A Method and System for Digital Human Vascular Modeling and Representation Based on Multimodal Large Model
By employing a multimodal large-scale digital human vascular modeling method, the problems of multi-source data integration and feature fusion are solved, generating high-precision 3D vascular models and constructing knowledge graphs, thereby enabling accurate diagnosis and personalized treatment support for vascular diseases.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI PUDONG HOSPITAL
- Filing Date
- 2025-11-05
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies for vascular disease diagnosis suffer from difficulties in integrating multi-source data, outdated feature extraction and fusion methods, insufficient dynamism and accuracy, and a lack of a unified representation system, which prevent the full utilization of the complementarity and accuracy between multimodal data.
A digital human vascular modeling method based on multimodal large models is adopted. Through multimodal feature fusion encoding, 3D vascular modeling and multidimensional representation, combined with self-attention mechanism and cross-attention mechanism, a high-precision 3D vascular model is generated, and a knowledge graph is constructed for unified representation.
It achieves deep fusion of multimodal data, improves the spatial accuracy and dynamic response capability of vascular models, provides interpretable decision support, and offers a powerful tool for early screening, accurate diagnosis and personalized treatment of vascular diseases.
Smart Images
Figure CN121148716B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, and in particular to a method and system for digital human blood vessel modeling and characterization based on a multimodal large model. Background Technology
[0002] In the clinical diagnosis and treatment of vascular diseases, accurate vascular modeling and comprehensive characterization are core prerequisites for achieving precision medicine. However, current technologies still face the following key bottlenecks when dealing with the complex, dynamic, and highly individualized physiological structure of human blood vessels:
[0003] Multi-source data integration is challenging: Clinical practice has accumulated a wealth of multimodal vascular data, including but not limited to imaging data such as computed tomography angiography (CTA) and magnetic resonance angiography (MRA), as well as functional data such as pulse wave velocity (PWV). However, existing methods often treat these data as isolated information, lacking effective mechanisms for deep fusion, thus failing to fully utilize the complementarity between data to improve the overall performance of the model.
[0004] Feature extraction and fusion methods are outdated: Traditional feature extraction methods (such as those based on hand-designed filters) struggle to adaptively capture complex vascular morphology and functional features. Meanwhile, the fusion of multimodal features often remains at the level of simple concatenation or weighted summation, failing to establish high-order correlations between different modal information and limiting the model's ability to finely characterize vascular states.
[0005] The dynamism and accuracy of 3D modeling are insufficient: Existing 3D vascular reconstruction techniques typically rely on static data at a single time point, failing to reflect the dynamic changes of blood vessels under different physiological states. Furthermore, the model construction process often suffers from insufficient detail in depicting vascular layering, boundary regions, and microenvironment features, affecting the model's accuracy and its guidance for clinical details.
[0006] The lack of a unified knowledge representation system: Analysis results of blood vessels are typically presented as fragmented numerical indicators (such as vessel diameter and stenosis rate), lacking a unified representation framework that organically integrates anatomical structure, physiological function, pathological state, and clinical information. This makes it difficult for physicians to understand the health status of blood vessels from a holistic perspective and limits the interpretability and depth of application of artificial intelligence models in clinical decision support.
[0007] Therefore, there is an urgent need in this field for a technical solution that can deeply integrate multimodal data, dynamically construct high-precision three-dimensional models, and comprehensively represent them in a knowledge-based manner, so as to break through the limitations of existing technologies and provide strong technical support for the early screening, accurate diagnosis and personalized treatment of vascular diseases. Summary of the Invention
[0008] The purpose of this invention is to provide a method and system for digital human vascular modeling and characterization based on a multimodal large model, so as to solve the problems existing in the background technology.
[0009] To achieve the above objectives, this invention provides a digital human blood vessel modeling and characterization method based on a multimodal large model, comprising the following steps:
[0010] S1. Multimodal vascular data acquisition and preprocessing: Acquire multi-source modal data containing human vascular information and perform standardized preprocessing.
[0011] S2. Multimodal feature fusion coding: Feature extraction and fusion coding are performed on the preprocessed modal data to obtain multimodal fusion features;
[0012] S3, 3D modeling of blood vessels: Based on multimodal fusion features, a high-precision 3D model of blood vessels is generated through 3D reconstruction.
[0013] S4. Multidimensional characterization of blood vessels: Analyze the three-dimensional model of blood vessels and the original multimodal data using a multimodal large model to extract the morphological, functional and pathological features of blood vessels and form a comprehensive characterization vector.
[0014] S5. Characterization optimization and storage: The comprehensive characterization vector is optimized and stored in association with the 3D vascular model for subsequent vascular disease analysis and application.
[0015] Preferably, the specific content of S2 is as follows:
[0016] For each modality, a feature extraction sub-network is established to generate the first feature of each modality. Each feature extraction sub-network adopts a ResNet network with a self-attention mechanism.
[0017] By utilizing the first feature, a multimodal cross-attention mechanism is introduced among multiple modalities to generate a second feature for each modality;
[0018] The second features of each modality are fused to generate a fused feature K;
[0019] The fused feature K is subjected to global average pooling and global max pooling respectively to generate the first channel vector G1 and the second channel vector G2.
[0020] The first channel vector G1 and the second channel vector G2 are convolved in one dimension and their features are added together. The result of the addition is then passed through the Softmax function to generate cross-channel weight coefficients corresponding to each modality.
[0021] The second feature of each mode is weighted using the cross-channel weight coefficients corresponding to the mode, and the third feature of each mode is generated.
[0022] The final multimodal fusion feature is generated using the third feature of each modality.
[0023] Preferably, at least one Bottleneck layer in the ResNet network that incorporates a self-attention mechanism includes, in sequence along the input direction, a 1x1 convolutional layer, a first self-attention mechanism layer, a 3x3 convolutional layer, a second self-attention mechanism layer, and a 1x1 convolutional layer.
[0024] Preferably, the first self-attention mechanism layer and the second self-attention mechanism layer receive the feature map of the previous layer, calculate the weight of each position in the feature map of the previous layer using the self-attention mechanism layer, and perform a weighted average of the feature map of the previous layer using the weight of each position, thereby generating the output of the corresponding first self-attention mechanism layer or second self-attention mechanism layer.
[0025] Preferably, the generation of the second feature for each modality includes:
[0026] Cross-attention features for each modality are calculated using a cross-attention mechanism;
[0027] Perform a 1x1 convolution on the cross-attention features of each modality, and add the result of the convolution to the first feature of the corresponding modality to obtain the second feature of each modality.
[0028] Preferably, the cross-attention feature for each modality is calculated using a cross-attention mechanism, including:
[0029] Calculate the attention score between any modalities , ;
[0030] The attention score is normalized to generate a normalized result. , ;
[0031] Determine the cross-attention weights for any modality , ;
[0032] in, Representing modes The query vector; express The key vector of a mode; express k The value vector of the modality; This represents the total number of modes.
[0033] Preferably, S3 includes:
[0034] Based on the standard vascular model, the 3D vascular model is divided into multiple layers, and the vascular characterization data corresponding to each layer, as well as the vascular cell type change characteristics and blood component data characteristics at the boundaries of each layer are obtained.
[0035] Based on the vascular representation data corresponding to each layer, the acquired fused multimodal data is classified to obtain multimodal datasets corresponding to each layer; the standard vascular model is adjusted based on the multimodal datasets to determine the 3D vascular model of each layer corresponding to the current vascular model, thus obtaining a primary vascular model.
[0036] Simultaneously, based on the characteristics of changes in vascular cell types and blood component data at the boundaries of each layer, the three-dimensional model of the vascular body at each layer corresponding to the current vascular body is used to determine the decomposition characteristics of each layer of the current vascular body.
[0037] Based on the decomposition features of each layer of the current blood vessel, the corresponding layers of the primary blood vessel model are fused to obtain a real-time three-dimensional model of the current blood vessel environment.
[0038] Preferably, S4 includes:
[0039] The three-dimensional vascular model, the optimized comprehensive representation vector, and key information from clinical information are identified as entities in the knowledge graph. A unique identifier is assigned to each entity, the types of relationships between entities are determined, and associations are established.
[0040] A digital human vascular modeling and representation system based on a multimodal large model includes:
[0041] The data acquisition module is used to collect multi-source modal data containing human vascular information;
[0042] Data preprocessing module: Used to perform standardized preprocessing on the collected multi-source modal data, including data format conversion, image denoising, contrast enhancement and other processing steps;
[0043] The multimodal fusion coding module is used to extract and fuse the features of the preprocessed modal data to achieve deep fusion of multimodal features and obtain multimodal fusion features.
[0044] The 3D modeling module is used to generate high-precision 3D models of blood vessels through 3D reconstruction based on multimodal fusion features.
[0045] The multi-dimensional representation module is used to analyze the three-dimensional model of blood vessels and the original multi-modal data using a large multi-modal model, extract the multi-dimensional features of blood vessels, and form a comprehensive representation vector.
[0046] The storage and application module is used to optimize the comprehensive representation vector and store it in association with the 3D vascular model, providing an interface for vascular disease analysis and application.
[0047] Preferably, the 3D modeling module includes a 3D model building unit and a model updating and visualization unit:
[0048] A 3D model building unit is used to construct a real-time 3D model of the vascular environment based on fused multimodal data using 3D modeling technology; it includes:
[0049] The model hierarchical subunit is used to divide the 3D vascular model into multiple layers based on the standard vascular model, and to obtain the vascular characterization data corresponding to each layer, as well as the vascular cell type change characteristics and blood component data characteristics at the boundaries of each layer.
[0050] The model building subunit is used to classify the acquired fused multimodal data based on the vascular representation data corresponding to each layer, and obtain the multimodal dataset corresponding to each layer respectively.
[0051] The standard vascular model is adjusted based on a multimodal dataset to determine the 3D model of the vascular system at each level corresponding to the current vascular system, thus obtaining a primary vascular model.
[0052] Simultaneously, based on the characteristics of changes in vascular cell types and blood component data at the boundaries of each layer, the three-dimensional model of the vascular body at each layer corresponding to the current vascular body is used to determine the decomposition characteristics of each layer of the current vascular body.
[0053] Based on the decomposition features of each layer of the current blood vessel, the corresponding layers of the primary blood vessel model are fused to obtain the real-time three-dimensional model of the current blood vessel environment.
[0054] The model update and visualization unit is used to receive new multi-source data in real time, recalculate weights according to the dynamic weight allocation algorithm, and perform data fusion.
[0055] Therefore, the present invention employs the above-mentioned digital human blood vessel modeling and characterization method and system based on a multimodal large model, which has the following beneficial effects:
[0056] (1) This enables the system to automatically learn and focus on the most critical information for blood vessel modeling in different modal data and establish high-order correlations between modalities, thereby generating multimodal fusion features that are more discriminative and robust than traditional methods.
[0057] (2) Significantly improved the spatial accuracy and structural details of the vascular model. Simultaneously, through the dynamic weight allocation algorithm, the model can receive new data and update in real time, thus more realistically reflecting the dynamic physiological state of blood vessels;
[0058] (3) By constructing a knowledge graph, the model, features and clinical information are linked, providing doctors with interpretable and reasonable decision support basis;
[0059] (4) It provides a powerful quantitative tool for early screening, accurate diagnosis, risk assessment and personalized treatment of vascular diseases, which helps to promote the development and application of smart healthcare.
[0060] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0061] Figure 1 This is a flowchart of the digital human blood vessel modeling and characterization method based on a multimodal large model according to the present invention;
[0062] Figure 2 This is a schematic diagram illustrating the specific weight calculation in an embodiment of the present invention;
[0063] Figure 3 This is a schematic diagram of the cross-attention mechanism in an embodiment of the present invention. Detailed Implementation
[0064] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0065] Please see Figures 1-3 A digital human vascular modeling and characterization method based on a multimodal large model includes the following steps:
[0066] S1. Multimodal vascular data acquisition and preprocessing: Acquire multi-source modal data containing human vascular information and perform standardized preprocessing.
[0067] S2. Multimodal feature fusion coding: Feature extraction and fusion coding are performed on the preprocessed modal data to obtain multimodal fusion features.
[0068] The specific content is as follows:
[0069] For each modality, a feature extraction sub-network is established to generate the first feature of each modality. Each feature extraction sub-network adopts a ResNet network with a self-attention mechanism.
[0070] By utilizing the first feature, a multimodal cross-attention mechanism is introduced among multiple modalities to generate a second feature for each modality;
[0071] The second features of each modality are fused to generate a fused feature K;
[0072] The fused feature K is subjected to global average pooling and global max pooling respectively to generate the first channel vector G1 and the second channel vector G2.
[0073] The first channel vector G1 and the second channel vector G2 are convolved in one dimension and their features are added together. The result of the addition is then passed through the Softmax function to generate cross-channel weight coefficients corresponding to each modality.
[0074] The second feature of each mode is weighted using the cross-channel weight coefficients corresponding to the mode, and the third feature of each mode is generated.
[0075] The final multimodal fusion feature is generated using the third feature of each modality.
[0076] The first feature is a high-level feature rich in internal modality context information, obtained after each modality passes through its independent, enhanced feature extraction network. The second feature is a feature enhanced and enriched by information from other modalities after the first feature of each modality undergoes cross-attention interaction. The third feature is a refined feature obtained by weighting the second feature of each modality according to its channel importance (dynamically calculated from the fused features).
[0077] At least one Bottleneck layer in a ResNet network that incorporates a self-attention mechanism contains, in sequence along the input direction, a 1x1 convolutional layer, a first self-attention mechanism layer, a 3x3 convolutional layer, a second self-attention mechanism layer, and a 1x1 convolutional layer.
[0078] The first and second self-attention mechanism layers receive the feature map from the previous layer, calculate the weight of each position in the feature map using the self-attention mechanism layer, and perform a weighted average of the feature map using the weight of each position to generate the output of the corresponding first or second self-attention mechanism layer.
[0079] The generation of the second feature for each modality includes:
[0080] Cross-attention features for each modality are calculated using a cross-attention mechanism;
[0081] Perform a 1x1 convolution on the cross-attention features of each modality, and add the result of the convolution to the first feature of the corresponding modality to obtain the second feature of each modality.
[0082] The cross-attention mechanism is used to calculate the cross-attention features for each modality, including:
[0083] Calculate the attention score between any modalities , ;
[0084] The attention score is normalized to generate a normalized result. , ;
[0085] Determine the cross-attention weights for any modality , ;
[0086] in, Representing modes The query vector; express The key vector of a mode; express The value vector of the modality; This represents the total number of modes.
[0087] S3, 3D modeling of blood vessels: Based on multimodal fusion features, a high-precision 3D model of blood vessels is generated through 3D reconstruction.
[0088] include:
[0089] Based on the standard vascular model, the 3D vascular model is divided into multiple layers, and the vascular characterization data corresponding to each layer, as well as the vascular cell type change characteristics and blood component data characteristics at the boundaries of each layer are obtained.
[0090] Based on the vascular representation data corresponding to each layer, the acquired fused multimodal data is classified to obtain multimodal datasets corresponding to each layer; the standard vascular model is adjusted based on the multimodal datasets to determine the 3D vascular model of each layer corresponding to the current vascular model, thus obtaining a primary vascular model.
[0091] Simultaneously, based on the characteristics of changes in vascular cell types and blood component data at the boundaries of each layer, the three-dimensional model of the vascular body at each layer corresponding to the current vascular body is used to determine the decomposition characteristics of each layer of the current vascular body.
[0092] Based on the decomposition features of each layer of the current blood vessel, the corresponding layers of the primary blood vessel model are fused to obtain a real-time three-dimensional model of the current blood vessel environment.
[0093] S4. Multidimensional characterization of blood vessels: Analyze the three-dimensional model of blood vessels and the original multimodal data using a multimodal large model to extract the morphological, functional and pathological features of blood vessels and form a comprehensive characterization vector.
[0094] S5. Characterization optimization and storage: The comprehensive characterization vector is optimized and stored in association with the 3D vascular model for subsequent vascular disease analysis and application.
[0095] A digital human vascular modeling and representation system based on a multimodal large model includes:
[0096] The data acquisition module is used to collect multi-source modal data containing human vascular information;
[0097] Data preprocessing module: Used to perform standardized preprocessing on the collected multi-source modal data, including data format conversion, image denoising, contrast enhancement and other processing steps;
[0098] The multimodal fusion coding module is used to extract and fuse the features of the preprocessed modal data to achieve deep fusion of multimodal features and obtain multimodal fusion features.
[0099] The 3D modeling module is used to generate high-precision 3D models of blood vessels through 3D reconstruction based on multimodal fusion features.
[0100] The multi-dimensional representation module is used to analyze the three-dimensional model of blood vessels and the original multi-modal data using a large multi-modal model, extract the multi-dimensional features of blood vessels, and form a comprehensive representation vector.
[0101] The storage and application module is used to optimize the comprehensive representation vector and store it in association with the 3D vascular model, providing an interface for vascular disease analysis and application.
[0102] The 3D modeling module includes a 3D model building unit and a model updating and visualization unit:
[0103] A 3D model building unit is used to construct a real-time 3D model of the vascular environment based on fused multimodal data using 3D modeling technology; it includes:
[0104] The model hierarchical subunit is used to divide the 3D vascular model into multiple layers based on the standard vascular model, and to obtain the vascular characterization data corresponding to each layer, as well as the vascular cell type change characteristics and blood component data characteristics at the boundaries of each layer.
[0105] The model building subunit is used to classify the acquired fused multimodal data based on the vascular representation data corresponding to each layer, and obtain the multimodal dataset corresponding to each layer respectively.
[0106] The standard vascular model is adjusted based on a multimodal dataset to determine the 3D model of the vascular system at each level corresponding to the current vascular system, thus obtaining a primary vascular model.
[0107] Simultaneously, based on the characteristics of changes in vascular cell types and blood component data at the boundaries of each layer, the three-dimensional model of the vascular body at each layer corresponding to the current vascular body is used to determine the decomposition characteristics of each layer of the current vascular body.
[0108] Based on the decomposition features of each layer of the current blood vessel, the corresponding layers of the primary blood vessel model are fused to obtain the real-time three-dimensional model of the current blood vessel environment.
[0109] The model update and visualization unit is used to receive new multi-source data in real time, recalculate weights according to the dynamic weight allocation algorithm, and perform data fusion.
[0110] Therefore, the present invention employs the above-mentioned digital human vascular modeling and characterization method and system based on multimodal large models, providing a powerful quantitative tool for early screening, accurate diagnosis, risk assessment and personalized treatment planning of vascular diseases, which helps to promote the development and application of intelligent healthcare.
[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for digital human vascular modeling and characterization based on multi-modal large models, characterized in that, Includes the following steps: S1. Multimodal vascular data acquisition and preprocessing: Acquire multi-source modal data containing human vascular information and perform standardized preprocessing. S2. Multimodal feature fusion coding: Feature extraction and fusion coding are performed on the preprocessed modal data to obtain multimodal fusion features; S3. Three-dimensional modeling of blood vessels: Based on multimodal fusion features, high-precision three-dimensional models of blood vessels are generated through three-dimensional reconstruction; including: Based on the standard vascular model, the 3D vascular model is divided into multiple layers, and the vascular characterization data corresponding to each layer, as well as the vascular cell type change characteristics and blood component data characteristics at the boundaries of each layer are obtained. Based on the vascular representation data corresponding to each layer, the acquired fused multimodal data is classified to obtain multimodal datasets corresponding to each layer; the standard vascular model is adjusted based on the multimodal datasets to determine the 3D vascular model of each layer corresponding to the current vascular model, thus obtaining a primary vascular model. Simultaneously, based on the characteristics of changes in vascular cell types and blood component data at the boundaries of each layer, the three-dimensional model of the vascular body at each layer corresponding to the current vascular body is used to determine the decomposition characteristics of each layer of the current vascular body. Based on the decomposition features of each layer of the current blood vessel, the corresponding layers of the primary blood vessel model are fused to obtain the real-time three-dimensional model of the current blood vessel environment. S4. Multidimensional characterization of blood vessels: Analyze the three-dimensional model of blood vessels and the original multimodal data using a multimodal large model to extract the morphological, functional and pathological features of blood vessels and form a comprehensive characterization vector. S5. Representation Optimization and Storage: The comprehensive representation vector is optimized and stored in association with the 3D blood vessel model.
2. The multi-modal large model based digital human vascular modeling and characterization method according to claim 1, wherein, The specific content of S2 is as follows: For each modality, a feature extraction sub-network is established to generate the first feature of each modality. Each feature extraction sub-network adopts a ResNet network with a self-attention mechanism. By utilizing the first feature, a multimodal cross-attention mechanism is introduced among multiple modalities to generate a second feature for each modality; The second features of each modality are fused to generate a fused feature K; The fused feature K is subjected to global average pooling and global max pooling respectively to generate the first channel vector G1 and the second channel vector G2. The first channel vector G1 and the second channel vector G2 are convolved in one dimension and their features are added together. The result of the addition is then passed through the Softmax function to generate cross-channel weight coefficients corresponding to each modality. The second feature of each mode is weighted using the cross-channel weight coefficients corresponding to the mode, and the third feature of each mode is generated. The final multimodal fusion feature is generated using the third feature of each modality.
3. The digital human blood vessel modeling and characterization method based on a multimodal large model according to claim 2, characterized in that: At least one Bottleneck layer in a ResNet network that incorporates a self-attention mechanism contains, in sequence along the input direction, a 1x1 convolutional layer, a first self-attention mechanism layer, a 3x3 convolutional layer, a second self-attention mechanism layer, and a 1x1 convolutional layer.
4. The digital human blood vessel modeling and characterization method based on a multimodal large model according to claim 3, characterized in that: The first and second self-attention mechanism layers receive the feature map from the previous layer, calculate the weight of each position in the feature map using the self-attention mechanism layer, and perform a weighted average of the feature map using the weight of each position to generate the output of the corresponding first or second self-attention mechanism layer.
5. The digital human blood vessel modeling and characterization method based on a multimodal large model according to claim 4, characterized in that, The generation of the second feature for each modality includes: Cross-attention features for each modality are calculated using a cross-attention mechanism; Perform a 1x1 convolution on the cross-attention features of each modality, and add the result of the convolution to the first feature of the corresponding modality to obtain the second feature of each modality.
6. The digital human blood vessel modeling and characterization method based on a multimodal large model according to claim 5, characterized in that, The cross-attention mechanism is used to calculate the cross-attention features for each modality, including: Calculate the attention score between any modalities , ; The attention score is normalized to generate a normalized result. , ; Determine the cross-attention weights for any modality , ; in, Representing modes The query vector; express The key vector of a mode; express The value vector of the modality; This represents the total number of modes.
7. The digital human blood vessel modeling and characterization method based on a multimodal large model according to claim 1, characterized in that, S4 includes: The three-dimensional vascular model, the optimized comprehensive representation vector, and key information from clinical information are identified as entities in the knowledge graph. A unique identifier is assigned to each entity, the types of relationships between entities are determined, and associations are established.
8. A digital human vascular modeling and characterization system based on a multimodal large model, employing the digital human vascular modeling and characterization method based on a multimodal large model as described in any one of claims 1-7, characterized in that: The data acquisition module is used to collect multi-source modal data containing human vascular information; Data preprocessing module: used to perform standardized preprocessing on the collected multi-source modal data, including data format conversion, image denoising, and contrast enhancement processing steps; The multimodal fusion coding module is used to extract and fuse the features of the preprocessed modal data to achieve deep fusion of multimodal features and obtain multimodal fusion features. The 3D modeling module is used to generate high-precision 3D models of blood vessels through 3D reconstruction based on multimodal fusion features. The multi-dimensional representation module is used to analyze the three-dimensional model of blood vessels and the original multi-modal data using a large multi-modal model, extract the multi-dimensional features of blood vessels, and form a comprehensive representation vector. The storage and application module is used to optimize the comprehensive representation vector and store it in association with the 3D vascular model, providing an interface for vascular disease analysis and application.
9. The digital human vascular modeling and characterization system based on a multimodal large model according to claim 8, characterized in that, The 3D modeling module includes a 3D model building unit and a model updating and visualization unit: A 3D model building unit is used to construct a real-time 3D model of the vascular environment based on fused multimodal data using 3D modeling technology; it includes: The model hierarchical subunit is used to divide the 3D vascular model into multiple layers based on the standard vascular model, and to obtain the vascular characterization data corresponding to each layer, as well as the vascular cell type change characteristics and blood component data characteristics at the boundaries of each layer. The model building subunit is used to classify the acquired fused multimodal data based on the vascular representation data corresponding to each layer, and obtain the multimodal dataset corresponding to each layer respectively. The standard vascular model is adjusted based on a multimodal dataset to determine the 3D model of the vascular system at each level corresponding to the current vascular system, thus obtaining a primary vascular model. Simultaneously, based on the characteristics of changes in vascular cell types and blood component data at the boundaries of each layer, the three-dimensional model of the vascular body at each layer corresponding to the current vascular body is used to determine the decomposition characteristics of each layer of the current vascular body. Based on the decomposition features of each layer of the current blood vessel, the corresponding layers of the primary blood vessel model are fused to obtain the real-time three-dimensional model of the current blood vessel environment. The model update and visualization unit is used to receive new multi-source data in real time, recalculate weights according to the dynamic weight allocation algorithm, and perform data fusion.