Dynamic graph convolution network driven organ and vessel topology optimized surface reconstruction method

Abstract

The application provides an organ and blood vessel topological optimization surface reconstruction method driven by a dynamic graph convolution network, and belongs to the technical field of medical treatment.The organ and blood vessel topological optimization surface reconstruction method driven by the dynamic graph convolution network comprises the following steps: step one: acquiring a CT image of a target part of a human body, performing segmentation processing on the CT image, obtaining a binary segmentation mask graph corresponding to organs and blood vessels, extracting an anatomical skeleton point of the binary segmentation mask graph, and constructing an organ-blood vessel mixed topological graph based on the anatomical skeleton point; and step two: building an adaptive dynamic adjacency matrix graph convolution network, dynamically adjusting network node weights based on blood vessel diameters and organ spatial distances, fusing CT image gray scale features and topological graph connectivity features, and extracting multi-scale topological fusion features.The application can completely retain microvessel branches and real organ morphologies, does not need manual post-processing, is suitable for image reconstruction of multiple human body parts, has high three-dimensional visualization precision, and has extremely high clinical application value.

Description

Technical Field

[0001] This invention belongs to the field of medical treatment technology, specifically a method for topological optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks. Background Technology

[0002] Three-dimensional reconstruction of medical CT images is a key technology for converting two-dimensional tomographic images into three-dimensional models, and it is widely used in clinical surgery, medical teaching, virtual simulation, and artificial intelligence-assisted diagnosis. Currently, the mainstream three-dimensional reconstruction technologies for medical organs and blood vessels are mainly divided into two categories: threshold segmentation reconstruction and traditional convolutional neural network (CNN) reconstruction. Both types of technologies have inherent technical defects and cannot meet the needs of high-precision clinical applications.

[0003] Traditional thresholding reconstruction techniques rely on pixel grayscale thresholds to complete image segmentation and reconstruction. They can only utilize the grayscale information of single pixels and local neighborhoods, completely ignoring the natural topological relationships of human anatomical structures. Human blood vessels have multi-level branching, gradual changes in thickness, and extremely strong connectivity. There is a fixed spatial attachment topological relationship between organs and blood vessels. Thresholding methods, which only model through grayscale differences, are prone to causing the breakage of small blood vessel branches and the loss of blood vessel terminals. At the same time, they cause unevenness on the surface of organs and local morphological distortion, making it impossible to restore the true human anatomical structure.

[0004] Compared to thresholding methods, traditional static CNN reconstruction techniques can extract shallow and deep image texture features, improving reconstruction accuracy to some extent. However, they still have core drawbacks: First, traditional CNNs are static network structures with fixed weights and fixed receptive fields, which cannot adapt to the multi-scale features of human blood vessels. They cannot adaptively adjust the modeling strategy to address the morphological differences between large main blood vessels, medium-sized branch blood vessels, and micron-sized microvessels. The extraction of small blood vessel branch features fails, resulting in severe loss of microvessels after reconstruction. Second, CNNs only focus on two-dimensional spatial features of images and cannot model the three-dimensional spatial connectivity topology between organs and blood vessels, severing the overall correlation of anatomical structures and resulting in structural defects in the reconstruction model.

[0005] More critically, all current mainstream reconstruction techniques lack post-reconstruction topology optimization mechanisms. After initial reconstruction, models commonly exhibit artifacts such as internal vascular cavities, organ surface burrs, broken vascular branches, and separation of organ-vascular junctions. These require manual post-construction smoothing, finishing, and finishing, which is not only time-consuming and labor-intensive but also prone to subjective errors, making standardized and high-precision reconstruction impossible. These shortcomings result in insufficient accuracy of existing reconstruction models, failing to meet the stringent precision requirements of precision surgical planning, accurate 3D visualization, and medical simulation teaching, thus severely limiting the practical application of 3D reconstruction technology in precision medicine.

[0006] Therefore, this invention proposes a dynamic graph convolutional network-driven method for topology optimization surface reconstruction of organs and blood vessels to compensate for and improve the shortcomings of existing technologies. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention proposes a method for topology optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks.

[0008] To achieve the above objectives, the present invention employs the following technical solution: A dynamic graph convolutional network-driven method for topology-optimized surface reconstruction of organs and blood vessels includes the following steps: Step 1: Acquire CT images of the target body part, segment the CT images to obtain binary segmentation masks corresponding to organs and blood vessels, extract the anatomical skeleton points of the binary segmentation masks, and construct organ-blood vessel hybrid topology maps based on the anatomical skeleton points. Step 2: Construct an adaptive dynamic adjacency matrix graph convolutional network, dynamically adjust the network node weights based on blood vessel diameter and organ spatial distance, fuse CT image grayscale features and topology graph connectivity features, and extract multi-scale topology fusion features; Step 3: Input the multi-scale topology fusion features into the 3D reconstruction module to generate the initial 3D mesh surface model of the organ and blood vessel; Step 4: Verify and repair the initial 3D mesh surface model based on topological connectivity, complete hole filling, branch completion, and surface smoothing, constrain the topological continuity of the organ and blood vessel junction, and output a high-precision 3D reconstruction model.

[0009] Preferably, in step one, the method for constructing the organ-blood vessel hybrid topology map includes: The binary segmentation mask image is denoised and morphologically optimized to remove isolated noise points; the organ contour skeleton and the blood vessel midline skeleton are extracted, and several three-dimensional skeleton feature nodes are obtained by uniform sampling; each skeleton feature node is used as a topological node, and the three-dimensional spatial adjacency relationship and anatomical connectivity relationship between nodes are used as topological edges to quantify the spatial distance and structural correlation of nodes, and to construct a hybrid topological map covering the global structure of organs and blood vessels.

[0010] Preferably, in step two, the rules for dynamically adjusting the weights of network nodes include: For large blood vessel trunks and core nodes of major organs, the weight of the neighborhood receptive field is increased to enhance the learning of main structural features; for microvascular branches and organ edge nodes, the weight of topological features is refined to accurately capture subtle anatomical structures; and node weights are updated in real time through a dynamic adjacency matrix to adapt to the boundary features of blood vessels and irregular organs at multiple scales.

[0011] Preferably, in step two, the extraction method of the multi-scale topological fusion feature is as follows: simultaneously extract the two-dimensional grayscale texture features of the CT image and the three-dimensional global connectivity features of the topological map, and obtain a fusion feature vector that has both local details and global topological correlation through feature splicing and weighted fusion.

[0012] Preferably, in step three, the three-dimensional reconstruction module adopts the Marching Cube algorithm to complete the initial reconstruction of the three-dimensional surface of organs and blood vessels through voxel traversal, isosurface extraction, and triangular patch fitting.

[0013] Preferably, in step four, the topology connectivity verification and repair includes: Based on the standard anatomical connectivity rules of the hybrid topology graph, the model identifies ruptured blood vessels, internal cavities, surface burrs, and structural separation artifacts. The model completes the ruptured blood vessel branches through interpolation algorithms, repairs the cavitary regions through topological constraint filling algorithms, removes surface burrs through feature-preserving smoothing algorithms, and calibrates the boundaries between organs and blood vessels to ensure overall topological continuity.

[0014] Preferably, the method is applicable to three-dimensional reconstruction of CT images of the lungs, heart, thoracic and abdominal organs, and human blood vessels, and can be applied to preoperative surgical planning, medical three-dimensional visualization, teaching simulation, and precise lesion localization.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Advantage of topological integrity: By constructing organ-blood vessel hybrid topology map, it is the first to realize the fusion modeling of local grayscale features of medical images and global anatomical topology features of the human body, completely solving the defect of traditional technology that ignores topological correlation, completely preserving microvascular branches at all levels, and the reconstructed model has no broken blood vessels or missing branches, with extremely high anatomical structure restoration.

[0016] 2. Multi-scale adaptive advantage: The innovative design of the dynamic adjacency matrix graph convolutional network can adaptively adjust the feature extraction weights according to the thickness of blood vessels, organ morphology, and spatial location, perfectly adapting to multi-scale anatomical structures such as thick blood vessels, thin branches, and irregular organ boundaries, solving the pain points of poor adaptability and failure of fine structure reconstruction in traditional static networks.

[0017] 3. Model accuracy and image quality advantages: Through the post-topology intelligent repair module, the voids, burrs, artifacts and structural separation problems of the reconstructed model are automatically eliminated, the organ surface is smooth and uniform, the blood vessel wall is regular, the three-dimensional visualization accuracy is greatly improved, and the morphological error is far lower than that of traditional reconstruction algorithms.

[0018] 4. Advantages in versatility and adaptability: The algorithm of this invention can be adapted to the reconstruction of CT images of multiple parts such as the lungs, heart, thoracic and abdominal organs, and blood vessels throughout the body. It is not limited by differences in organ morphology and blood vessel distribution, and has a wide range of applicable scenarios.

[0019] 5. High efficiency and intelligent advantages: The entire reconstruction and optimization process is automated, eliminating the need for manual post-construction polishing, completion, and correction. This significantly reduces labor costs and operational errors, improving reconstruction efficiency by more than 60%. It can be directly applied to core clinical scenarios such as preoperative 3D surgical planning, medical teaching simulation, precise lesion localization, and postoperative effect evaluation. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of the present invention; Detailed Implementation

[0022] To make the technical means, creative features, achieved objectives, and effects of this invention readily understandable, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope defined by this application.

[0023] I. Data Preprocessing and Topology Graph Construction Enhanced CT images of the lungs of clinical patients were acquired with a slice thickness of 1 mm and a resolution of 512×512. The U-Net segmentation network was used to accurately segment the lung parenchyma and pulmonary arteries and veins, resulting in a binary segmentation mask image.

[0024] Morphological denoising was performed on the binary mask to remove isolated noise points and small interference areas, ensuring the basic accuracy of the image. A thinning algorithm was used to extract the central skeleton curves of organs and blood vessels, and a large number of skeleton feature nodes were obtained by uniform sampling. The three-dimensional spatial coordinates, the structure to which the node belongs (organ / blood vessel), and the corresponding blood vessel diameter parameters of each node were recorded.

[0025] Using each skeleton node as a topological node, when the three-dimensional spatial distance between two nodes is less than a preset threshold (2mm) and conforms to the human anatomical connectivity logic, an undirected topological edge is constructed, ultimately forming a hybrid topological graph of lung organs and blood vessels, which fully covers the topological relationships of the main blood vessels, secondary branches, microvessels and lung parenchyma of the lung.

[0026] II. Feature Extraction from Dynamic Graph Convolutional Networks An adaptive dynamic graph convolutional network was constructed, comprising an input layer, a dynamic adjacency matrix calculation layer, a graph convolutional feature extraction layer, a feature fusion layer, and an output layer. The input layer simultaneously imports features from the original grayscale CT image and the constructed hybrid topology graph data. The dynamic adjacency matrix calculation layer reads the vessel diameter and spatial distance parameters of each node in real time. For main vessel nodes with a diameter greater than 3 mm, the neighborhood weight coefficient is set to 0.8-1.0 to enhance the main structural features; for medium branch nodes with a diameter of 0.5-3 mm, the weight coefficient is set to 0.4-0.8; for microvascular nodes with a diameter less than 0.5 mm, the weight coefficient is set to 0.1-0.4 to refine the extraction of subtle branch features; the weight of organ core nodes is relatively high, while the weight of irregular edge nodes is adaptively fine-tuned.

[0027] The graph convolutional layer traverses all topological nodes and edges based on a dynamic weight matrix to extract global topological features. The feature fusion layer concatenates and fuses the grayscale texture features of the two-dimensional image with the three-dimensional topological connectivity features, outputting a 128-dimensional multi-scale fused feature vector.

[0028] III. Initial Reconstruction of 3D Surface The multi-scale fused feature vector output by the network is input into the Marching Cubes reconstruction module. The voxel resolution is set to 0.5 mm. Through voxel traversal, isosurface extraction, and triangular patch fitting, an initial three-dimensional mesh model of the lung organs and blood vessels is generated.

[0029] At this point, the initial model can completely restore the overall shape, but there are a few common artifacts such as broken microvessels, tiny burrs on the surface of organs, and local voxel cavities, which provide a basic model for subsequent optimization and repair.

[0030] IV. Topology Optimization and Automatic Mesh Repair The topology optimization and repair module is activated. Using the standard connected topology of the hybrid topology graph as the verification benchmark, the initial mesh model is globally traversed for verification and repair. For the broken blood vessel nodes identified by the topology verification, the missing branches are automatically filled in based on the spatial distribution and topological association of adjacent nodes, ensuring the continuity of blood vessel connectivity. For closed void regions inside the model, a topological voxel filling algorithm is used to close and repair the voids; for burrs and protrusions on the surface of organs and blood vessel walls, a feature-preserving adaptive smoothing algorithm is used to remove noise protrusions and smooth surface textures while preserving structural edge features. Finally, topological constraint calibration is performed on the junction area between lung organs and blood vessels to eliminate boundary misalignment and structural separation problems, and output the final high-precision three-dimensional reconstruction model.

[0031] V. Verification of Implementation Results Verified by multiple sets of clinical CT images, the lung organ and blood vessel models reconstructed by this method have a microvascular branch preservation rate that is more than 35% higher than that of traditional CNN algorithms. There are no breaks or cavitation artifacts, the smoothness of the organ surface is significantly improved, and the morphological reconstruction error is controlled within 0.2mm, which fully meets the high-precision requirements of lung surgery planning, lesion localization, and medical simulation.

[0032] Meanwhile, this implementation method operates entirely automatically without human intervention, and its reconstruction efficiency is significantly better than that of traditional technologies.

[0033] In addition to lung imaging, the technical solution of this invention can be directly adapted to medical image reconstruction of multiple sites such as the heart, liver, kidneys, and thoracic and abdominal blood vessels. High-precision reconstruction can be achieved simply by fine-tuning the topology node thresholds and weight parameters according to the target anatomical structure, making it highly versatile.

[0034] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for topology optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks, characterized in that, Includes the following steps: Step 1: Acquire CT images of the target body part, segment the CT images to obtain binary segmentation masks corresponding to organs and blood vessels, extract the anatomical skeleton points of the binary segmentation masks, and construct organ-blood vessel hybrid topology maps based on the anatomical skeleton points. Step 2: Construct an adaptive dynamic adjacency matrix graph convolutional network, dynamically adjust the network node weights based on blood vessel diameter and organ spatial distance, fuse CT image grayscale features and topology graph connectivity features, and extract multi-scale topology fusion features; Step 3: Input the multi-scale topology fusion features into the 3D reconstruction module to generate the initial 3D mesh surface model of the organ and blood vessel; Step 4: Verify and repair the initial 3D mesh surface model based on topological connectivity, complete hole filling, branch completion, and surface smoothing, constrain the topological continuity of the organ and blood vessel junction, and output a high-precision 3D reconstruction model.

2. The method for topological optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks according to claim 1, characterized in that, In step one, the method for constructing the organ-blood vessel hybrid topology map includes: The binary segmentation mask image is denoised and morphologically optimized to remove isolated noise points; the organ contour skeleton and the blood vessel midline skeleton are extracted, and several three-dimensional skeleton feature nodes are obtained by uniform sampling; each skeleton feature node is used as a topological node, and the three-dimensional spatial adjacency relationship and anatomical connectivity relationship between nodes are used as topological edges to quantify the spatial distance and structural correlation of nodes, and to construct a hybrid topological map covering the global structure of organs and blood vessels.

3. The method for topological optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks according to claim 1, characterized in that, In step two, the rules for dynamically adjusting the weights of network nodes include: For large blood vessel trunks and core nodes of major organs, the weight of the neighborhood receptive field is increased to enhance the learning of main structural features; for microvascular branches and organ edge nodes, the weight of topological features is refined to accurately capture subtle anatomical structures; and node weights are updated in real time through a dynamic adjacency matrix to adapt to the boundary features of blood vessels and irregular organs at multiple scales.

4. The method for organ and blood vessel topology optimization surface reconstruction driven by dynamic graph convolutional networks according to claim 1, characterized in that, In step two, the extraction method of the multi-scale topological fusion features is as follows: simultaneously extract the two-dimensional grayscale texture features of the CT image and the three-dimensional global connectivity features of the topological map, and obtain a fusion feature vector that has both local details and global topological correlation through feature splicing and weighted fusion.

5. The method for organ and blood vessel topology optimization surface reconstruction driven by dynamic graph convolutional networks according to claim 1, characterized in that, In step three, the three-dimensional reconstruction module adopts the Marching Cube algorithm to complete the initial reconstruction of the three-dimensional surface of organs and blood vessels through voxel traversal, isosurface extraction, and triangular patch fitting.

6. The method for topology optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks according to claim 1, characterized in that, In step four, the topological connectivity verification and repair includes: identifying model vessel rupture, internal cavities, surface burrs, and structural separation artifacts based on standard anatomical connectivity rules of the hybrid topological graph; completing ruptured vessel branches through interpolation algorithms, repairing cavities through topological constraint filling algorithms, removing surface burrs through feature-preserving smoothing algorithms, calibrating the boundaries between organs and vessels, and ensuring overall topological continuity.

7. The method for topology optimization and surface reconstruction of organs and blood vessels driven by dynamic graph convolutional networks according to claim 1, characterized in that, The method is applicable to the three-dimensional reconstruction of CT images of the lungs, heart, thoracic and abdominal organs, and human blood vessels, and can be applied to preoperative surgical planning, medical three-dimensional visualization, teaching simulation, and precise lesion localization.

Images