confidence, and solution residual correction parameters respectively. For example, when the aortic valve orifice jet velocity is continuously underestimated, the system increases the estimated aortic valve orifice outlet flow rate and decreases the corresponding local resistance parameter; when ultrasound section offset causes fluctuations in the orifice velocity observation, the system decreases the modal confidence of the ultrasound observation in the current phase; when there is a large difference between the local high-fidelity solution and the output of the physically constrained neural operator, the system updates the residual correction parameters for that region.

[0103] After parameter adjustment, the system re-imagines the blood flow state field based on the adjusted digital twin cardiac hemodynamics model. During the re-imagine process, the system updates the boundary node states and topological edge states of relevant phases along a unified cardiac cycle index, and rapidly propagates the global blood flow state through graph-based order reduction solutions. For local regions where the deviation still exceeds the threshold, the system calls a local high-fidelity solution for secondary correction, and then completes the full field completion using a physically constrained neural operator. The system simultaneously records the source of each deviation, the object being adjusted, the adjustment magnitude, the residual change, and the updated confidence state, forming a deviation correction record. Finally, the system combines the updated velocity field, pressure field, wall shear stress field, local vortex state, parameter confidence, and deviation correction record to output real-time hemodynamic simulation results.

[0104] Through the above embodiments, the system can perform phase alignment, deviation identification, and parameter closed-loop update of the blood flow state field using the current cardiac observation data in each cardiac cycle or preset time window. This enables the individualized dynamic boundary condition set, modal reliability, and solution residual correction parameters to be continuously adjusted as the observation changes, thereby improving the accuracy, stability, and dynamic adaptability of real-time hemodynamic simulation results.

[0105] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0106] Figure 2 This is a schematic diagram of the structure of the multimodal digital twin real-time simulation system for cardiac blood flow provided in an embodiment of this application. Figure 2 As shown, the system includes: The acquisition module 201 is used to acquire multimodal cardiac image data and cardiac synchronization data of the target object, perform quality evaluation, coordinate registration and cardiac phase normalization on the multimodal cardiac image data, and generate a unified cardiac cycle index. The generation module 202 is used to perform anatomical structure recognition, motion boundary extraction and blood flow observation mapping on multimodal cardiac image data based on a unified cardiac cycle index, and generate a cardiac structure motion fusion representation with modal credibility. The inversion module 203 is used to construct a cardiac anatomical topology map based on the cardiac structure motion fusion representation, and to perform parameter inversion on the inlet state, outlet state, valve state and wall motion state based on the cardiac anatomical topology map to generate an individualized dynamic boundary condition set. Module 204 is used to construct a digital twin cardiac hemodynamic model based on a cardiac anatomical topology map and a set of individualized dynamic boundary conditions. The deduction module 205 is used to input the digital twin cardiac hemodynamics model into the hybrid hemodynamics solver and use local high-fidelity solution, graph structure order reduction solution and physical constraint neural operators to jointly deduce the blood flow state field. The output module 206 is used to update the individualized dynamic boundary condition set, modal confidence and solution residual correction parameters based on the simulation deviation between the current cardiac observation data and the blood flow state field, and output the real-time hemodynamic simulation results.

[0107] In some embodiments, Figure 2 The acquisition module 201 receives multimodal cardiac imaging data and cardiac synchronization data according to a preset acquisition protocol, and generates an image data packet with acquisition time, modality type and cardiac event identifier; it evaluates the quality of the image data packet based on image integrity, motion artifacts and cardiac synchronization consistency, and generates modal quality weights; it performs cross-modal coordinate registration of the image data packet according to the modal quality weights, and maps the registered image data packet to the standard cardiac phase axis according to the cardiac event identifier, and generates a unified cardiac cycle index.

[0108] In some embodiments, Figure 2 The generation module 202 extracts cardiac structural features under each cardiac phase according to a unified cardiac cycle index and determines the cardiac anatomical boundary representation based on cross-modal consistency constraints; it generates dynamic motion boundaries corresponding to the cardiac anatomical boundary representation based on the structural displacement relationship between consecutive cardiac phases; it maps blood flow observation information in multimodal cardiac imaging data to the dynamic motion boundary and corresponding anatomical region, and determines modal credibility based on modal quality weights and mapping residuals, and fuses to generate a cardiac structural motion fusion representation.

[0109] In some embodiments, Figure 2 The inversion module 203 performs anatomical functional partitioning analysis on the cardiac structure motion fusion representation to determine the structural nodes corresponding to the cardiac blood flow path; generates topological relationship edges based on the spatial connectivity and blood flow transmission relationships between each structural node, and associates the dynamic motion boundary and blood flow observation information to the corresponding structural node or topological relationship edge; performs consistency verification on the associated structural nodes and topological relationship edges based on modal credibility, and generates a cardiac anatomical topology map for boundary condition inversion.

[0110] In some embodiments, Figure 2The inversion module 203 determines the blood flow boundary nodes and associated topological relationships based on the cardiac anatomical topology map, and writes the dynamic motion boundary, blood flow observation information and modal confidence into the corresponding blood flow boundary nodes; based on the blood flow conservation constraint and the cardiac cycle continuity constraint, a boundary parameter inversion model is constructed with the goal of minimizing the observation residual; the boundary parameter inversion model is used to jointly estimate the blood flow state parameters and motion constraint parameters of the blood flow boundary nodes, and generate an individualized set of dynamic boundary conditions with parameter confidence.

[0111] In some embodiments, Figure 2 The building module 204 establishes a computational domain corresponding to the cardiac blood flow path based on the cardiac anatomical topology map, and loads the individualized dynamic boundary condition set to the boundary nodes of the computational domain; based on the computational domain, boundary nodes and topological relationships, it constructs a multi-layer model structure containing geometric state, motion state and blood flow state; modal confidence and parameter confidence are written into the correction layer of the multi-layer model structure to generate a digital twin cardiac hemodynamic model for real-time simulation.

[0112] In some embodiments, Figure 2 The derivation module 205 divides the global blood flow path and local fine calculation region based on the digital twin cardiac hemodynamics model, and generates a hierarchical solution task; it uses graph structure reduced-order solution to quickly propagate the global blood flow state, and calls local high-fidelity solution to correct the blood flow state of the corresponding local fine calculation region when the local residual meets the trigger condition; it inputs the hierarchical solution result and individualized dynamic boundary condition set into the physical constraint neural operator, performs field completion based on physical residual constraints, and generates the blood flow state field.

[0113] In some embodiments, Figure 2 The derivation module 205 maps the computational domain in the digital twin cardiac hemodynamics model to a reduced-order blood flow graph, and initializes the node and edge states of the reduced-order blood flow graph according to the individualized dynamic boundary condition set; iteratively propagates the global blood flow state based on the topological relationship and physical constraints of the reduced-order blood flow graph to generate global reduced-order solution results; it determines the trigger region based on the local residual between the global reduced-order solution results and blood flow observation information, and calls the local high-fidelity solution to refine the trigger region, and writes the correction results back to the reduced-order blood flow graph.

[0114] In some embodiments, Figure 2 The output module 206 aligns the current cardiac observation data with the blood flow state field according to the unified cardiac cycle index, and generates a simulation deviation characterization. Based on the simulation deviation characterization, it identifies the source of deviation and adjusts the individualized dynamic boundary condition set, modal confidence and solution residual correction parameters according to the source of deviation. Based on the adjusted digital twin cardiac hemodynamic model, it re-deduces the blood flow state field and generates real-time hemodynamic simulation results by combining the deviation correction record.

[0115] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0116] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments described above.

[0117] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0118] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0119] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0120] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0121] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0122] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for real-time simulation of cardiac blood flow using a multimodal digital twin, characterized in that, include: Acquire multimodal cardiac imaging data and cardiac synchronization data of the target object, perform quality evaluation, coordinate registration and cardiac phase normalization on the multimodal cardiac imaging data, and generate a unified cardiac cycle index; Based on a unified cardiac cycle index, anatomical structure identification, motion boundary extraction, and blood flow observation mapping are performed on multimodal cardiac imaging data to generate a cardiac structure motion fusion representation with modal credibility. Based on the cardiac structure motion fusion representation, a cardiac anatomical topology map is constructed, and based on the cardiac anatomical topology map, the parameters of the inlet state, outlet state, valve state and wall motion state are inverted to generate an individualized dynamic boundary condition set. A digital twin cardiac hemodynamic model is constructed based on a cardiac anatomical topology map and a set of individualized dynamic boundary conditions. The digital twin cardiac hemodynamic model is input into a hybrid hemodynamic solver, which uses local high-fidelity solving, graph structure order reduction solving, and physically constrained neural operators to jointly infer the blood flow state field. Based on the simulation deviation between the current cardiac observation data and the blood flow state field, the individualized dynamic boundary condition set, modal confidence, and solution residual correction parameters are updated, and real-time hemodynamic simulation results are output.

2. The method according to claim 1, characterized in that, The process of acquiring multimodal cardiac imaging data and cardiac synchronization data of the target object, performing quality evaluation, coordinate registration, and cardiac phase normalization on the multimodal cardiac imaging data, and generating a unified cardiac cycle index includes: According to the preset acquisition protocol, multimodal cardiac imaging data and cardiac synchronization data are received, and an image data package with acquisition time, modality type and cardiac event identifier is generated; The quality of image data packets is evaluated based on image integrity, motion artifacts, and cardiac synchronization consistency, and modal quality weights are generated. Cross-modal coordinate registration of image data packets is performed based on modal quality weights, and the registered image data packets are mapped to the standard cardiac phase axis based on cardiac event identifiers to generate a unified cardiac cycle index.

3. The method according to claim 1, characterized in that, The method, based on a unified cardiac cycle index, performs anatomical structure identification, motion boundary extraction, and blood flow observation mapping on multimodal cardiac imaging data to generate a cardiac structure motion fusion representation with modal credibility, including: The cardiac structural features under each cardiac phase are extracted according to the unified cardiac cycle index, and the cardiac anatomical boundary representation is determined based on cross-modal consistency constraints. Based on the structural displacement relationship between consecutive cardiac phases, a dynamic motion boundary corresponding to the cardiac anatomical boundary representation is generated. The blood flow observation information in the multimodal cardiac imaging data is mapped to the dynamic motion boundary and the corresponding anatomical region, and the modal credibility is determined based on the modal quality weight and mapping residual, and the motion fusion representation of the cardiac structure is generated by fusion.

4. The method according to claim 1, characterized in that, The construction of a cardiac anatomical topology map based on the motion fusion representation of cardiac structures includes: Anatomical functional zoning analysis was performed on the cardiac structural motion fusion representation to determine the structural nodes corresponding to the cardiac blood flow pathway. Topological relationship edges are generated based on the spatial connectivity and blood flow transmission relationships between the structural nodes, and dynamic motion boundaries and blood flow observation information are associated with the corresponding structural nodes or topological relationship edges. Based on modal credibility, consistency checks are performed on the associated structural nodes and topological relationship edges to generate a cardiac anatomical topology map for boundary condition inversion.

5. The method according to claim 4, characterized in that, The method involves parametric inversion of the inlet, outlet, valve, and wall motion states based on the cardiac anatomical topology map to generate an individualized set of dynamic boundary conditions, including: Based on the cardiac anatomical topology map, blood flow boundary nodes and associated topological relationship edges are determined, and the dynamic motion boundary, blood flow observation information and modal confidence are written into the corresponding blood flow boundary nodes; Based on blood flow conservation constraints and cardiac cycle continuity constraints, a boundary parameter inversion model is constructed with the goal of minimizing observation residuals; The blood flow state parameters and motion constraint parameters of the blood flow boundary nodes are jointly estimated using the boundary parameter inversion model to generate an individualized set of dynamic boundary conditions with parameter confidence.

6. The method according to claim 1, characterized in that, The digital twin cardiac hemodynamic model, constructed based on cardiac anatomical topology and a set of individualized dynamic boundary conditions, includes: A computational domain corresponding to the cardiac blood flow path is established based on the cardiac anatomical topology map, and the individualized dynamic boundary condition set is loaded onto the boundary nodes of the computational domain; Based on the computational domain, the boundary nodes, and the topological relationships, a multi-layer model structure containing geometric state, motion state, and blood flow state is constructed. Modal confidence and parameter confidence are written into the correction layer of the multi-layer model structure to generate a digital twin cardiac hemodynamic model for real-time simulation.

7. The method according to claim 1, characterized in that, The process of inputting a digital twin cardiac hemodynamic model into a hybrid hemodynamic solver, and using local high-fidelity solving, graph-structured order reduction solving, and physically constrained neural operators to collaboratively deduce the blood flow state field includes: Based on the digital twin cardiac hemodynamics model, the global blood flow path and local fine calculation regions are divided to generate a hierarchical solution task; The global blood flow state is rapidly propagated by using graph structure-based order reduction solution, and the blood flow state of the corresponding local fine calculation region is corrected by calling local high-fidelity solution when the local residual meets the triggering condition; The hierarchical solution results and the set of individualized dynamic boundary conditions are input into the physical constraint neural operator, and the field is completed based on the physical residual constraints to generate the blood flow state field.

8. The method according to claim 7, characterized in that, The method of using graph structure-based order reduction to rapidly propagate the global blood flow state, and calling local high-fidelity solutions to correct the blood flow state of the corresponding local fine-grained calculation region when local residuals meet triggering conditions, includes: The computational domain in the digital twin cardiac hemodynamics model is mapped to a reduced-order blood flow graph, and the node and edge states of the reduced-order blood flow graph are initialized according to the individualized dynamic boundary condition set. The global blood flow state is iteratively propagated based on the topological relationships and physical constraints of the reduced-order blood flow graph to generate a global reduced-order solution result; The trigger region is determined based on the local residual between the global reduced-order solution and the blood flow observation information. The local high-fidelity solution is then called to refine the trigger region, and the correction result is written back to the reduced-order blood flow map.

9. The method according to claim 1, characterized in that, The process involves updating the individualized dynamic boundary condition set, modal reliability, and residual correction parameters based on the simulation deviation between the current cardiac observation data and the blood flow state field, and outputting real-time hemodynamic simulation results, including: The current cardiac observation data is phase-aligned with the blood flow state field according to the unified cardiac cycle index to generate a simulation deviation characterization. The source of deviation is identified based on the simulation deviation characterization, and the individualized dynamic boundary condition set, the modal confidence level, and the solution residual correction parameters are adjusted according to the source of deviation. The blood flow state field was re-inferred based on the adjusted digital twin cardiac hemodynamic model, and real-time hemodynamic simulation results were generated by combining the deviation correction record.

10. A multimodal digital twin real-time simulation system for cardiac blood flow, characterized in that, include: The acquisition module is used to acquire multimodal cardiac imaging data and cardiac synchronization data of the target object, perform quality evaluation, coordinate registration and cardiac phase normalization on the multimodal cardiac imaging data, and generate a unified cardiac cycle index. The generation module is used to perform anatomical structure recognition, motion boundary extraction, and blood flow observation mapping on multimodal cardiac imaging data based on a unified cardiac cycle index, and generate a cardiac structure motion fusion representation with modal credibility. The inversion module is used to construct a cardiac anatomical topology map based on the cardiac structure motion fusion representation, and to perform parameter inversion on the inlet state, outlet state, valve state and wall motion state based on the cardiac anatomical topology map to generate an individualized dynamic boundary condition set. Build modules are used to construct digital twin cardiac hemodynamic models based on cardiac anatomical topology maps and individualized dynamic boundary condition sets; The deduction module is used to input the digital twin cardiac hemodynamics model into the hybrid hemodynamics solver and use local high-fidelity solution, graph structure order reduction solution and physical constraint neural operators to jointly deduce the blood flow state field. The output module is used to update the individualized dynamic boundary condition set, modal confidence and solution residual correction parameters based on the simulation deviation between the current cardiac observation data and the blood flow state field, and output the real-time hemodynamic simulation results.

Images