Cerebrovascular three-dimensional model construction method and system based on cerebrovascular image segmentation

Abstract

The invention relates to the technical field of image processing, in particular to a cerebrovascular three-dimensional model construction method and system based on cerebrovascular image segmentation. The method comprises the following steps: acquiring an initial cerebrovascular image set; registering the initial cerebrovascular image set to obtain a cerebrovascular image set; performing weight mapping feature level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; performing multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; performing spatial attention capillary segmentation on the enhanced cerebrovascular feature map to obtain a capillary segmentation feature map; performing vascular morphological constraint fusion on the large vessel segmentation feature map and the capillary segmentation feature map to obtain a complete cerebrovascular mask map; and performing geodesic distance field calculation on the complete cerebrovascular mask image to obtain a cerebrovascular center line candidate point set. According to the invention, the accuracy and efficiency of cerebrovascular image segmentation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for constructing a cerebral blood vessel three-dimensional model based on cerebral blood vessel image segmentation. Background Art

[0002] In the field of image processing technology, especially in medical image analysis and diagnosis, the construction of cerebral vascular 3D models is an important research direction. However, existing cerebral vascular 3D reconstruction technology faces some challenges, especially in terms of accuracy and efficiency. The cerebral vascular system is composed of a large number of arteries, veins and capillaries, which are distributed throughout the brain in a highly branched and intertwined manner. This complex network structure makes it extremely difficult to accurately identify and segment each blood vessel in medical images. At the same time, the diameter, direction and degree of curvature of cerebral blood vessels vary. Some blood vessels are very small and difficult to detect, while others may have abnormal morphology due to lesions (such as aneurysms). This morphological diversity further increases the difficulty of segmentation. In practical application scenarios, such as the diagnosis of cerebral vascular malformations or aneurysms, accurate vascular segmentation is crucial for the subsequent 3D model construction. If the segmentation result is inaccurate, the reconstructed 3D model will deviate from the actual anatomical structure. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a method and system for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation comprises the following steps: Step S1: obtaining an initial cerebrovascular image set; registering the initial cerebrovascular image set to obtain a cerebrovascular image set; performing weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; Step S2: performing multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; performing spatial attention microvascular segmentation on the enhanced cerebrovascular feature map to obtain a microvascular segmentation feature map; performing vascular morphology-constrained fusion on the large vessel segmentation feature map and the microvascular segmentation feature map to obtain a complete cerebrovascular mask map; Step S3: performing geodesic distance field calculation on the complete cerebral vascular mask image to obtain a set of candidate points for the cerebral vascular centerline; performing fast marching tree structure construction on the set of candidate points for the cerebral vascular centerline to obtain an initial cerebral vascular tree model; performing vascular bifurcation point optimization on the initial cerebral vascular tree model to obtain a complete cerebral vascular network topology map; Step S4: performing double ellipse cross-section fitting modeling on the cerebral blood vessels based on the complete cerebral vascular network topology to obtain a standard cerebral blood vessel cross-section model; performing active contour caliber evaluation on the standard cerebral blood vessel cross-section model to obtain cerebral blood vessel caliber evaluation data; performing implicit surface reconstruction on the cerebral blood vessels based on the cerebral blood vessel caliber evaluation data to obtain an optimized cerebral blood vessel surface model; Step S5: positioning the vascular lesion area on the optimized cerebrovascular surface model to obtain a cerebrovascular lesion area positioning map; classifying the lesions on the cerebrovascular lesion area positioning map to obtain lesion type label data; reconstructing the lesion mesh on the optimized cerebrovascular surface model according to the lesion type label data to obtain a complete cerebrovascular network three-dimensional model.

[0005] The present invention can enhance the cerebrovascular feature map by adopting weight mapping feature level fusion, so that both large and small blood vessels can be effectively segmented. This method is suitable for processing multimodal image data, and can make full use of complementary information in different image modalities, thereby improving the accuracy of blood vessel segmentation, especially in the intersection and stenosis of blood vessels. By performing multi-scale large blood vessel segmentation and spatial attention microvascular segmentation on the enhanced cerebrovascular feature map, the microvascular structure in the cerebrovascular network can be captured, which is often difficult to achieve in previous technologies. This accurate capture of the microvascular structure makes the reconstructed three-dimensional model more complete and refined. By constructing a fast-moving tree structure and optimizing the vascular bifurcation point, the cerebrovascular network topology map is effectively constructed. This not only improves the speed of model construction, but also ensures the correctness of the topological structure of the model. By locating the vascular lesion area and classifying the lesions on the optimized cerebrovascular surface model, the cerebrovascular lesion area can be accurately identified and classified. Through lesion grid reconstruction, a complete cerebrovascular network three-dimensional model containing lesion information is obtained. This can not only show the anatomical structure of the cerebrovascular, but also reflect the pathological changes of the blood vessels. In summary, the present invention not only improves the accuracy and efficiency of cerebrovascular image segmentation, but also enhances the detail capture capability of the cerebrovascular three-dimensional model.

[0006] Preferably, the present invention further provides a cerebrovascular three-dimensional model construction system based on cerebrovascular image segmentation, which is used to execute the cerebrovascular three-dimensional model construction method based on cerebrovascular image segmentation as described above, and the cerebrovascular three-dimensional model construction system based on cerebrovascular image segmentation comprises: The feature enhancement module is used to obtain an initial cerebrovascular image set; register the initial cerebrovascular image set to obtain a cerebrovascular image set; perform weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; The vascular segmentation module is used to perform multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; perform spatial attention microvascular segmentation on the enhanced cerebrovascular feature map to obtain a microvascular segmentation feature map; perform vascular morphology constraint fusion on the large vessel segmentation feature map and the microvascular segmentation feature map to obtain a complete cerebrovascular mask map; The topology construction module is used to calculate the geodesic distance field of the complete cerebrovascular mask map to obtain the candidate point set of the cerebrovascular centerline; to construct a fast marching tree structure on the candidate point set of the cerebrovascular centerline to obtain an initial cerebrovascular tree model; to optimize the vascular bifurcation points of the initial cerebrovascular tree model to obtain a complete cerebrovascular network topology map; The surface model building module is used to perform double elliptical cross-section fitting modeling on cerebral blood vessels based on the complete cerebral vascular network topology map to obtain a standard cerebral vascular cross-section model; perform active contour caliber evaluation on the standard cerebral vascular cross-section model to obtain cerebral vascular caliber evaluation data; perform implicit surface reconstruction on cerebral blood vessels based on the cerebral vascular caliber evaluation data to obtain an optimized cerebral vascular surface model; The model reconstruction module is used to locate the vascular lesion area on the optimized cerebrovascular surface model to obtain a cerebrovascular lesion area positioning map; classify the lesions on the cerebrovascular lesion area positioning map to obtain lesion type label data; reconstruct the lesion grid of the optimized cerebrovascular surface model according to the lesion type label data to obtain a complete cerebrovascular network three-dimensional model.

[0007] The present invention uses a feature enhancement module to perform precise registration and weight mapping feature-level fusion on the image set, thereby enhancing the recognition of cerebrovascular features and improving the accuracy of vascular segmentation. The vascular segmentation module can simultaneously process the segmentation of large vessels and microvessels, and through multi-scale and spatial attention mechanisms, it ensures that the structural information of vessels of different sizes is effectively captured. The topology map construction module accurately restores the centerline and bifurcation points of cerebrovascular vessels through geodesic distance field calculation and fast marching tree structure construction, providing a basis for constructing an accurate cerebrovascular network topology map. The surface model construction module adopts double elliptical cross-section fitting modeling to standardize the shape of the cerebrovascular cross-section, making the model more consistent with the actual anatomical structure and improving the biological authenticity of the model. Through active contour evolution technology, the caliber of cerebrovascular vessels can be accurately evaluated. By adopting implicit surface reconstruction, the smoothness and detail expression of the cerebrovascular surface model are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Other features, objects and advantages of the present invention will become more apparent from the detailed description made with reference to the following drawings: Figure 1 A schematic flow chart of the steps of a method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to an embodiment is shown.

[0009] Figure 2 A detailed flowchart diagram of step S13 of an embodiment is shown.

[0010] Figure 3 A detailed flow chart of step S24 of an embodiment is shown. DETAILED DESCRIPTION

[0011] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0012] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0013] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0014] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation, comprising the following steps: Step S1: obtaining an initial cerebrovascular image set; registering the initial cerebrovascular image set to obtain a cerebrovascular image set; performing weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; Step S2: performing multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; performing spatial attention microvascular segmentation on the enhanced cerebrovascular feature map to obtain a microvascular segmentation feature map; performing vascular morphology-constrained fusion on the large vessel segmentation feature map and the microvascular segmentation feature map to obtain a complete cerebrovascular mask map; Step S3: performing geodesic distance field calculation on the complete cerebral vascular mask image to obtain a set of candidate points for the cerebral vascular centerline; performing fast marching tree structure construction on the set of candidate points for the cerebral vascular centerline to obtain an initial cerebral vascular tree model; performing vascular bifurcation point optimization on the initial cerebral vascular tree model to obtain a complete cerebral vascular network topology map; Step S4: performing double ellipse cross-section fitting modeling on the cerebral blood vessels based on the complete cerebral vascular network topology to obtain a standard cerebral blood vessel cross-section model; performing active contour caliber evaluation on the standard cerebral blood vessel cross-section model to obtain cerebral blood vessel caliber evaluation data; performing implicit surface reconstruction on the cerebral blood vessels based on the cerebral blood vessel caliber evaluation data to obtain an optimized cerebral blood vessel surface model; Step S5: positioning the vascular lesion area on the optimized cerebrovascular surface model to obtain a cerebrovascular lesion area positioning map; classifying the lesions on the cerebrovascular lesion area positioning map to obtain lesion type label data; reconstructing the lesion mesh on the optimized cerebrovascular surface model according to the lesion type label data to obtain a complete cerebrovascular network three-dimensional model.

[0015] In this embodiment, first, a medical imaging scanning device (such as CT, MRI) is used to obtain the patient's initial cerebrovascular image set, including digital subtraction angiography (DSA), tomographic angiography (CTA) and magnetic resonance angiography (MRA) images. These images are stored in DICOM format and imported into professional medical image processing software, such as 3D Slicer. In 3D Slicer, these images are registered to ensure that they are aligned in the same coordinate system. The registration process involves rigid registration (such as translation and rotation) and non-rigid registration (such as elastic deformation) to handle deformation caused by patient movement or equipment differences. The registered images are sent to the feature-level fusion module, which uses a weighted method to combine the advantages of different imaging modalities to generate an enhanced cerebrovascular feature map. Next, a deep learning method, especially a convolutional neural network (CNN), is used to perform multi-scale large vessel segmentation and spatial attention microvascular segmentation on the enhanced cerebrovascular feature map. The CNN model is trained on a large amount of labeled data to identify and segment vascular structures of different scales. The segmentation results are fused to form a complete cerebrovascular mask map. The geodesic distance field of the complete cerebrovascular mask was calculated, and the bwdist function in the image processing software MATLAB was used to obtain the candidate point set of the cerebrovascular centerline. These point sets were used to construct a fast marching tree structure, and the initial cerebrovascular tree model was obtained by the bwtraceboundary function of MATLAB. Furthermore, the initial cerebrovascular tree model was optimized for vascular bifurcation points, and the bifurcation points were identified and refined in MATLAB using graph theory algorithms to obtain a complete cerebrovascular network topology map. Based on this topology map, the standard cerebrovascular cross-section model was obtained by nonlinear least squares fitting using the double ellipse section fitting modeling method. The standard cerebrovascular cross-section model was evaluated for active contour caliber, and the activecontour function of MATLAB was used to obtain cerebrovascular caliber evaluation data. These data were used for implicit surface reconstruction, and the optimized cerebrovascular surface model was obtained by the isosurface function of MATLAB. Finally, the optimized cerebrovascular surface model was used to locate the vascular lesion area, and the lesion area was classified using a machine learning classifier such as a support vector machine (SVM) to obtain lesion type label data. Based on these data, three-dimensional reconstruction techniques such as stereolithography (SLA) or selective laser melting (SLM) are used to reconstruct the lesion mesh of the optimized cerebrovascular surface model to obtain a complete three-dimensional model of the cerebrovascular network.

[0016] Preferably, step S1 comprises the following steps: Step S11: obtaining a cerebral vascular imaging projection angle, and performing angiography scanning on the patient according to the cerebral vascular imaging projection angle to obtain an initial cerebral vascular image set; wherein the initial cerebral vascular image set includes a digital subtraction cerebral vascular angiography image, a tomography cerebral angiography image, and a magnetic resonance angiography image; Specifically, a CT machine or MRI machine with high-resolution imaging capabilities can be used to obtain the projection angle of the patient's cerebrovascular imaging. For example, a 64-row CT scanner is used, and the imaging plane is parallel to the patient's basilar cerebral artery by adjusting the angle setting of the scanner. Under this setting, the patient is subjected to a vascular angiography scan. During the scan, a series of continuous two-dimensional image slices are collected, which together constitute the initial cerebrovascular image set. Specifically, the image set obtained contains three types of images: digital subtraction cerebral angiography (DSA) images, which are obtained through digital subtraction technology and can clearly display the vascular structure; tomographic cerebral angiography (CTA) images, which are obtained through CT scanning and provide cross-sectional images of blood vessels; magnetic resonance angiography (MRA) images, which are obtained through MRI scanning and combined with specific vascular flow velocity sensitive sequences, can display the three-dimensional structure of blood vessels.

[0017] Step S12: performing vascular tree segmentation and enhancement on the digital subtraction cerebral angiography image to obtain an enhanced digital subtraction vascular tree map; Specifically, professional medical image processing software, such as 3D Slicer, can be used to load DSA images. A segmentation algorithm based on graph cuts is applied to segment the vascular tree. This algorithm requires selecting several points on the inner and outer boundaries of the vascular tree as seed points, and then the algorithm divides the blood vessels and background according to local image features and global image information. After the segmentation is completed, the segmentation results are post-processed, including morphological operations (such as opening and closing operations) to remove small noise and fill small holes inside the blood vessels, as well as edge enhancement to further improve the clarity of the blood vessel contours. Finally, an enhanced digital subtraction vascular tree image is obtained, which clearly shows the structure of the vascular tree.

[0018] Step S13: performing voxel density adaptive segmentation on the tomographic cerebral angiography image to obtain a segmented tomographic vascular structure map; Specifically, please refer to the sub-steps of step S13 for the detailed implementation process of this embodiment.

[0019] Step S14: performing phase contrast enhancement on the magnetic resonance angiography image to obtain an enhanced magnetic resonance vascular network map; Specifically, a 3.0T MRI scanner can be used to obtain magnetic resonance angiography images of patients. A technique called "phase contrast MRA" is used, which uses the phase changes caused by blood flowing in blood vessels to enhance vascular signals. Set the MRI scanning parameters and select the appropriate flip angle and echo time (TE). For example, a flip angle of 20 degrees and an echo time of 30 milliseconds are selected. During the scan, two sets of data are collected: one set is flow compensated (Flow Compensated) images, and the other set is non-flow compensated (Non-Flow Compensated) images. Flow compensated images are used to reduce phase artifacts caused by blood flow, while non-flow compensated images are used to capture phase changes caused by blood flow. Use professional medical image analysis software, such as FSL (FMRIB Software Library), to process these images. Apply a phase contrast enhancement algorithm, which extracts phase change information caused by blood flow by comparing flow compensated and non-flow compensated images. Specifically, the phase difference between the two sets of images is calculated and mapped into a single image, which highlights the vascular structure because blood flow in the blood vessels will produce high signals in the phase difference image. The final result is an enhanced magnetic resonance vascular network image, which clearly shows the structure of the vascular network, including large and small blood vessels.

[0020] Step S15: performing deformation field calculation on the enhanced digital subtraction vascular tree map, the segmented tomography vascular structure map and the enhanced magnetic resonance vascular network map to obtain a multimodal image deformation mapping field, and performing elastic deformation registration on the enhanced digital subtraction vascular tree map, the segmented tomography vascular structure map and the enhanced magnetic resonance vascular network map according to the multimodal image deformation mapping field to obtain a cerebral vascular image set; wherein the cerebral vascular image set includes the registered digital subtraction vascular tree map, the registered tomography vascular structure map and the registered magnetic resonance vascular network map; Specifically, medical image processing software such as Elastix, an open source tool for medical image registration, can be used to process enhanced digital subtraction angiograms (DSA), segmented tomographic angiograms (CTA), and enhanced magnetic resonance angiograms (MRA). Elastix supports multiple types of registration, including rigid, affine, and non-rigid (elastic) registration. First, load the three image datasets in Elastix. MRA is selected as the reference image because it provides clear vascular structures and high soft tissue contrast. For DSA and CTA images, use Elastix's non-rigid registration module and select the B-spline transformation model to calculate the deformation field. The B-spline model is selected for its ability to control local deformation. Set the registration optimization method to mutual information, which is a common metric used to evaluate the similarity of two image data, especially when their contents are not exactly aligned. Run the registration process and Elastix will automatically calculate the deformation field from one image to the other. This process involves iterative optimization until the deformation field is stable, ensuring that the vascular structures in the DSA and CTA images are aligned with the corresponding structures in the MRA image. Once the deformation fields are calculated, they are used to elastically register the DSA and CTA images. This process involves applying the calculated deformations to the original DSA and CTA images to match their spatial layout to the MRA images. Using Elastix’s Transformation module, the deformation fields are applied to the DSA and CTA images. During this process, the pixels of each image are translated, scaled and rotated to match the reference MRA image as guided by the deformation field. The registered images are saved and a registered digital subtraction vessel tree map (registered DSA) and a registered tomographic vessel structure map (registered CTA) are obtained. The registered DSA and CTA images are combined with the original MRA images to obtain a set of cerebrovascular images of all three modalities, which are now in the same coordinate system.

[0021] Step S16: performing weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map.

[0022] Specifically, please refer to the sub-steps of step S16 for the detailed implementation process of this embodiment.

[0023] The present invention realizes the complementary advantages of multimodal images by integrating digital subtraction cerebral angiography images, tomographic cerebral angiography images and magnetic resonance angiography images, and significantly improves the recognition accuracy of cerebrovascular structure, especially in capturing vascular details and lesion characteristics. By performing specific enhancement processing on each imaging modality, such as vascular tree segmentation enhancement, voxel density adaptive segmentation and phase contrast enhancement, the image quality is significantly improved, making the vascular structure clearer. Through deformation field calculation and elastic deformation registration, accurate alignment between images of different modalities is achieved, ensuring the consistency of image data in space, and further improving the accuracy of the three-dimensional reconstruction model. Finally, the weight mapping feature-level fusion step enables the image data of different modalities to complement each other, enhances the information content of the cerebrovascular feature map, especially in the areas of vascular intersection and stenosis, and improves the accuracy of vascular segmentation.

[0024] Preferably, step S13 comprises the following steps: Step S131: performing multi-peak density detection on the tomographic cerebral angiography image to obtain a cerebral vascular density distribution map; Specifically, the CTA image sequence can be imported using the medical image processing software ITK-SNAP. The software is capable of reading image data in DICOM format and converting it into grayscale images. Histogram analysis is applied to each CTA image, and the density peaks of different tissue types are identified using the histogram tool of ITK-SNAP. By observing the histogram, the peaks representing blood vessels are manually marked, which are usually located in higher density areas. The software automatically calculates the density values ​​of these peaks and generates a cerebral vascular density distribution map containing all marked peaks, which shows the distribution of blood vessels at different density levels.

[0025] Step S132: Calculating the blood vessel density threshold for the tomographic cerebral angiography image based on the cerebral blood vessel density distribution map to obtain a blood vessel density segmentation threshold set; Specifically, the MATLAB software can be used to open the cerebral vascular density distribution map. The graythresh function in the image processing toolbox of MATLAB is used to automatically calculate a global threshold based on the Otsu method to distinguish between vascular and non-vascular areas. The imthreshold function is used to apply this global threshold to generate a binary image in which the vascular area is marked as white and the non-vascular area is marked as black. In order to improve the accuracy of segmentation, the adaptthresh function can be used to dynamically adjust the threshold according to the density distribution in each small area in combination with the local threshold adjustment. Finally, a vascular density segmentation threshold set containing all CTA images is obtained.

[0026] Step S133: extracting voxel cube neighborhood from the tomographic cerebral angiography image to obtain three-dimensional voxel neighborhood data; Specifically, CTA image sequences can be imported into 3D Slicer and loaded into the software's workspace. Use the software's volume rendering tool to view and analyze image data. Use the "Label Statistics" tool in 3D Slicer, which allows the definition of a cubic region and calculates the statistics of the voxels within the region. Manually define a cubic region centered on a blood vessel, or use the software's automatic segmentation tool to assist in determining this region. Set the side length of the cube, for example to 3 voxels, to ensure that sufficient surrounding tissue information is included. 3D Slicer automatically calculates and extracts all voxel data within this cube, including their density values ​​and spatial positions, to obtain three-dimensional voxel neighborhood data.

[0027] Step S134: performing density-connected region growing segmentation on the tomographic cerebral angiography image based on the blood vessel density segmentation threshold set and the three-dimensional voxel neighborhood data to obtain an initial cerebral blood vessel segmentation map; Specifically, the 3D voxel neighborhood data and the vessel density segmentation threshold set can be imported into ITK-SNAP. The "Region Growing" tool of the software is used to perform region growing segmentation. This tool allows the definition of a starting point (seed point), and a voxel in the center of the vessel is selected as the starting point. The parameters of the region growing are set, including the density threshold and the connectivity criterion. The density threshold is the vessel density segmentation threshold set, which is input into ITK-SNAP as the density range for region growing. The connectivity criterion defines the conditions under which voxels are considered connected, and 6-neighbor connectivity is selected, which means that if the six facet neighbor voxels of a voxel meet the density threshold condition, they will be included in the segmentation region. After starting the region growing algorithm, ITK-SNAP will start from the seed point and gradually add neighboring voxels that meet the density threshold condition to the segmentation region until no more voxels meet the condition. This process generates an initial cerebral vascular segmentation map in which the vascular structures are marked.

[0028] Step S135: performing vascular structure morphology screening on the initial cerebral vascular segmentation image to obtain a set of cerebral vascular candidate regions; Specifically, the image processing software FIJI (an open source software based on ImageJ) can be used to perform morphological screening of vascular structures. The initial cerebral vascular segmentation map is imported into FIJI and operated using its morphological processing tools. First, a morphological opening operation is applied, which involves the selection of a structural element, for example, a 3x3 rectangular structural element. This operation helps to remove small noise points and separate contacting objects. A morphological closing operation is performed, using a slightly larger structural element, such as a 5x5 rectangle, to fill small holes inside the vascular structure and connect adjacent vascular segments. Through these operations, vascular structures that meet specific morphological characteristics are screened out, such as larger, continuous vascular structures are retained, while small, isolated noise is removed. In this way, a set of cerebral vascular candidate regions is obtained.

[0029] Step S136: locating the vascular structure boundary of the initial cerebral vascular segmentation map according to the cerebral vascular candidate region set to obtain a cerebral vascular boundary positioning map, and performing density distribution mapping on the cerebral vascular boundary positioning map to obtain a segmented tomographic vascular structure map.

[0030] Specifically, FIJI's "edge detection" function, such as the Sobel operator, can be used to identify the boundaries of the set of cerebrovascular candidate regions. This operator can highlight the edges in the image, thereby helping to determine the location of the vascular structure. After edge detection, a cerebrovascular boundary localization map is obtained, which shows the approximate outline of the vascular structure. Next, use FIJI's "density slicing" tool to generate multiple binary images by setting different density thresholds, each corresponding to a specific density range. The distribution of vascular structures at different density levels can be obtained. Finally, these binary images are merged to form a complete segmented tomographic vascular structure map.

[0031] The present invention accurately captures the density distribution of cerebral blood vessels through multi-peak density detection, providing key density information for blood vessel segmentation. Automated blood vessel density threshold calculation reduces manual intervention, enhances the efficiency and consistency of the segmentation process, and ensures the comparability of segmentation results between different image data. By extracting the voxel cubic neighborhood, more detailed three-dimensional voxel neighborhood data can be obtained to accurately depict the three-dimensional structure of cerebral blood vessels. The blood vessels and non-vascular areas are effectively distinguished by the density-connected regional growth segmentation method, and an accurate initial cerebral blood vessel segmentation map is generated. The recognition of vascular candidate areas is further optimized by morphological screening, and the accuracy and reliability of blood vessel segmentation are improved. The cerebral blood vessel boundaries can be determined by precise blood vessel boundary positioning and density distribution mapping. The present invention significantly improves the segmentation quality of tomographic cerebral angiography images.

[0032] Preferably, step S16 comprises the following steps: Step S161: respectively evaluating the vascular clarity of the registered digital subtraction vascular tree map, the registered tomographic vascular structure map, and the registered magnetic resonance vascular network map to obtain the digital subtraction vascular clarity, the tomographic vascular clarity, and the magnetic resonance vascular clarity; Specifically, image analysis software, such as MATLAB, can be used to process the registered DSA images. DSA images are imported into MATLAB, and MATLAB's edge function, such as the Canny edge detection algorithm, is used to identify the edges of blood vessels. By calculating the proportion of edge pixels in the edge detection results, a quantitative index of blood vessel clarity can be obtained. The average contrast between blood vessels and background can also be calculated, which can be achieved by measuring the gray value difference between blood vessel pixels and surrounding background pixels. These indicators together constitute the blood vessel clarity assessment results of DSA images. For CTA images, FIJI software is used for blood vessel clarity assessment. CTA images are imported into FIJI, and its threshold segmentation tool is used to separate blood vessels and background. Then, the "Analyze Particles" function of FIJI is used to measure the area, perimeter, and shape parameters of the blood vessel region, such as circularity and ellipticity. For example, a blood vessel with higher clarity will have a higher area to perimeter ratio and shape parameters close to a circle. Through these measurement results, the blood vessel clarity assessment of CTA images can be obtained. For MRA images, 3D Slicer software is used for blood vessel clarity assessment. Import MRA images into 3D Slicer and use its volume rendering tool to visually observe the vascular structure. Use 3D Slicer's "Label Statistics" tool to calculate the signal intensity distribution of the vascular region, including the mean, standard deviation, and signal intensity range. You can also calculate the average signal intensity ratio of the blood vessel to the background, which can be achieved by comparing the signal intensity of the vascular region with the surrounding background area. These signal intensity parameters provide important information about the clarity of the blood vessels. By combining these parameters, the vascular clarity evaluation results of the MRA image can be obtained.

[0033] Step S162: normalizing and weighting the digital subtraction blood vessel clarity, the tomography blood vessel clarity and the magnetic resonance blood vessel clarity to obtain a cerebrovascular structure feature weight set; wherein the cerebrovascular structure feature weight set includes the digital subtraction image feature weight, the tomography image feature weight and the magnetic resonance image feature weight; Specifically, MATLAB software can be used to process the data of digital subtraction vessel clarity. The vessel clarity indicators of DSA images, such as the edge pixel ratio and the average contrast between the blood vessels and the background, are calculated. These indicators are normalized using the minmax function of MATLAB to convert the raw data into a range of 0 to 1. Next, weights are assigned according to the importance of vessel clarity. For example, if the edge pixel ratio is considered more important than the contrast, a weight of 0.6 is assigned to the edge pixel ratio, while a weight of 0.4 is assigned to the contrast. For CTA images, FIJI software is used to calculate vessel clarity indicators, such as the area to perimeter ratio and shape parameters. Then, these indicators are imported into MATLAB for normalization. Using the minmax function of MATLAB, the vessel clarity indicators of CTA images are normalized to a range of 0 to 1. In terms of weight assignment, the area to perimeter ratio is considered more critical for evaluating vessel clarity, so a higher weight, such as 0.7, is given, while the shape parameter is assigned a weight of 0.3. For MRA images, 3D Slicer software is used to calculate signal intensity parameters, and these data are exported to MATLAB for further processing. In MATLAB, the minmax function was used to normalize the vascular clarity index of the MRA image. In terms of weight allocation, the signal intensity range is considered to be the most important for evaluating vascular clarity, so it is given a weight of 0.5, while the average signal intensity ratio is assigned a weight of 0.5. Through the above steps, a cerebrovascular structure feature weight set including DSA, CTA and MRA image feature weights is obtained.

[0034] Step S163: performing wavelet domain weighted enhancement on the registered digital subtraction vascular tree map according to the digital subtraction image feature weights to obtain a weighted digital subtraction vascular feature map, and performing pyramid domain weighted enhancement on the registered tomography vascular structure map according to the tomography image feature weights to obtain a weighted tomography vascular feature map; performing Gabor domain weighted enhancement on the registered magnetic resonance vascular network map according to the magnetic resonance image feature weights to obtain a weighted magnetic resonance vascular feature map; Specifically, the DSA image can be read and the wavelet transform function wavedec2 in the image processing toolbox of MATLAB is used to perform a two-dimensional discrete wavelet transform on the image to decompose it into sub-band images of different scales. Select a suitable wavelet basis, such as db4 (Daubechies wavelet), and determine the number of decomposition layers, such as 2 layers. Next, each sub-band image is weighted enhanced according to the digital subtraction image feature weight determined in the previous step. For example, if the weight of the vascular edge is high, the sub-band image corresponding to the high-frequency information will be enhanced to highlight the details of the vascular edge. Threshold processing is applied using the wthresh function to suppress noise and retain important vascular features. Then, the weighted sub-band image is inversely transformed using the waverec2 function to reconstruct the weighted digital subtraction vascular feature map. For the registered tomographic vascular structure image (CTA), pyramid domain weighted enhancement is performed using FIJI software. First, the CTA image is opened in FIJI and the image is Gaussian blurred using the "Gaussian Blur" tool under the "Filters" menu to reduce noise. Then, a Laplacian pyramid was constructed using the “Laplace” option in the “Stacks” tool under the “Tools” menu of FIJI. Next, each layer of the pyramid was weighted according to the weight of the tomographic image features. For example, if the large-area features of the blood vessels are more important, a higher weight would be given to the lower layers of the pyramid (representing large-scale features). The “ImageCalculator” plugin of FIJI was used to merge the weighted pyramid layers and reconstruct the weighted tomographic vascular feature map. For the registered magnetic resonance vascular network map (MRA), Gabor domain weighted enhancement was performed using Python programming language with OpenCV and scikit-image library. First, the MRA image was read and the Gabor filter was generated using the GaborKernel function of OpenCV. Appropriate Gabor filter parameters such as wavelength, direction, and spatial scale were selected according to the weight of the magnetic resonance image features. Then, the Gabor filter was applied to the MRA image using the gabor function in the scikit-image library to extract vascular features of different directions and scales. According to the weight of the magnetic resonance image features, the extracted features were weighted to enhance important vascular structures. Finally, the weighted feature maps are merged to obtain the weighted magnetic resonance vascular feature map, and finally three weighted enhanced image feature maps are obtained.

[0035] Step S164: performing feature fusion on the weighted digital subtraction vascular feature map, the weighted tomography vascular feature map and the weighted magnetic resonance vascular feature map to obtain an enhanced cerebral vascular feature map.

[0036] Specifically, the three weighted image feature maps can be imported into Photoshop as different layers. These layers correspond to the feature maps of DSA, CTA, and MRA images, respectively. In Photoshop, first adjust the transparency of each layer to reflect the relative importance of each imaging modality in the feature fusion. For example, if the edge information of the DSA image is considered to be the most important, set its transparency to 100%, while the transparency of the CTA and MRA images is set to 80% and 60%, respectively. Next, use Photoshop's layer blending mode to further fuse the feature maps. After adjusting the transparency and blending mode, if necessary, you can manually adjust the brightness and contrast of the layer, or use Photoshop's "Curves" tool to fine-tune the contrast of the image. Finally, merge the fused layers into a single enhanced cerebrovascular feature map.

[0037] The present invention can identify the difference in clarity of vascular structures in each image by independently evaluating the vascular clarity of images of different modalities. The feature weight distribution is optimized through normalization and weight distribution steps to ensure that the contributions of different imaging modalities in the feature fusion process are reasonably balanced, thereby enhancing the representativeness of the features and the quality of the overall image. Through weighted enhancement techniques in the wavelet domain, pyramid domain, and Gabor domain, the key vascular features in each imaging modality are highlighted, the details of the vascular structure are enhanced, and the readability of the image is improved. The weighted multimodal image features are integrated through the feature fusion step to generate an enhanced cerebrovascular feature map that combines the advantages of various images.

[0038] Preferably, step S2 comprises the following steps: Step S21: performing pyramid feature recursive extraction on the enhanced cerebrovascular feature map to obtain a multi-level cerebrovascular feature atlas; Specifically, the enhanced cerebrovascular feature map can be imported into the MATLAB environment, and the pyramid construction function pyrdown in its image processing toolbox can be used to recursively construct a Gaussian pyramid of the image. This function reduces the resolution of the image layer by layer through Gaussian filtering and downsampling while retaining the key features of the image. The number of pyramid layers is set to 4, which means that starting from the original image, the resolution will be gradually reduced until four different levels of feature maps are reached. The feature map of each layer is obtained by applying the pyrdown function to the feature map of the previous layer. In this process, multi-scale information from low resolution to high resolution is retained. Through this process, a cerebrovascular feature atlas containing four levels can be obtained, each of which reflects the cerebrovascular structure information at different scales.

[0039] Step S22: extracting encoder features from the multi-level cerebrovascular feature atlas to obtain a multi-level encoded cerebrovascular feature set; Specifically, a convolutional neural network (CNN) model can be defined, which contains multiple convolutional layers and pooling layers. Each convolutional layer uses an appropriate filter size (e.g., 3x3 or 5x5) and depth (i.e., the number of channels of the feature map) to extract features from cerebrovascular feature maps at different levels. Pooling layers (usually maximum pooling) are used to reduce the spatial dimension of the features while increasing the level of abstraction of the features. The multi-level cerebrovascular feature map set is used as input data and input into the CNN model layer by layer. At each level, CNN extracts features through convolution and pooling operations and passes the extracted features to the next level. In this way, a multi-level encoded cerebrovascular feature set is obtained, and the features of each level have been subjected to deep feature extraction by the encoder. During the training process, the cross entropy loss function and the Adam optimizer are used to optimize the weights of the model. After the training is completed, the trained model is used to forward propagate the feature maps of each level to obtain the final multi-level encoded cerebrovascular feature set.

[0040] Step S23: reorganizing the multi-level coded cerebrovascular feature set to obtain a reorganized coded cerebrovascular feature set, and performing decoder feature recovery on the reorganized coded cerebrovascular feature set to obtain a large vessel segmentation feature map; Specifically, the deep learning framework Keras can be used to build an autoencoder network, which consists of two parts: an encoder and a decoder. First, the encoder part of the autoencoder is defined using Keras, which consists of multiple convolutional layers and pooling layers, and a multi-level encoded cerebrovascular feature set has been successfully extracted. Next, these features are reorganized. In the decoder part of the autoencoder, a series of upsampling (transposed convolution) layers and convolutional layers are designed. The upsampling layer is used to increase the spatial resolution of the feature map, while the convolutional layer is used to extract features while increasing the resolution. The feature reorganization process is finely controlled by adjusting the step size and kernel size of the upsampling layer, as well as the number and size of the filters of the convolutional layer. In the specific operation, the spatial resolution of the feature map of the lowest level is increased by an upsampling layer, and then it is spliced ​​or added (feature fusion) with the upper feature map. In this way, the deep features extracted in the encoder are gradually combined with the shallow features to restore a more complete cerebrovascular feature map. The upsampling and feature fusion operations of each layer are carefully designed according to the size and depth of the feature map to ensure the effective fusion of information and the gradual recovery of resolution. After the feature reorganization is completed, the decoder part of the autoencoder is used to restore the reorganized encoded cerebrovascular feature set. The last layer of the decoder is a convolutional layer, which converts the fused multi-layer feature map into the final large vessel segmentation feature map, which has the same resolution as the original input image.

[0041] Step S24: performing spatial attention microvascular segmentation on the enhanced cerebral vascular feature map to obtain a microvascular segmentation feature map; Specifically, please refer to the sub-steps of step S24 for the detailed implementation process of this embodiment.

[0042] Step S25: performing morphological constraint extraction on the large vessel segmentation feature map and the microvessel segmentation feature map to obtain a cerebrovascular morphological constraint set, and performing vascular connectivity detection on the large vessel segmentation feature map and the microvessel segmentation feature map based on the cerebrovascular morphological constraint set to obtain a cerebrovascular connectivity map; Specifically, the macrovascular segmentation feature map and microvascular segmentation feature map can be imported into FIJI, and its morphological toolbox is used to extract the cerebrovascular morphological constraint set. Morphological operations, such as opening and closing operations, are applied to enhance the vascular structure in the image and reduce noise. The opening operation (erosion followed by dilation) is used to remove small noise points and separate the contacting vascular structures, while the closing operation (dilation followed by erosion) is used to fill the small holes inside the vascular structure and connect the adjacent vascular segments. A structural element, such as a 3x3 rectangular or circular kernel, is selected to perform these morphological operations. Through these operations, a cerebrovascular morphological constraint set can be obtained, which contains the macrovascular and microvascular feature maps after morphological processing. Next, a custom algorithm is used to analyze the feature maps in the cerebrovascular morphological constraint set to detect the connectivity between vascular structures. Specifically, each pixel in the cerebrovascular morphological constraint set is traversed and the pixel values ​​in its neighborhood are checked. If a pixel represents a vascular structure and there are other pixels representing vascular structures in its neighborhood, these vascular structures are considered to be connected. All connected vascular structures are recorded and a cerebrovascular connectivity map is constructed, which shows the connectivity relationship between vascular structures. In practice, a threshold is set to determine which pixels are considered vascular structures. The size and shape of the neighborhood are defined, which is usually based on the expected size and morphological characteristics of the blood vessels. In this way, a detailed cerebrovascular connectivity map can be obtained.

[0043] Step S26: morphologically reconstructing the large vessel segmentation feature map and the microvessel segmentation feature map according to the cerebral vascular connectivity map to obtain a complete cerebral vascular mask map.

[0044] Specifically, the cerebrovascular connectivity map, the macrovascular segmentation feature map, and the microvascular segmentation feature map can be imported into ImageJ. In ImageJ, the “Reconstruct” function in the “Morphology” toolbar is used to perform morphological reconstruction. This function allows the segmentation feature map to be reconstructed based on the connectivity map. In the specific operation, the macrovascular segmentation feature map is first selected as the base image, and the cerebrovascular connectivity map is reconstructed as the guide image using the “Reconstruct” function. This process will identify and connect all connected vascular structures in the macrovascular segmentation feature map. Next, the above steps are repeated for the microvascular segmentation feature map to ensure that the microvascular structure is also correctly reconstructed. During the reconstruction process, the size and shape of the structural elements need to be adjusted to match the vascular structures of different sizes. For example, for microvessels, a smaller circular structural element is selected, while for macrovessels, a larger structural element is selected. After the reconstruction is completed, the reconstruction results of macrovessels and microvessels are merged to obtain a complete cerebrovascular mask. This mask accurately shows the structure of the cerebrovascular network, including macrovessels and microvessels.

[0045] The present invention realizes multi-level feature capture through pyramid feature recursive extraction, and comprehensively understands the structure and morphology of cerebral blood vessels, especially at different scales. The information content of the multi-level encoded cerebral vascular feature set is enhanced through deep learning processing of encoders and decoders, and the accuracy of large blood vessel segmentation is improved. The spatial attention microvascular segmentation technology focuses on the precise segmentation of microvascular structure. The connectivity between large and small blood vessels is ensured by strengthening morphological constraints, which is crucial for building a complete cerebral vascular network model. The feature maps of large and micro vessels are integrated through the morphological reconstruction step to generate a complete cerebral vascular mask map.

[0046] Preferably, step S24 comprises the following steps: Step S241: performing multi-channel feature decomposition on the enhanced cerebrovascular feature map to obtain a multi-channel cerebrovascular feature set; Specifically, the enhanced cerebrovascular feature map can be input into a pre-trained convolutional neural network (CNN), which can automatically learn and extract features from images. Select a CNN model with multiple convolutional layers, such as VGG-16 or ResNet-50. In actual operation, the enhanced cerebrovascular feature map is input into the CNN model, and a global average pooling layer is added after each convolutional layer, so that the feature map of each convolutional layer can be converted into a feature vector. These feature vectors represent the features of different channels, which are combined to form a multi-channel cerebrovascular feature set. For example, if a CNN model with 5 convolutional layers is selected, and each convolutional layer is followed by a global average pooling layer, then after inputting an enhanced cerebrovascular feature map, 5 feature vectors are obtained, each feature vector corresponds to the features of a channel, thus forming a multi-channel cerebrovascular feature set.

[0047] Step S242: re-arranging the vascular space geometric feature dimensions of the cerebral blood vessels according to the multi-channel cerebral blood vessel feature set to obtain a re-arranged geometric feature matrix; Specifically, the feature vectors in the multi-channel cerebrovascular feature set can be imported into the Python environment and processed using the NumPy library. These feature vectors are rearranged into a geometric feature matrix, in which rows represent different vascular structures and columns represent different feature channels. In actual operation, the position and size of each structure in the cerebrovascular are first determined, which can be achieved by analyzing the enhanced cerebrovascular feature map or using the cerebrovascular connectivity map obtained in the previous step. Then, according to the spatial geometric features of these structures, such as length, width, curvature, etc., the feature vectors in the multi-channel feature set are rearranged. For example, assuming there are 3 cerebrovascular structures and 5 feature channels, a 3x5 matrix will be created, in which each row corresponds to a cerebrovascular structure and each column corresponds to a feature channel. The order of the feature vectors in the matrix will be determined according to the spatial geometric features of each cerebrovascular structure, such as length and curvature. In this way, a reordered geometric feature matrix is ​​obtained.

[0048] Step S243: performing autocorrelation calculation on the reordered geometric feature matrix to obtain a microvascular feature correlation matrix; Specifically, the reordered geometric feature matrix can be imported into the R environment. This matrix contains multiple features of cerebrovascular structures, each of which is extracted from a multi-channel cerebrovascular feature set. Use R's built-in function cor() to calculate the correlation between each pair of features in the matrix. This function will provide a correlation matrix in which each element represents the correlation coefficient between the corresponding feature pair. In actual operation, each feature in the matrix is ​​compared pairwise to calculate their correlation. For example, if the matrix has 5 features, a 5x5 correlation matrix will be obtained, in which each element is the correlation coefficient between two features.

[0049] Step S244: performing nonlinear mapping on the microvascular feature correlation matrix to obtain a multi-scale attention weight matrix, and performing spatial dependency mapping on the enhanced cerebrovascular feature map according to the multi-scale attention weight matrix to obtain a microvascular spatial attention weight map; Specifically, the microvascular feature correlation matrix can be imported into the PyTorch environment, and a nonlinear activation function such as ReLU (Rectified Linear Unit) is used to nonlinearly map each element in the matrix. This operation will enhance the weights of feature pairs with strong correlation, while suppressing the weights of feature pairs with weak correlation. A custom multi-scale attention mechanism is used to generate a multi-scale attention weight matrix. This mechanism will consider cerebrovascular features at different scales and assign a weight to each feature, which reflects the importance of the feature in the overall cerebrovascular structure. A softmax-based attention layer is used, which can dynamically adjust the weights according to the correlation of the features. Finally, the multi-scale attention weight matrix is ​​applied to enhance the cerebrovascular feature map for spatial dependency mapping. This process involves multiplying the feature vector of each pixel with the corresponding attention weight to highlight important vascular features and suppress unimportant features, and finally obtain the microvascular spatial attention weight map.

[0050] Step S245: performing microvascular structure enhancement on the enhanced cerebrovascular feature map according to the microvascular spatial attention weight map to obtain an enhanced microvascular feature set; Specifically, the microvascular spatial attention weight map and the enhanced cerebrovascular feature map can be imported into MATLAB. Next, the two images are multiplied pixel by pixel using the function immultiply in MATLAB's image processing toolbox. This operation will enhance or weaken the corresponding features in the enhanced cerebrovascular feature map according to the weights in the microvascular spatial attention weight map. Specifically, the features corresponding to the areas with higher values ​​in the weight map will be strengthened in the enhanced cerebrovascular feature set, while the features corresponding to the areas with lower values ​​in the weight map will be weakened. In this way, an enhanced microvascular feature set can be obtained, which highlights the microvascular structure and suppresses unimportant background noise.

[0051] Step S246: reconstructing the enhanced microvascular feature set to obtain a microvascular segmentation feature map.

[0052] Specifically, a threshold operation can be applied to the enhanced microvascular feature set to further separate the microvascular structure and the background. Use the imthreshold function of MATLAB and select an appropriate threshold, which is usually determined based on the intensity distribution of the microvascular structure. The threshold operation divides the pixels in the enhanced microvascular feature set into two categories: pixels above the threshold are considered to be part of the microvascular structure, while pixels below the threshold are considered to be the background. Finally, the imclearborder function is used to remove the microvascular structure close to the edge of the image, ensuring that only the internal microvascular structure is retained. After these processes, the microvascular segmentation feature map is finally obtained, which clearly shows the boundaries and structures of the microvessels.

[0053] The present invention extracts rich vascular feature information from the enhanced cerebrovascular feature map through multi-channel feature decomposition, providing multi-dimensional data support for microvascular analysis. The feature matrix is ​​optimized by rearranging the dimensionality of the vascular space geometric features, so that the spatial structure of the cerebrovascular can be captured more accurately. The intrinsic connection between the microvascular features is strengthened by autocorrelation calculation, making the microvascular network structure clearer. By applying multi-scale attention weights, attention is paid to microvascular features at different scales, and the microvascular structure is highlighted through spatial dependency mapping, further improving the accuracy of segmentation. The microvascular features are made more obvious through microvascular structure enhancement technology, providing a clear feature set, and the feature reconstruction step synthesizes the enhanced microvascular feature set into a microvascular segmentation feature map, providing accurate microvascular structure information for the construction of a cerebrovascular three-dimensional model, thereby significantly improving the accuracy and reliability of microvascular segmentation.

[0054] Preferably, step S3 comprises the following steps: Step S31: performing Euclidean distance transformation on the complete cerebral vascular mask image to obtain an initial cerebral vascular distance field image; Specifically, the complete cerebrovascular mask can be imported into MATLAB, and the bwdist function in its image processing toolbox can be used to calculate the Euclidean distance transform. This function calculates the distance from each background pixel in the mask to the nearest foreground (cerebrovascular) pixel. In the bwdist function, set the parameter 'Type', 'cityblock' to specify the use of city block (Manhattan) distance, which is a form of Euclidean distance and is suitable for distance transform of binary images. After executing the bwdist function, an initial cerebrovascular distance field map is obtained, in which each pixel value represents the distance from the pixel to the nearest cerebrovascular edge.

[0055] Step S32: performing anisotropic diffusion filtering on the complete cerebral vascular mask image based on the initial cerebral vascular distance field image to obtain a smoothed cerebral vascular distance field image; Specifically, the anisdiff function of MATLAB can be used to perform anisotropic diffusion filtering on the initial cerebral vascular distance field map. Set the parameters of the anisdiff function, including the intensity and direction of the diffusion. For example, a weaker diffusion intensity and a direction consistent with the main flow direction of the cerebral blood vessels can be selected to smooth the noise in the distance field map without blurring the edges of the cerebral blood vessels. After executing the anisdiff function, a smoothed cerebral vascular distance field map can be obtained.

[0056] Step S33: Solving the geodesic equation for the smoothed cerebral vascular distance field map to obtain the cerebral vascular geodesic distance field map, and extracting the local maximum of the cerebral vascular structure based on the cerebral vascular geodesic distance field map to obtain a cerebral vascular centerline candidate point set; Specifically, the smoothed cerebrovascular distance field map can be imported into ITK-SNAP and its built-in geodesic distance transform function can be used. This function calculates the distance from each pixel in the image to the nearest vessel boundary by solving the geodesic equation. In ITK-SNAP, select the "Geodesic Distance Transform" tool and set appropriate parameters, such as considering the curvature and local shape of the vessel boundary. After performing the geodesic distance transform, the cerebrovascular geodesic distance field map is obtained. Each pixel value in this image represents the shortest path distance from the point to the nearest vessel boundary. Next, the local maximum value extraction of the cerebrovascular geodesic distance field map is performed using the "Local Maxima" tool of ITK-SNAP. This tool identifies potential candidate points for the vascular centerline by analyzing local changes in the distance field map. Set the threshold parameters of the tool to identify significant local maximum points in the distance field map. After performing the local maximum value extraction, a set of candidate points for the cerebrovascular centerline is obtained. These points are potential centerline locations in the cerebrovascular network.

[0057] Step S34: constructing spatial connectivity of the cerebral vascular centerline candidate point set to obtain a vascular centerline connection relationship diagram; Specifically, a set of candidate points for cerebrovascular centerlines can be imported into FIJI and its "Analyze Skeleton" function can be used to construct the spatial connectivity of vascular centerlines. This function determines which points are interconnected by analyzing the distance and direction relationship between points in the candidate point set. In practice, a threshold is first set to define the conditions for points to be considered connected, which is based on the typical diameter of cerebrovascular vessels and the expected centerline spacing. The "Skeletonize" tool is applied, which simplifies the vascular structure into a skeleton with a single pixel width while maintaining the connectivity information of the vessels. Through this process, a vascular centerline connectivity graph can be obtained, which shows the connectivity relationship between cerebrovascular centerline candidate points. This connectivity graph provides a clear path for each centerline candidate point, indicating how they are connected to each other.

[0058] Step S35: constructing a minimum spanning tree for the cerebral vascular centerline candidate point set according to the vascular centerline connection relationship diagram to obtain an initial cerebral vascular tree structure diagram; Specifically, the vascular centerline connection relationship graph can be imported into MATLAB and converted into a graph object in graph theory. Use MATLAB's graph function to create this graph object, where the vascular centerline candidate points are used as the nodes of the graph and the connectivity between them is used as the edges of the graph. Use MATLAB's minspantree function to calculate the minimum spanning tree of this graph. This function finds the shortest path connecting all nodes (vascular centerline candidate points) while ensuring the connectivity of the entire network, thereby constructing an initial cerebrovascular tree structure. In actual operation, it is necessary to set weight parameters to reflect the length or importance of different vascular segments, so that the minspantree function can select paths based on these weights. After executing the minimum spanning tree construction, the initial cerebrovascular tree structure diagram is finally obtained, which shows the connection relationship between the cerebrovascular centerline candidate points in a tree form.

[0059] Step S36: extracting morphological features from the complete cerebral vascular mask image to obtain cerebral vascular morphological feature data, and performing fast marching reconstruction on the initial cerebral vascular tree structure image based on the cerebral vascular morphological feature data to obtain an initial cerebral vascular tree model; Specifically, FIJI can be used to open the complete cerebrovascular mask map and apply its morphological tools to extract the morphological features of the cerebrovascular vessels. Use the "Analyze Particles" function to measure the area, perimeter, shape, and direction of each connected vascular region. These measurements will be used as cerebrovascular morphological feature data. Export these morphological feature data as a CSV file for further processing in MATLAB. In MATLAB, load the CSV file and use these data to guide the fast marching reconstruction of the initial cerebrovascular tree structure diagram. Use MATLAB's "active contour" method, which is an iterative process that adjusts the tree structure based on morphological feature data. In the specific operation, set the initial conditions of the fast marching algorithm, including the starting point and end conditions of the tree structure. The algorithm starts from the entry point of the cerebrovascular vessel, moves in the direction indicated by the morphological feature data, and forks according to the feature data when encountering a branch point. In this way, an initial cerebrovascular tree model that matches the actual cerebrovascular morphology can be constructed. Finally, a detailed initial cerebrovascular tree model is obtained.

[0060] Step S37: performing bifurcation point detection on the initial cerebrovascular tree model to obtain a cerebrovascular bifurcation point candidate set, and performing vascular direction identification on the initial cerebrovascular tree model to obtain cerebrovascular direction information; Specifically, the initial cerebrovascular tree model can be loaded into MATLAB, and the bwmorph function in MATLAB's image processing toolbox can be applied to detect bifurcation points using the "branch points" morphological operation. This function can identify branch areas in the tree structure, that is, bifurcation points. In actual operation, the parameters of the bwmorph function are set to specify the type of operation to be performed as "branch points", and a structural element, such as a 3x3 square or circular kernel, is selected to define the neighborhood range. After executing the function, a binary image containing a set of bifurcation point candidates is obtained, in which the bifurcation points are marked as 1 and other areas are 0. Next, the regionprops function of MATLAB is used to calculate the directional characteristics of each vascular segment, such as the main axis direction. This function can provide the geometric properties of each connected region, including the main axis direction. By analyzing the main axis direction of each vascular segment, the cerebrovascular direction information can be obtained, including the starting and ending directions of each vascular segment.

[0061] Step S38: reconstructing the topological structure of the initial cerebrovascular tree model according to the candidate set of cerebrovascular bifurcation points and the cerebrovascular direction information to obtain a complete cerebrovascular network topology map.

[0062] Specifically, the bifurcation point candidate set and cerebrovascular direction information can be loaded into MATLAB. Use the graph function in MATLAB's graph theory toolbox to construct a graph object, in which the center lines of the cerebrovascular vessels are used as the edges of the graph and the bifurcation points are used as the nodes of the graph. Next, adjust the edges of the graph according to the cerebrovascular direction information to ensure that the direction of each vascular segment is consistent with the identified direction. Use the addedge function to add or modify edges to reflect the actual direction of the blood vessels. During the reconstruction process, pay attention to the processing of bifurcation points to ensure that each bifurcation point is correctly connected to its corresponding vascular segment. Use the updategraph function to update the graph object, add new nodes and edges, and form a complete cerebrovascular network topology. Finally, use MATLAB's plot function to visualize the complete cerebrovascular network topology. This diagram clearly shows the connection relationship of the cerebrovascular network, including the starting point, end point, and bifurcation point of the blood vessels.

[0063] The present invention generates an accurate cerebrovascular distance field map through Euclidean distance transformation, providing a key basis for vascular structure analysis and centerline extraction. The image is further smoothed by anisotropic diffusion filtering to reduce noise and artifacts and improve image quality. The cerebrovascular centerline is accurately located by solving geodesic equations and extracting local maximum values, providing key geometric information for constructing a cerebrovascular network. The connectivity of the vascular network is enhanced by spatial connectivity construction and a minimum spanning tree algorithm to form a vascular centerline connection relationship diagram. The construction of the cerebrovascular tree structure is accelerated by a fast marching reconstruction technology based on morphological feature data, thereby improving the efficiency of model construction. The branch information of the cerebrovascular network is accurately revealed by bifurcation point detection and vascular direction identification. Finally, the topological structure reconstruction step integrates bifurcation point and direction information to generate a complete cerebrovascular network topology map.

[0064] Preferably, step S4 comprises the following steps: Step S41: performing orthogonal plane sampling on the vascular centerline based on the complete cerebral vascular network topology map to obtain a cerebral vascular cross-section sampling plane set; Specifically, the complete cerebrovascular network topology map can be imported into 3D Slicer, and the vascular centerline data can be loaded. These data define the path of each blood vessel in the cerebrovascular network. In 3D Slicer, use the "Resample Fibers" tool, which allows uniform or custom interval sampling along the vascular centerline. Set the sampling parameters and select the orthogonal plane sampling method, which means that each sampling plane is perpendicular to the direction of the vascular centerline. Define the interval between the sampling planes, for example, sample once every 1 mm to ensure that the details of the vascular cross-section can be captured. 3D Slicer automatically moves along the vascular centerline and creates a plane perpendicular to the centerline at each specified interval, thereby generating a series of orthogonal sampling planes. Through this process, a set of cerebrovascular cross-section sampling planes is finally obtained, each of which represents a vascular cross-section at a specific location in the vascular network.

[0065] Step S42: extract edge points from the cerebral blood vessel cross-section sampling plane set to obtain a cerebral blood vessel cross-section edge point cloud set; Specifically, the "Edge Detection" tool of 3D Slicer can be used to process each cerebrovascular section sampling plane. This tool applies an edge detection algorithm, such as the Canny algorithm, to identify the edges of the blood vessels in each section. The Canny algorithm enhances edge features through Gaussian filtering, gradient calculation, and non-maximum suppression, and uses double thresholds to determine strong edges and weak edges. In actual operation, the parameters of the Canny algorithm are adjusted, such as the standard deviation of the Gaussian filter, the gradient threshold, and the area threshold. 3D Slicer automatically performs edge detection on each sampling plane and marks the edges of the blood vessels. After performing edge detection, a point cloud set of the cerebrovascular section edge is finally obtained, which contains the location information of the blood vessel edge in each section.

[0066] Step S43: Calculating the ellipse fitting parameters of the cerebral blood vessel section based on the cerebral blood vessel section edge point cloud set to obtain a cerebral blood vessel ellipse interface parameter set, and constructing inner and outer double ellipse boundaries of the cerebral blood vessel section based on the cerebral blood vessel ellipse interface parameter set to obtain a cerebral blood vessel double ellipse section contour set; Specifically, the edge point cloud set of the cerebrovascular section can be imported into MATLAB. Use the fit function of MATLAB and select the ellipse as the fitting model to calculate the ellipse fitting parameters of each section. The fit function can automatically determine the optimal ellipse parameters based on the provided point cloud data, including the coordinates of the center point, the lengths of the major and minor axes, and the rotation angle of the ellipse. In actual operation, the edge point cloud of each cerebrovascular section is iteratively fitted to find the best matching ellipse parameters. It is necessary to set initial parameters or constraints, such as the approximate ratio range of the major and minor axes. During the fitting process, the fit function will adjust the ellipse parameters to minimize the difference between the point cloud data and the ellipse model. After obtaining the ellipse fitting parameters of each cerebrovascular section, the inner and outer double ellipse boundaries are constructed according to these parameters. First, an inner ellipse boundary is constructed, which fits the inner edge of the vascular section closely, and then an outer ellipse boundary is constructed, which covers the outer edge of the vascular section. Finally, the inner and outer ellipse boundary parameters of all cerebrovascular sections are combined to form a cerebrovascular double ellipse section contour set. This contour set contains the inner and outer ellipse parameters of each cerebrovascular section.

[0067] Step S44: acquiring cerebrovascular morphological feature data, and fusing the cerebrovascular double elliptical cross-sectional contour set with the cerebrovascular morphological feature data to obtain a standard cerebrovascular cross-sectional model; Specifically, cerebrovascular morphological feature data can be imported from a database or file, which includes information such as the width, curvature, and branch angle of the blood vessel. At the same time, the cerebrovascular double ellipse section profile set is also loaded. Use Python to write a function that can match and fuse each double ellipse section profile with the corresponding morphological feature data. For each section profile, the parameters of the ellipse are adjusted according to the morphological feature data. For example, if the feature data shows that a certain blood vessel section is wide, the length of the major axis and minor axis of the corresponding ellipse will be adjusted to reflect this feature. In actual operation, the double ellipse model needs to be fine-tuned to ensure that it more accurately conforms to the actual cerebrovascular morphology. This involves adjusting the center point, major axis, minor axis, and rotation angle of the ellipse. Use the NumPy library to perform these numerical calculations, and use the OpenCV library to process and display images. After the fusion is completed, the standard cerebrovascular section model is obtained.

[0068] Step S45: performing active contour evolution on the standard cerebral blood vessel cross-section model to obtain a standard cerebral blood vessel cross-section contour set; Specifically, the standard cerebrovascular cross-section model can be imported into MATLAB, and its "Active Contour" tool can be used to perform active contour evolution. This tool allows an initial contour to be defined and automatically evolves the contour based on the gradient information of the image so that it fits the boundary of the target object more closely. In actual operation, an initial contour is selected, which can be a double elliptical boundary in the standard cerebrovascular cross-section model. Then, the parameters of the evolution, such as the number of iterations, the evolution speed, and the gradient calculation method, are set. MATLAB automatically executes the active contour evolution algorithm, and gradually adjusts the position and shape of the contour based on the image gradient information of the cerebrovascular cross-section. After the evolution is completed, a set of standard cerebrovascular cross-section contours is finally obtained, which contains optimized cerebrovascular cross-section contours.

[0069] Step S46: performing radial distance mapping calculation on the standard cerebral blood vessel cross-sectional contour set to obtain cerebral blood vessel caliber assessment data; Specifically, a set of standard cerebrovascular cross-sectional profiles can be imported into MATLAB, and a coordinate system can be defined for each cross-sectional profile. The center of each profile is selected as the coordinate origin, and the radial distance from each point on the profile to the center point is calculated. In MATLAB, these distances are calculated using the sqrt function and coordinate difference. Next, the radial distances of all points are statistically analyzed, and the average radial distance and standard deviation are calculated, which will serve as evaluation data for cerebrovascular caliber. Finally, these cerebrovascular caliber evaluation data are organized into tables or graphs.

[0070] Step S47: constructing an implicit function of the cerebral blood vessel surface based on the cerebral blood vessel caliber assessment data to obtain a cerebral blood vessel implicit surface function, and performing isosurface extraction on the cerebral blood vessel implicit surface function to obtain cerebral blood vessel surface mesh data; Specifically, the cerebral vascular caliber assessment data can be imported into MeshLab, and these data can be used to define the implicit surface function of the cerebral blood vessels. In MeshLab, use the "Create Implicit Surface" function in the "Filter" menu, input the cerebral vascular caliber assessment data as the function parameter, and define an implicit function such as f(x, y, z) = 0, where f is defined based on the size and shape of the cerebral blood vessels. Next, use the "Isosurface Extraction" function of MeshLab to extract the isosurface, that is, the surface of the cerebral blood vessels, from the implicit function. Set the isovalue parameters of the isosurface, which corresponds to the average caliber of the cerebral blood vessels. After performing the isosurface extraction, the cerebral vascular surface mesh data can be obtained, which is a three-dimensional mesh composed of multiple vertices, edges, and faces.

[0071] Step S48: reconstructing the standard cerebral blood vessel cross-section model with surface constraints according to the cerebral blood vessel morphological feature data and the cerebral blood vessel surface mesh data to obtain an optimized cerebral blood vessel surface model.

[0072] Specifically, the cerebrovascular morphological feature data and the cerebrovascular surface mesh data can be imported into 3D Slicer, and the standard cerebrovascular cross-section model can be loaded in the software. Next, a Python script is used to write code in the Python console of 3D Slicer to adjust the geometric parameters of the cerebrovascular cross-section model according to the cerebrovascular morphological feature data. For example, if the morphological feature data shows that a certain section of blood vessels is wider, the elliptical model parameters of the cross-section of the blood vessel will be adjusted accordingly. In actual operation, the optimization tools in the SciPy library are used to minimize the difference between the cerebrovascular cross-section model and the cerebrovascular surface mesh data. Define an objective function that calculates the difference between the surface mesh predicted by the model and the actual surface mesh, and use gradient descent or other optimization algorithms to adjust the model parameters until the best fit solution is found. Through this process, an optimized cerebrovascular surface model can be obtained, which more accurately reflects the actual morphology of cerebrovascular vessels in terms of geometric features.

[0073] The present invention accurately obtains the cross section of cerebral blood vessels through orthogonal plane sampling, providing a solid geometric foundation for subsequent analysis. The blood vessel boundary is made clearer through edge point extraction and enhancement. The shape of the cerebral blood vessel cross section is accurately simulated by double ellipse boundary construction, which improves the biosimulation degree of the model. The double ellipse cross section contour and morphological feature data are integrated through cerebral blood vessel cross section feature fusion to generate an accurate and complete standard cerebral blood vessel cross section model. The model is adapted to the change of blood vessel shape through active contour evolution technology, which enhances the adaptability and accuracy of the model. The cerebral blood vessel caliber evaluation data obtained by radial distance mapping calculation provides an important basis for diagnosing vascular lesions. The implicit surface function constructed based on the caliber evaluation data provides a continuous and smooth mathematical description for the cerebral blood vessel surface, which is convenient for surface analysis and processing. The cerebral blood vessel surface mesh data is accurately extracted from the implicit surface function through the isosurface extraction technology, providing a high-quality model for applications such as three-dimensional printing and surgical simulation. Finally, the surface constraint reconstruction combined with the morphological feature data and the surface mesh data generates an optimized cerebral blood vessel surface model, which further improves the accuracy and practicality of the model.

[0074] Preferably, step S5 comprises the following steps: Step S51: extracting the cerebral vascular surface curvature features from the optimized cerebral vascular surface model to obtain a cerebral vascular surface curvature feature map; Specifically, the optimized cerebral vascular surface model can be imported into Rhinoceros 3D, which allows for precise editing and analysis of complex three-dimensional models. In Rhinoceros 3D, use its built-in "Measure" tool to calculate the curvature of each point on the model surface. This tool can provide the Gaussian curvature and mean curvature of each point, and these curvature values ​​reflect the degree of curvature of the vascular surface at that point. Next, export these curvature data as a text file and then use MATLAB for further analysis and processing. In MATLAB, load the curvature data and use the surf or mesh function to visualize these data as a three-dimensional curvature feature map. This feature map shows the curvature size of different areas on the cerebral vascular surface, and the darker the color, the greater the curvature. Through this process, a cerebral vascular surface curvature feature map can be obtained.

[0075] Step S52: locating the lesion area on the cerebral blood vessel surface based on the cerebral blood vessel surface curvature characteristic map to obtain a cerebral blood vessel lesion area positioning map; Specifically, the cerebral vascular surface curvature feature map can be loaded in MATLAB, and a Gaussian filter can be applied using the imfilter function to smooth the curvature feature map. Next, the graythresh function is used to calculate the Otsu threshold to determine the boundary for distinguishing normal and abnormal curvatures. The imbinarize function is used to convert the curvature feature map into a binary image based on the calculated threshold, where the area above the threshold is marked as a candidate lesion area. Finally, the detected lesion area is marked on the original cerebral vascular surface model to generate a cerebral vascular lesion area localization map.

[0076] Step S53: extracting multi-scale lesion features from the cerebrovascular lesion area localization map to obtain a cerebrovascular lesion feature vector set, and performing hierarchical clustering on the cerebrovascular lesion feature vector set to obtain lesion type label data; Specifically, the cerebrovascular lesion area localization map can be imported into the Python environment, and then a series of multi-scale feature extraction methods, such as Gabor filter, LBP (local binary pattern) and HOG (histogram of oriented gradients), are defined to extract different features of the lesion area. For each lesion area, these feature extraction methods are applied and feature vectors are generated. Each feature vector contains lesion features extracted at different scales and directions. The feature vectors of all lesion areas are combined to form a set of cerebrovascular lesion feature vectors. Next, the hierarchical clustering algorithms in scikit-learn, such as wardlinkage or average linkage, are used to cluster the set of cerebrovascular lesion feature vectors. The clustering algorithm groups the feature vectors according to the similarity between them, and each group represents a type of lesion. A lesion type label is assigned to each lesion area according to the clustering results, and these labels constitute the lesion type label data.

[0077] Step S54: performing topological spatial relationship mapping on the lesion type label data based on the optimized cerebrovascular surface model to obtain a topological map of the cerebrovascular lesion area; Specifically, the optimized cerebrovascular surface model and lesion type label data can be imported into MeshLab. In MeshLab, the "Edit Attributes" function is used to attach lesion type label data to each mesh point or mesh cell. The surface points of each lesion area are labeled with the corresponding lesion type. Next, Python scripts are used to interact with the command line interface of MeshLab to write scripts to automate the process of topological spatial relationship mapping. Based on the lesion type label data, the script calculates the spatial distance and connectivity between lesion areas and constructs a topological graph in which nodes represent lesion areas and edges represent spatial relationships between areas. In Python, the network analysis library NetworkX is used to construct and analyze this topological graph. Topological features such as degree (number of connected edges), clustering coefficient, and path length of each lesion area are calculated, and these features are associated with lesion types. Finally, a topological map of cerebrovascular lesion areas is obtained.

[0078] Step S55: reconstructing the mesh of the lesion area of ​​the optimized cerebral vascular surface model according to the cerebrovascular lesion area topology map to obtain a complete cerebrovascular network three-dimensional model.

[0079] Specifically, the optimized cerebrovascular surface model and the topological map of the cerebrovascular lesion area can be imported into Blender. Next, the mesh of the lesion area can be finely adjusted using Blender's mesh editing tools, such as the "Move", "Rotate" and "Scale" tools in the "Edit Mode". The shape and size of the mesh are adjusted according to the nature of the lesion (e.g., stenosis, dilation or obstruction). A Python script is used to interact with Blender's API. The script automatically selects the corresponding mesh part based on the data in the topological map of the cerebrovascular lesion area and applies predefined deformation or modification operations. The NumPy library is used to process large amounts of mesh data. During the reconstruction process, the results are constantly previewed and adjusted in Blender to ensure that the mesh of the lesion area matches the actual cerebrovascular lesion characteristics. Finally, the reconstructed cerebrovascular surface model is exported to a standard 3D model format to obtain a complete 3D model of the cerebrovascular network.

[0080] The present invention provides a geometric basis for locating the lesion area by extracting the surface curvature features of cerebral blood vessels, thereby quickly and accurately locating the cerebral vascular lesion area. Multi-scale lesion feature extraction comprehensively captures lesion features of different scales, enhances the understanding of lesion complexity, and efficiently classifies lesion types through hierarchical clustering to generate lesion type label data. The topological relationship of the lesion area is clearly expressed through topological spatial relationship mapping. The grid reconstruction technology based on the topological map of the lesion area accurately reconstructs the lesion area, improving the clinical application value of the model. Finally, a complete three-dimensional model of the cerebral vascular network is constructed, providing accurate anatomical information for surgical planning.

[0081] Preferably, the present invention further provides a cerebrovascular three-dimensional model construction system based on cerebrovascular image segmentation, which is used to execute the cerebrovascular three-dimensional model construction method based on cerebrovascular image segmentation as described above, and the cerebrovascular three-dimensional model construction system based on cerebrovascular image segmentation comprises: The feature enhancement module is used to obtain an initial cerebrovascular image set; register the initial cerebrovascular image set to obtain a cerebrovascular image set; perform weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; The vascular segmentation module is used to perform multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; perform spatial attention microvascular segmentation on the enhanced cerebrovascular feature map to obtain a microvascular segmentation feature map; perform vascular morphology constraint fusion on the large vessel segmentation feature map and the microvascular segmentation feature map to obtain a complete cerebrovascular mask map; The topology construction module is used to calculate the geodesic distance field of the complete cerebrovascular mask map to obtain the candidate point set of the cerebrovascular centerline; to construct a fast marching tree structure on the candidate point set of the cerebrovascular centerline to obtain an initial cerebrovascular tree model; to optimize the vascular bifurcation points of the initial cerebrovascular tree model to obtain a complete cerebrovascular network topology map; The surface model building module is used to perform double elliptical cross-section fitting modeling on cerebral blood vessels based on the complete cerebral vascular network topology map to obtain a standard cerebral vascular cross-section model; perform active contour caliber evaluation on the standard cerebral vascular cross-section model to obtain cerebral vascular caliber evaluation data; perform implicit surface reconstruction on cerebral blood vessels based on the cerebral vascular caliber evaluation data to obtain an optimized cerebral vascular surface model; The model reconstruction module is used to locate the vascular lesion area on the optimized cerebrovascular surface model to obtain a cerebrovascular lesion area positioning map; classify the lesions on the cerebrovascular lesion area positioning map to obtain lesion type label data; reconstruct the lesion grid of the optimized cerebrovascular surface model according to the lesion type label data to obtain a complete cerebrovascular network three-dimensional model.

[0082] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0083] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation, characterized in that: The following steps are involved: Step S1: obtaining an initial cerebrovascular image set; registering the initial cerebrovascular image set to obtain a cerebrovascular image set; performing weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; Step S2: performing multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; performing spatial attention microvascular segmentation on the enhanced cerebrovascular feature map to obtain a microvascular segmentation feature map; performing vascular morphology-constrained fusion on the large vessel segmentation feature map and the microvascular segmentation feature map to obtain a complete cerebrovascular mask map; Step S3: performing geodesic distance field calculation on the complete cerebral vascular mask image to obtain a set of candidate points for the cerebral vascular centerline; performing fast marching tree structure construction on the set of candidate points for the cerebral vascular centerline to obtain an initial cerebral vascular tree model; performing vascular bifurcation point optimization on the initial cerebral vascular tree model to obtain a complete cerebral vascular network topology map; Step S4: performing double ellipse cross-section fitting modeling on the cerebral blood vessels based on the complete cerebral vascular network topology to obtain a standard cerebral blood vessel cross-section model; performing active contour caliber evaluation on the standard cerebral blood vessel cross-section model to obtain cerebral blood vessel caliber evaluation data; performing implicit surface reconstruction on the cerebral blood vessels based on the cerebral blood vessel caliber evaluation data to obtain an optimized cerebral blood vessel surface model; Step S5: positioning the vascular lesion area on the optimized cerebrovascular surface model to obtain a cerebrovascular lesion area positioning map; classifying the lesions on the cerebrovascular lesion area positioning map to obtain lesion type label data; reconstructing the lesion mesh on the optimized cerebrovascular surface model according to the lesion type label data to obtain a complete cerebrovascular network three-dimensional model.

2. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: obtaining a cerebral vascular imaging projection angle, and performing angiography scanning on the patient according to the cerebral vascular imaging projection angle to obtain an initial cerebral vascular image set; wherein the initial cerebral vascular image set includes a digital subtraction cerebral vascular angiography image, a tomography cerebral angiography image, and a magnetic resonance angiography image; Step S12: performing vascular tree segmentation and enhancement on the digital subtraction cerebral angiography image to obtain an enhanced digital subtraction vascular tree map; Step S13: performing voxel density adaptive segmentation on the tomographic cerebral angiography image to obtain a segmented tomographic vascular structure map; Step S14: performing phase contrast enhancement on the magnetic resonance angiography image to obtain an enhanced magnetic resonance vascular network map; Step S15: performing deformation field calculation on the enhanced digital subtraction vascular tree map, the segmented tomography vascular structure map and the enhanced magnetic resonance vascular network map to obtain a multimodal image deformation mapping field, and performing elastic deformation registration on the enhanced digital subtraction vascular tree map, the segmented tomography vascular structure map and the enhanced magnetic resonance vascular network map according to the multimodal image deformation mapping field to obtain a cerebral vascular image set; wherein the cerebral vascular image set includes the registered digital subtraction vascular tree map, the registered tomography vascular structure map and the registered magnetic resonance vascular network map; Step S16: performing weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map.

3. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1, characterized in that: Step S13 includes the following steps: Step S131: performing multi-peak density detection on the tomographic cerebral angiography image to obtain a cerebral vascular density distribution map; Step S132: Calculating the blood vessel density threshold for the tomographic cerebral angiography image based on the cerebral blood vessel density distribution map to obtain a blood vessel density segmentation threshold set; Step S133: extracting voxel cube neighborhood from the tomographic cerebral angiography image to obtain three-dimensional voxel neighborhood data; Step S134: performing density-connected region growing segmentation on the tomographic cerebral angiography image based on the blood vessel density segmentation threshold set and the three-dimensional voxel neighborhood data to obtain an initial cerebral blood vessel segmentation map; Step S135: performing vascular structure morphology screening on the initial cerebral vascular segmentation image to obtain a set of cerebral vascular candidate regions; Step S136: locating the vascular structure boundary of the initial cerebral vascular segmentation map according to the cerebral vascular candidate region set to obtain a cerebral vascular boundary positioning map, and performing density distribution mapping on the cerebral vascular boundary positioning map to obtain a segmented tomographic vascular structure map.

4. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 2, characterized in that: Step S16 includes the following steps: Step S161: respectively evaluating the vascular clarity of the registered digital subtraction vascular tree map, the registered tomographic vascular structure map, and the registered magnetic resonance vascular network map to obtain the digital subtraction vascular clarity, the tomographic vascular clarity, and the magnetic resonance vascular clarity; Step S162: normalizing and weighting the digital subtraction blood vessel clarity, the tomography blood vessel clarity and the magnetic resonance blood vessel clarity to obtain a cerebrovascular structure feature weight set; wherein the cerebrovascular structure feature weight set includes the digital subtraction image feature weight, the tomography image feature weight and the magnetic resonance image feature weight; Step S163: performing wavelet domain weighted enhancement on the registered digital subtraction vascular tree map according to the digital subtraction image feature weights to obtain a weighted digital subtraction vascular feature map, and performing pyramid domain weighted enhancement on the registered tomography vascular structure map according to the tomography image feature weights to obtain a weighted tomography vascular feature map; performing Gabor domain weighted enhancement on the registered magnetic resonance vascular network map according to the magnetic resonance image feature weights to obtain a weighted magnetic resonance vascular feature map; Step S164: performing feature fusion on the weighted digital subtraction vascular feature map, the weighted tomography vascular feature map and the weighted magnetic resonance vascular feature map to obtain an enhanced cerebral vascular feature map.

5. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing pyramid feature recursive extraction on the enhanced cerebrovascular feature map to obtain a multi-level cerebrovascular feature atlas; Step S22: extracting encoder features from the multi-level cerebrovascular feature atlas to obtain a multi-level encoded cerebrovascular feature set; Step S23: reorganizing the multi-level coded cerebrovascular feature set to obtain a reorganized coded cerebrovascular feature set, and performing decoder feature recovery on the reorganized coded cerebrovascular feature set to obtain a large vessel segmentation feature map; Step S24: performing spatial attention microvascular segmentation on the enhanced cerebral vascular feature map to obtain a microvascular segmentation feature map; Step S25: performing morphological constraint extraction on the large vessel segmentation feature map and the microvessel segmentation feature map to obtain a cerebrovascular morphological constraint set, and performing vascular connectivity detection on the large vessel segmentation feature map and the microvessel segmentation feature map based on the cerebrovascular morphological constraint set to obtain a cerebrovascular connectivity map; Step S26: morphologically reconstructing the large vessel segmentation feature map and the microvessel segmentation feature map according to the cerebral vascular connectivity map to obtain a complete cerebral vascular mask map.

6. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 5, characterized in that: Step S24 includes the following steps: Step S241: performing multi-channel feature decomposition on the enhanced cerebrovascular feature map to obtain a multi-channel cerebrovascular feature set; Step S242: re-arranging the vascular space geometric feature dimensions of the cerebral blood vessels according to the multi-channel cerebral blood vessel feature set to obtain a re-arranged geometric feature matrix; Step S243: performing autocorrelation calculation on the reordered geometric feature matrix to obtain a microvascular feature correlation matrix; Step S244: performing nonlinear mapping on the microvascular feature correlation matrix to obtain a multi-scale attention weight matrix, and performing spatial dependency mapping on the enhanced cerebrovascular feature map according to the multi-scale attention weight matrix to obtain a microvascular spatial attention weight map; Step S245: performing microvascular structure enhancement on the enhanced cerebral vascular feature map according to the microvascular spatial attention weight map to obtain an enhanced microvascular feature set; Step S246: reconstructing the enhanced microvascular feature set to obtain a microvascular segmentation feature map.

7. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing Euclidean distance transformation on the complete cerebral vascular mask image to obtain an initial cerebral vascular distance field image; Step S32: performing anisotropic diffusion filtering on the complete cerebral vascular mask image based on the initial cerebral vascular distance field image to obtain a smoothed cerebral vascular distance field image; Step S33: Solving the geodesic equation for the smoothed cerebral vascular distance field map to obtain the cerebral vascular geodesic distance field map, and extracting the local maximum of the cerebral vascular structure based on the cerebral vascular geodesic distance field map to obtain a cerebral vascular centerline candidate point set; Step S34: constructing spatial connectivity of the cerebral vascular centerline candidate point set to obtain a vascular centerline connection relationship diagram; Step S35: constructing a minimum spanning tree for the cerebral vascular centerline candidate point set according to the vascular centerline connection relationship diagram to obtain an initial cerebral vascular tree structure diagram; Step S36: extracting morphological features from the complete cerebral vascular mask image to obtain cerebral vascular morphological feature data, and performing fast marching reconstruction on the initial cerebral vascular tree structure image based on the cerebral vascular morphological feature data to obtain an initial cerebral vascular tree model; Step S37: performing bifurcation point detection on the initial cerebrovascular tree model to obtain a cerebrovascular bifurcation point candidate set, and performing vascular direction identification on the initial cerebrovascular tree model to obtain cerebrovascular direction information; Step S38: reconstructing the topological structure of the initial cerebrovascular tree model according to the candidate set of cerebrovascular bifurcation points and the cerebrovascular direction information to obtain a complete cerebrovascular network topology map.

8. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing orthogonal plane sampling on the vascular centerline based on the complete cerebral vascular network topology map to obtain a cerebral vascular cross-section sampling plane set; Step S42: extract edge points from the cerebral blood vessel cross-section sampling plane set to obtain a cerebral blood vessel cross-section edge point cloud set; Step S43: Calculating the ellipse fitting parameters of the cerebral blood vessel section based on the cerebral blood vessel section edge point cloud set to obtain a cerebral blood vessel ellipse interface parameter set, and constructing inner and outer double ellipse boundaries of the cerebral blood vessel section based on the cerebral blood vessel ellipse interface parameter set to obtain a cerebral blood vessel double ellipse section contour set; Step S44: acquiring cerebrovascular morphological feature data, and fusing the cerebrovascular double elliptical cross-sectional contour set with the cerebrovascular morphological feature data to obtain a standard cerebrovascular cross-sectional model; Step S45: performing active contour evolution on the standard cerebral blood vessel cross-section model to obtain a standard cerebral blood vessel cross-section contour set; Step S46: performing radial distance mapping calculation on the standard cerebral blood vessel cross-sectional contour set to obtain cerebral blood vessel caliber assessment data; Step S47: constructing an implicit function of the cerebral blood vessel surface based on the cerebral blood vessel caliber assessment data to obtain a cerebral blood vessel implicit surface function, and performing isosurface extraction on the cerebral blood vessel implicit surface function to obtain cerebral blood vessel surface mesh data; Step S48: reconstructing the standard cerebral blood vessel cross-section model with surface constraints according to the cerebral blood vessel morphological feature data and the cerebral blood vessel surface mesh data to obtain an optimized cerebral blood vessel surface model.

9. The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: extracting the curvature features of the cerebral blood vessel surface from the optimized cerebral blood vessel surface model to obtain a cerebral blood vessel surface curvature feature map; Step S52: locating the lesion area on the cerebral blood vessel surface based on the cerebral blood vessel surface curvature characteristic map to obtain a cerebral blood vessel lesion area positioning map; Step S53: extracting multi-scale lesion features from the cerebrovascular lesion area localization map to obtain a cerebrovascular lesion feature vector set, and performing hierarchical clustering on the cerebrovascular lesion feature vector set to obtain lesion type label data; Step S54: performing topological spatial relationship mapping on the lesion type label data based on the optimized cerebrovascular surface model to obtain a topological map of the cerebrovascular lesion area; Step S55: reconstructing the mesh of the lesion area of ​​the optimized cerebral vascular surface model according to the topological map of the cerebral vascular lesion area to obtain a complete 3D model of the cerebral vascular network.

10. A 3D cerebrovascular model construction system based on cerebrovascular image segmentation, characterized in that: The method for constructing a 3D cerebral vascular model based on cerebral vascular image segmentation according to claim 1 is used to execute the cerebral vascular 3D model construction method based on cerebral vascular image segmentation according to claim 1, and the cerebral vascular 3D model construction system based on cerebral vascular image segmentation comprises: The feature enhancement module is used to obtain an initial cerebrovascular image set; register the initial cerebrovascular image set to obtain a cerebrovascular image set; perform weight mapping feature-level fusion on the cerebrovascular image set to obtain an enhanced cerebrovascular feature map; The vascular segmentation module is used to perform multi-scale large vessel segmentation on the enhanced cerebrovascular feature map to obtain a large vessel segmentation feature map; perform spatial attention microvascular segmentation on the enhanced cerebrovascular feature map to obtain a microvascular segmentation feature map; perform vascular morphology constraint fusion on the large vessel segmentation feature map and the microvascular segmentation feature map to obtain a complete cerebrovascular mask map; The topology construction module is used to calculate the geodesic distance field of the complete cerebrovascular mask map to obtain the candidate point set of the cerebrovascular centerline; to construct a fast marching tree structure on the candidate point set of the cerebrovascular centerline to obtain an initial cerebrovascular tree model; to optimize the vascular bifurcation points of the initial cerebrovascular tree model to obtain a complete cerebrovascular network topology map; The surface model building module is used to perform double elliptical cross-section fitting modeling on cerebral blood vessels based on the complete cerebral vascular network topology map to obtain a standard cerebral vascular cross-section model; perform active contour caliber evaluation on the standard cerebral vascular cross-section model to obtain cerebral vascular caliber evaluation data; perform implicit surface reconstruction on cerebral blood vessels based on the cerebral vascular caliber evaluation data to obtain an optimized cerebral vascular surface model; The model reconstruction module is used to locate the vascular lesion area on the optimized cerebrovascular surface model to obtain a cerebrovascular lesion area positioning map; classify the lesions on the cerebrovascular lesion area positioning map to obtain lesion type label data; reconstruct the lesion grid of the optimized cerebrovascular surface model according to the lesion type label data to obtain a complete cerebrovascular network three-dimensional model.

Images