A method for constructing a three-dimensional mask of a blood vessel tree and related products

Abstract

The application discloses a blood vessel tree three-dimensional mask construction method and related products. First, a target blood vessel tree structure containing a main aorta, main blood vessels and branch blood vessels is acquired, and a plurality of blood vessel center lines and their segmented information are extracted. Then, each center line is sampled to obtain a sampling line segment, and a set of inscribed circle physical radii corresponding to each blood vessel segment is obtained through morphological erosion detection. Next, a calculation task package is created for each blood vessel segment, and a unique geometric hash signature is calculated in parallel based on the center line coordinate point set in the task package. Then, according to the repeated matching relationship of the geometric hash signature, valid segments are screened out, and local blood vessel masks of each valid segment are generated by using the corresponding task package. Finally, the three-dimensional mask construction of the target blood vessel tree structure is completed by combining all the local blood vessel masks with the original structure parameters and morphological characteristics of the main aorta. The application improves the blood vessel label construction precision and the overall label construction efficiency.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method for constructing a three-dimensional mask for a vascular tree and related products. Background Technology

[0002] In the context of precision diagnosis and treatment of cardiovascular and cerebrovascular diseases, whether it's CT fractional flow reserve (FFRCT) calculation, interventional surgical path planning, or intravascular ultrasound navigation, all rely heavily on high-precision digital vascular models with anatomical segmentation markings for analysis and judgment. Clinically, these models are also essential for accurately distinguishing key vascular segments such as the right coronary artery (RCA), left anterior descending artery (LAD), and left circumflex artery (LCX). The core prerequisite for constructing high-quality vascular models is generating reliable vascular annotation labels; however, current mainstream vascular label generation technologies generally suffer from low label construction efficiency and insufficient label annotation accuracy. Summary of the Invention

[0003] Based on the above problems, this application provides a method for constructing a 3D mask for a blood vessel tree and related products, with the aim of improving the accuracy of blood vessel label construction and the overall efficiency of label construction.

[0004] The embodiments of this application disclose the following technical solutions: A method for constructing a 3D mask for a vascular tree, the method comprising: Obtain multiple centerlines from the target vascular tree structure, along with segment information corresponding to each centerline. The target vascular tree structure includes the aorta, multiple main trunk vessels, and multiple branch vessels. The main trunk vessels originate from the aorta and extend outwards, while the branch vessels branch off from the main trunk vessels. The starting point of each centerline is the root of the main trunk vessel to which the branch vessel belongs, and the ending point of each centerline is the terminal end of that branch vessel. Each centerline corresponds to multiple vascular segments. The segment information includes the segment index, the coordinates of the segment's starting point, the coordinates of the segment's ending point, and the coordinates of all centerline points from the segment's starting point to its ending point. Each centerline is sampled to obtain the corresponding sampling line segment, and the set of physical radii of the inscribed circle corresponding to each blood vessel segment on each sampling line segment is obtained based on morphological corrosion detection; each sampling line segment is composed of multiple centerline sampling points, and each blood vessel segment corresponds to multiple centerline sampling points; A computational task package is generated for each blood vessel segment, and a unique geometric hash signature corresponding to each blood vessel segment is calculated in parallel based on the centerline coordinate point set in each computational task package. The computational task package includes the segment information of the blood vessel segment, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the set of centerline coordinate points of the centerline where the blood vessel segment is located, and the set of voxel coordinate points of the original image corresponding to the blood vessel segment. Multiple first target segments are obtained by filtering all blood vessel segments based on the repeated matching relationship of multiple geometric hash signatures, and the local blood vessel mask corresponding to each first target segment is obtained based on the computation task package corresponding to each first target segment. Based on all local vascular masks, as well as the original structural parameters and morphological features of the aortic structure, a three-dimensional mask of the target vascular tree structure is constructed.

[0005] A 3D mask construction device for vascular trees, the device comprising: An information acquisition unit is used to acquire multiple centerlines in a target vascular tree structure, as well as segment information corresponding to each centerline. The target vascular tree structure includes the aorta, multiple main trunk vessels, and multiple branch vessels. The main trunk vessels originate from the aorta and extend outward, while the branch vessels branch off from the main trunk vessels. The starting point of the centerline is the root of the main trunk to which the branch vessel belongs, and the ending point of the centerline is the terminal end of the branch vessel. Each centerline corresponds to multiple vascular segments. The segment information includes the segment index, the coordinates of the segment's starting point, the coordinates of the segment's ending point, and the coordinates of all centerline points from the segment's starting point to its ending point. The sampling unit is used to sample each centerline to obtain the sampling line segment corresponding to each centerline; each sampling line segment is composed of multiple centerline sampling points, and each segment of the blood vessel corresponds to multiple centerline sampling points; The radius detection unit is used to obtain the set of physical radii of the inscribed circles of each blood vessel segment on each sampling line segment based on morphological corrosion detection. The task package generation unit is used to generate a calculation task package for each blood vessel segment. The calculation task package includes the segment information of the blood vessel segment, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the set of centerline coordinate points of the centerline where the blood vessel segment is located, and the set of voxel coordinate points of the original image corresponding to the blood vessel segment. The hash calculation unit is used to calculate the unique geometric hash signature corresponding to each blood vessel segment in parallel based on the centerline coordinate point set in each calculation task package. The segmentation filtering unit is used to filter all vascular segments based on the duplicate matching relationship of multiple geometric hash signatures to obtain multiple first target segments; The local mask generation unit is used to obtain the local vascular mask corresponding to each first target segment based on the computational task package corresponding to each first target segment. A three-dimensional mask construction unit is used to construct a three-dimensional mask of the target vascular tree structure based on all local vascular masks, as well as the original structural parameters and morphological features of the aortic structure.

[0006] A 3D mask construction device for vascular trees includes: a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the 3D mask construction method for vascular trees as described above.

[0007] Compared with the prior art, this application has the following beneficial effects: This embodiment first completes the standardized information acquisition of the vascular structure: acquiring multiple centerlines within the target vascular tree and the segment information corresponding to each centerline. Then, it calculates the diameter parameters of each vascular segment: sampling is performed on each centerline to obtain sampling segments composed of multiple centerline sampling points. Then, based on morphological erosion detection, the physical radius of the inscribed circle of the corresponding vascular segment on each sampling segment is calculated, ultimately obtaining the set of physical radii of the inscribed circle corresponding to each vascular segment, providing precise geometric constraints for subsequent mask construction. Next, it completes the parallel computation and redundancy screening of the segments: a standardized computation task package is generated for each vascular segment. The task package contains the segment information, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the set of all centerline coordinate points of the centerline to which the segment belongs, and the set of original image voxel coordinate points corresponding to the segment. Simultaneously, based on the centerline coordinate point set in each task package, the unique geometric hash signature corresponding to each vascular segment is calculated in parallel. Then, the first target segment without redundancy is obtained by filtering through the duplicate matching relationship of multiple geometric hash signatures. Next, local masks for each segment are generated: based on the computational task package corresponding to each first target segment, a corresponding local vascular mask is generated. Finally, the global 3D mask is constructed: by integrating all local vascular masks and combining them with the original structural parameters and morphological features of the aortic structure, a complete 3D mask of the target vascular tree can be obtained. Attached Figure Description

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

[0009] Figure 1 A schematic diagram illustrating an exemplary application scenario provided in this application embodiment; Figure 2 A flowchart illustrating the implementation of a 3D mask construction method for a blood vessel tree provided in this application embodiment; Figure 3 A flowchart illustrating the implementation of a method for obtaining the physical radius of the inscribed circle of a blood vessel, as provided in this application embodiment; Figure 4 A flowchart illustrating the implementation of a method for caching and mapping storage of blood vessel segment radius features provided in this application embodiment; Figure 5 A flowchart illustrating the implementation of a blood vessel segment radius feature retrieval and reuse method provided in this application embodiment; Figure 6 A flowchart illustrating the implementation of another multi-scale topology model construction method provided in this application embodiment; Figure 7 A flowchart illustrating the implementation of an initial mask generation method provided in this application embodiment; Figure 8 A flowchart illustrating the implementation of a 3D mask fusion and stitching method for a blood vessel tree, as provided in this application embodiment; Figure 9 This is a schematic diagram of a three-dimensional mask construction device for a vascular tree provided in an embodiment of this application. Detailed Implementation

[0010] As mentioned earlier, in the context of precision diagnosis and treatment of cardiovascular and cerebrovascular diseases, whether it's FFRCT calculation, interventional surgical path planning, or intravascular ultrasound navigation, all rely heavily on high-precision digital vascular models with anatomical segmentation markings for analysis and judgment. During clinical diagnosis and treatment, these models are also needed to accurately distinguish key coronary vessel segments such as the RCA, LAD, and LCX. The core prerequisite for constructing a high-quality digital vascular model lies in generating accurate and reliable vascular annotation labels. However, current mainstream vascular annotation label generation technologies still have many technical shortcomings, specifically in the following aspects: Low global computational efficiency: When dealing with complex whole-body vascular trees and fine coronary artery trees containing hundreds of segments, traditional serial reconstruction algorithms need to repeatedly traverse and scan massive three-dimensional voxel matrices and read and write data. The computation time usually reaches minutes or even longer, which is difficult to meet the processing needs of real-time clinical diagnosis and treatment. Radius estimation distortion: Conventional Euclidean distance transformation or cross-section fitting estimation methods cannot adapt to complex working conditions such as sudden changes in blood vessel diameter and irregular blood vessel morphology. This can easily cause the constructed blood vessel model to estimate the diameter too coarsely at the narrowing of the lesion, thus masking the true lesion characteristics, and to estimate the diameter too thinly at the bifurcation of the blood vessel, resulting in the problem of blood vessel structure breakage. Expansion overflow and noise interference: When using the three-dimensional expansion algorithm, it is easy to generate discrete spurs and isolated artifact areas in non-vascular anatomical areas, which will destroy the integrity of the real vascular anatomical structure. At the same time, it is easy to invade adjacent non-target tissue areas such as veins and bones, causing distortion of label annotation boundaries.

[0011] The aforementioned technical deficiencies can lead to problems such as low computational efficiency in vascular label generation and insufficient accuracy in anatomical label annotation.

[0012] Based on this, this application provides a method and related products for constructing a 3D vascular tree mask. First, a complete vascular tree structure including the aorta, main trunk vessels, and branch vessels is obtained. The start and end rules of the centerlines of each branch vessel are clarified, and the index, start and end coordinates, and centerline point coordinates of each vessel segment are fully collected, establishing a unified and standardized vascular anatomical segmentation basis from the source. Next, adaptive sampling is performed on each centerline, and morphological erosion detection is used to accurately obtain the physical radius of the inscribed circle of the corresponding sampling point of each vessel segment, providing a realistic and reliable diameter constraint for vascular contour construction. Then, a standardized computation task package containing segment information, sampling parameters, radius set, centerline coordinates, and voxel coordinates is encapsulated for each vessel segment. Based on the centerline coordinate point set within the task package, a unique geometric hash signature for each segment is obtained through parallel computation. The vascular segments are deduplicated and effectively filtered based on the duplicate matching relationship of the hash signatures, avoiding redundant calculations and repeated annotations. Finally, a precise local vascular mask is generated based on the filtered target segments and the corresponding task package. Finally, by combining all local vascular masks with the original structural parameters and morphological features of the aorta, a complete 3D vascular tree mask with clear topology, accurate anatomical positioning, realistic diameter fitting, and no redundant noise is constructed. While ensuring modeling accuracy and anatomical integrity, the construction efficiency and stability of the 3D vascular mask are greatly improved.

[0013] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0014] It should be understood that the technical solution of this application can be applied to the following scenarios, but is not limited to: In one feasible approach Figure 1 This is a schematic diagram illustrating an exemplary application scenario provided in an embodiment of this application, such as... Figure 1As shown, this application scenario may include electronic device 110 and network device 120. Electronic device 110 can establish a connection with network device 120 through a wired network or a wireless network.

[0015] For example, electronic device 110 may be a desktop computer, laptop computer, tablet computer, etc., but is not limited thereto; network device 120 may be a terminal device or a server, but is not limited thereto; in one embodiment of this application, electronic device 110 may send a vascular tree 3D mask construction processing request to network device 120, the processing request carrying target vascular tree structure image data, vascular centerline data and corresponding segment information; further, electronic device 110 may receive vascular tree 3D mask construction response data sent by network device 120, the response data including sampling line segment information, the set of physical radii of the inscribed circles of each vascular segment, vascular segment geometric hash signature, local vascular mask and the finally constructed target vascular tree structure 3D mask.

[0016] also, Figure 1 An electronic device 110 and a network device 120 are provided as examples, but other numbers of electronic devices and network devices may be included in practice, and this application does not limit this.

[0017] In another possible implementation, the technical solution of this application can also be executed by the aforementioned development tool environment 110, or the technical solution of this application can also be executed by the aforementioned code generation service module 120. This application does not limit this.

[0018] To facilitate understanding of this application, the following description, in conjunction with the accompanying drawings, illustrates a method for constructing a three-dimensional mask for a vascular tree, as provided in an embodiment of this application.

[0019] See Figure 2 As shown, this figure is a flowchart illustrating the implementation of a 3D mask construction method for a vascular tree according to an embodiment of this application. Figure 2 As shown, the method may include S201-S205: S201: Obtain multiple centerlines in the target vascular tree structure, as well as the segmentation information corresponding to each centerline.

[0020] To accurately construct a 3D mask of a vascular tree with a complete structure and clear segmentation, the first step is to collect basic data and perform structured analysis on the target vascular tree structure to obtain multiple center lines in the target vascular tree and the segmentation information corresponding to each center line.

[0021] The target vascular tree structure includes the aorta, multiple main trunk vessels, and multiple branch vessels. The main trunk vessels originate from the aorta and extend outwards, while the branch vessels further branch off from the main trunk vessels, collectively forming a hierarchical vascular tree topology. Each centerline corresponds to a branch vessel, starting at the root of the main trunk to which it belongs and ending at the terminal end of the branch vessel, thus comprehensively representing the course of a single branch vessel. Simultaneously, each centerline is divided into multiple vascular segments according to anatomical location and morphological characteristics. The segment information specifically includes the segment index, segment start coordinates, segment end coordinates, and the coordinate set of all centerline points from the segment start to the segment end, providing precise spatial location and topological basis for subsequent radius calculation, task encapsulation, and mask generation.

[0022] For example, let's analyze a branch vessel on a main trunk originating from the aorta, denoted as "branch vessel A". First, determine the centerline of branch vessel A, starting at the junction (root) on the main trunk and ending at the terminal end of branch vessel A. Then, divide this centerline into multiple consecutive segments, such as segment 1, segment 2, and segment 3. Each segment has its corresponding segment index (e.g., 1, 2, 3), start and end coordinates, and coordinates of all centerline points within the segment. If branch vessel A is divided into 3 segments, then its segment information includes detailed coordinate data for these 3 segments.

[0023] S202: Sample each centerline to obtain the corresponding sampling line segment, and obtain the set of physical radii of the inscribed circle of each blood vessel segment on each sampling line segment based on morphological corrosion detection.

[0024] To accurately obtain the true lumen radius of each segment of a blood vessel and avoid mask generation distortion caused by uneven vessel thickness or irregular shape, while improving processing efficiency while ensuring computational accuracy, the acquired vessel centerlines can be discretized and their radius detected. Uniform or adaptive sampling operations are performed on each centerline to obtain the corresponding sampled segment. Then, a morphological erosion detection algorithm is used to calculate the radius of each vessel segment on each sampled segment, thereby obtaining the set of physical radii of the inscribed circle corresponding to each vessel segment.

[0025] Each sampling line segment corresponds to multiple blood vessel segments. Each sampling line segment consists of multiple centerline sampling points distributed along the centerline. Each blood vessel segment corresponds to multiple consecutive centerline sampling points. Each centerline sampling point is equipped with an inscribed circle physical radius that matches its position, thus forming a complete set of radii that correspond to each point.

[0026] S203: Generate a computation task package for each blood vessel segment, and compute the unique geometric hash signature corresponding to each blood vessel segment in parallel based on the centerline coordinate point set in each computation task package.

[0027] To achieve parallel processing of vessel segmentation, avoid redundant computations, improve the efficiency of 3D mask generation, and ensure the independence and data integrity of the computation process for each vessel segment, a standardized computational unit can be encapsulated for each vessel segment, i.e., a dedicated computational task package is generated for each vessel segment. Then, based on the centerline coordinate point set within each computational task package, a unique geometric hash signature corresponding to each vessel segment is obtained through parallel computation using a hash algorithm, which is used for subsequent segment deduplication and validity screening.

[0028] The computational task package serves as a complete data carrier, integrating all the core information required for the blood vessel segmentation. Specifically, it includes: the segmentation information of the blood vessel segment, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the complete set of centerline coordinate points of the centerline where the blood vessel segment is located, and the set of voxel coordinate points of the original image corresponding to the blood vessel segment. This ensures that each task package can be started, computed, and output results independently, supporting efficient parallel execution of multiple processes.

[0029] S204: Based on the repeated matching relationship of multiple geometric hash signatures, all blood vessel segments are filtered to obtain multiple first target segments, and the local blood vessel mask corresponding to each first target segment is obtained based on the calculation task package corresponding to each first target segment.

[0030] To avoid redundant calculations, eliminate redundant segments, and ensure that vascular masks do not overlap, overflow, or conflict, while improving the accuracy and computational efficiency of local mask generation, the uniqueness and matching characteristics of geometric hash signatures can be used to deduplicate and filter the validity of all vascular segments. By comparing the geometric hash signatures of multiple vascular segments, duplicate, redundant, or spatially overlapping vascular segments are identified and eliminated. The filtering is completed based on the duplicate matching relationship of the signatures, resulting in multiple first target segments without redundancy or overlap. Then, using the computational task package corresponding to each first target segment as the data basis, and combining information such as the centerline coordinates, the physical radius of the inscribed circle, and the voxel coordinates, a local vascular mask corresponding to each first target segment is generated, providing clean and accurate local units for the subsequent fusion of the global 3D vascular tree mask.

[0031] S205: Based on all local vascular masks, as well as the original structural parameters and morphological features of the aortic structure, construct a three-dimensional mask of the target vascular tree structure.

[0032] To construct high-quality, high-precision, and structurally complete 3D vascular tree annotation data for medical imaging model training, the local vascular masks of all segments can be uniformly fused and normalized: the local vascular masks of all first target segments after deduplication, screening, and optimization are spatially stitched and fused, and the fusion results are corrected for integrity, repaired at the root, and normalized at the boundary by combining the original structural parameters and morphological features of the aortic structure. Finally, a non-overlapping, non-redundant, topologically continuous, and anatomically standard target vascular tree 3D mask is generated, which can serve as reliable 3D label data for medical imaging model training, verification, and inference.

[0033] Based on the descriptions in S201-S205, in this embodiment, firstly, multiple centerlines of the target vascular tree and their corresponding segment information are acquired. The target vascular tree consists of the aorta, multiple trunk vessels, and branch vessels. Next, these centerlines are sampled to obtain sampled segments, and the set of physical radii of the inscribed circles corresponding to each vessel segment is obtained using a morphological erosion detection method. Then, a computational task package is generated for each vessel segment. These task packages contain segment information, the number of sampling points, the set of inscribed circle radii, and the relevant coordinate point set. Using this information, the geometric hash signature of each vessel segment is calculated in parallel. By comparing the repetitive matching relationships of multiple geometric hash signatures, multiple first target segments are selected. Based on the computational task packages of these target segments, the corresponding local vascular masks can be obtained. Finally, by integrating all local vascular masks and combining them with the original parameters and morphological features of the aortic structure, a three-dimensional mask of the target vascular tree structure is constructed. This application, by standardizing the acquisition of vascular tree and segment information, and combining adaptive sampling and morphological erosion detection, accurately obtains the vessel radius, realistically restoring the anatomical structure. Simultaneously, task package encapsulation, geometric hash deduplication, and parallel computing are employed to improve efficiency. Combined with local masking and aortic repair, label overflow and noise are suppressed to ensure accurate and complete vascular annotation.

[0034] In one possible implementation, step S202 involves sampling each centerline to obtain the corresponding sampling line segment, specifically including: First, identify the total number of discrete spatial coordinate points that make up each centerline. Then, for centerlines with a total number of coordinates greater than or equal to a preset threshold, perform sparse sampling with a first step size (e.g., 10) to reduce computation while ensuring accuracy; for centerlines with a total number of coordinates less than the threshold, perform dense sampling with a second step size (e.g., 5) to preserve the complete morphological information of small blood vessel segments.

[0035] The first step is longer than the second step, thereby achieving adaptive centerline sampling with long sparse segments and short dense segments.

[0036] See Figure 3 , Figure 3 This application provides a flowchart of a method for obtaining the physical radius of the inscribed circle of a blood vessel. Accordingly, the step S202, which involves obtaining the set of physical radii of the inscribed circles corresponding to each blood vessel segment on each sampling line segment based on morphological erosion detection, can be implemented through steps S301-S304. S301: Sequentially use each blood vessel segment as the second target segment, and sequentially use each centerline sampling point corresponding to the second target segment as the first target sampling point.

[0037] To accurately and point-by-point obtain the true inner diameter of each blood vessel segment at every location, ensuring complete coverage of the entire blood vessel without omissions or skips in radius detection, an ordered traversal process can be performed on the blood vessel segments and centerline sampling points: all blood vessel segments are selected sequentially as the second target segment to be calculated according to a preset order, and within the current second target segment, all corresponding centerline sampling points are selected sequentially as the first target sampling points to be detected. This achieves a stable radius detection process of segment-by-segment processing and point-by-point calculation, ensuring that the accurate physical radius of the inscribed circle can be obtained at each location.

[0038] S302: Extract the local binary image centered on the first target sampling point to obtain the local image.

[0039] To avoid redundant calculations on the entire image, reduce the computational overhead of iterative erosion, and improve the accuracy of radius detection by focusing on local areas of blood vessels, the computational region can be effectively cropped before morphological erosion detection: for the currently determined first target sampling point, the image data containing only the local area around the sampling point is cropped from the original blood vessel image, and the corresponding local binary image is extracted. This allows subsequent erosion operations to be performed only within this small local area, which can significantly improve computational efficiency and avoid interference from surrounding irrelevant tissues on radius detection, ensuring the accuracy and stability of the calculation of the physical radius of the inscribed circle.

[0040] S303: The local image is iteratively eroded using spherical structural elements with progressively increasing radii until the first target sampling point is completely covered by the spherical structural elements. The radius of the spherical structural elements at the time of the last iterative erosion operation is taken as the physical radius of the inscribed circle of the first target sampling point.

[0041] To accurately and robustly obtain the true maximum inscribed circle radius of the blood vessel at each first target sampling point, avoiding overestimation or underestimation of the radius, and adapting to irregular cross-sections and stenotic lesions, adaptive iterative erosion calculation can be performed on the extracted local binary image. By using spherical structuring elements with radii starting from a small value and gradually increasing uniformly, the local image is iteratively eroded repeatedly, continuously shrinking the boundary of the blood vessel region. When the iteration reaches the point where the spherical structuring element exactly covers the blood vessel region where the first target sampling point is located, and the point no longer belongs to the foreground connected region, the erosion stops. The radius of the spherical structuring element at this point is determined as the physical radius of the inscribed circle corresponding to the sampling point, thus truly reflecting the maximum spherical diameter that the blood vessel lumen can accommodate at that location, providing a high-precision radius basis for subsequent mask generation.

[0042] In one possible implementation, to accurately convert the number of iterative erosions into a blood vessel radius that conforms to real physical space, this embodiment introduces a voxel spacing normalization and dual-coefficient compensation mechanism to accurately calculate the radius of the spherical structural element, specifically using the following formula: R=(N+offset)×Avg_Spacing×Compensate; Where R is the radius; N is the number of iterations, which is 0 when the first iteration of erosion is performed; offset is the compensation coefficient, which is 2; Avg_Spacing is the average voxel spacing; and Compensate is the dilation compensation coefficient, which is 1.6.

[0043] S304: Integrate the physical radii of the inscribed circles of all the first target sampling points on the second target segment to obtain the set of physical radii of the inscribed circles corresponding to the second target segment.

[0044] To fully record the vessel thickness information corresponding to each centerline sampling point on the second target segment, forming a continuous, reusable, and queryable radius data sequence, and providing point-by-point radius basis for subsequent local vessel mask generation, the radii of all calculated sampling points within the current segment can be systematically collected: according to the order of the centerline sampling points, the physical radii of the inscribed circles corresponding to all first target sampling points on the second target segment are sequentially summarized, organized, and stored to form an ordered radius dataset corresponding one-to-one with the sampling points. Finally, a set of physical radii of the inscribed circles specific to the second target segment is obtained, ensuring that the segment can accurately match the vessel radius at each location during mask generation.

[0045] Steps S301-S304 involve segment-by-segment and point-by-point morphological corrosion detection to accurately determine the physical radius of the actual inscribed circle of the blood vessel corresponding to each centerline sampling point, and to integrate the radius data according to the blood vessel segments. This provides accurate diameter constraints for subsequent local blood vessel mask generation, effectively avoiding problems such as distorted blood vessel radius estimation, contour breakage, or expansion overflow.

[0046] See Figure 4 , Figure 4 A flowchart illustrating the implementation of a method for caching and mapping storage of blood vessel segment radius features provided in this application embodiment is shown, and can be specifically implemented through steps S401-S403: S401: Count the number of sampling points of the centerline sampling points corresponding to each blood vessel segment, and integrate them based on the physical radius of the inscribed circle of each centerline sampling point corresponding to each blood vessel segment to obtain the radius feature set.

[0047] To standardize the feature extraction of the spatial scale and radius distribution of each vascular segment, forming feature data that can be quickly compared and reused, the basic geometric information of each segment can be quantified and collected first: First, the total number of centerline sampling points contained in each vascular segment is counted to characterize the length and spatial scale of the segment. Then, the physical radii of the inscribed circles corresponding to all centerline sampling points within the segment are integrated sequentially to form an ordered, continuous radius data sequence that corresponds one-to-one with the sampling points. This yields a set of radius features that can completely describe the thickness variation of the vascular segment, providing standardized feature input for subsequent construction of a fast query cache and the reuse of radius data.

[0048] S402: Construct segment index key-value pairs by using the number of sampling points corresponding to each blood vessel segment as the key and the segment index corresponding to each blood vessel segment as the value.

[0049] To achieve rapid indexing, efficient matching, and reuse of duplicate data for vascular segments, a standardized retrieval structure based on geometric features can be established. This allows for the rapid location of processed vascular segments in subsequent calculations, avoiding repeated corrosion detection and radius calculations. The number of sampling points, which intuitively reflects the length of a vascular segment, can be used as the retrieval keyword (key), and the segment index, which uniquely identifies a specific vascular segment, can be used as the identifier (value). Segment index key-value pairs are constructed according to the "key-value" correspondence, forming a simple, efficient, and fast-read / write index system. This provides a reliable index foundation for the mapping and caching of radius feature sets.

[0050] S403: Establish a mapping relationship between the segmented index key-value pairs and the corresponding radius feature sets, and store the mapping relationship in the storage module.

[0051] To achieve persistent storage, rapid retrieval, and efficient reuse of vascular segment radius data, and to avoid repeatedly performing morphological erosion and radius calculations on vascular segments with the same geometric features, thereby further improving the overall processing speed and system stability of the vascular tree, a method can be adopted to associate and bind the constructed segment index key-value pairs with the corresponding radius feature set for that segment. The segment index key-value pairs used for rapid localization serve as the retrieval index, while the radius feature set used for direct retrieval serves as the data content. A stable one-to-one mapping relationship is established between the two, and this mapping relationship is persistently stored in a dedicated storage module. This forms a radius feature caching system that can be read at any time, matched quickly, and is secure and reliable. When processing vascular segments with the same or similar structures subsequently, the radius data can be directly matched and retrieved from the storage module, significantly reducing the overhead of repeated calculations and ensuring the efficient and stable operation of the mask generation process.

[0052] Steps S401-S403 establish a cache mapping relationship with the number of sampling points as the key and the segment index and radius feature as the value. This enables rapid querying and reuse of radius data for segments of the same structural blood vessel, avoids repeated morphological erosion detection, significantly improves the efficiency of whole blood vessel tree radius calculation, and ensures the consistency and stability of radius data calls during subsequent mask generation.

[0053] See Figure 5 , Figure 5 A flowchart illustrating the implementation of a blood vessel segment radius feature retrieval and reuse method provided in this application embodiment is shown, and can be specifically implemented through steps S501-S504: S501: For each second target segment, use the number of sampling points corresponding to the second target segment as the key and the segment index corresponding to the second target segment as the value to construct a search key-value pair.

[0054] To quickly and accurately retrieve historical feature data matching the current second target segment from the storage module, enabling rapid matching and reuse of radius feature sets and avoiding blindly traversing all cached data, a unique retrieval identifier can be pre-built for each second target segment to be processed. Specifically, according to preset rules, the number of sampling points that can characterize the segment length feature is used as the retrieval primary key, and the segment index that can uniquely identify the segment identity is used as the corresponding value. A unified search key-value pair with consistent retrieval rules is independently built for each second target segment, which serves as the standard retrieval basis for subsequent data matching queries in the storage module.

[0055] S502: Query the segmented index key-value pairs that match the search key-value pairs in the storage module.

[0056] To enable rapid matching and reuse of vascular segment radius features based on stored historical mapping data and reduce the overhead of repetitive morphological erosion calculations, after constructing the search key-value pair for the current second target segment, the system can enter the preset storage module. Using this search key-value pair as the retrieval benchmark, it traverses and compares various segment index key-value pairs that have been fixed in the storage module, accurately compares the matching between the two key-value pairs, and determines whether there are historical key-value pairs with highly consistent structures and parameters. This provides a basis for subsequent direct reuse of the radius feature set or re-conducting radius detection.

[0057] S503: If a segmented index key-value pair matching the search key-value pair is found in the storage module, then the segmented index key-value pair is taken as the target key-value pair, and the set of radius features that have a mapping relationship with the target key-value pair is obtained as the set of physical radii of the inscribed circle of the second target segment corresponding to the search key-value pair.

[0058] To fully utilize cached historical calculation results and eliminate the tedious process of repeatedly performing morphological iterative corrosion detection, while ensuring accurate and reliable radius values ​​and significantly reducing computation time, when a segmented index key-value pair that perfectly matches the current search key-value pair is found in the storage module, that matching key-value pair can be identified as the target key-value pair. Then, relying on pre-established mapping relationships, the radius feature set bound to the target key-value pair is directly retrieved and assigned as the set of physical radii of the inscribed circle of the second target segment corresponding to the current search key-value pair. This enables direct reuse of existing feature data, eliminating the need to recalculate the radius point by point, and improving the overall processing efficiency.

[0059] S504: If no matching segment index key-value pair is found in the storage module, then the step S301 of taking each centerline sampling point corresponding to the second target segment as the first target sampling point and subsequent steps is executed until the set of physical radii of the inscribed circle corresponding to the second target segment is obtained.

[0060] To ensure that the radius detection of blood vessel segments without cache matching can still be completed completely and accurately, without missing the inscribed circle radius data of any blood vessel segment, and to ensure the integrity and reliability of the subsequently generated local blood vessel mask, when no segment index key-value pair matching the current search key-value pair can be retrieved in the storage module, it indicates that there is no historical cached feature data for this type of blood vessel segment that can be reused. At this time, step S301 is triggered, which sequentially uses each centerline sampling point corresponding to the second target segment as the first target sampling point and subsequent steps. All centerline sampling points corresponding to the current second target segment are sequentially used as the first target sampling points, and then a series of subsequent steps are executed, such as extracting local binary images, iterative erosion using spherical structural elements, determining the stopping iteration condition, and calculating the physical radius of the inscribed circle of a single point, until the radius calculation of all sampling points is completed and integrated, finally obtaining the set of physical radii of the inscribed circle corresponding to the second target segment. This realizes the de novo solution of the segment radius when there is no cache hit, taking into account both data integrity and algorithm generality.

[0061] Steps S501-S504 achieve cache hit and reuse of vessel segment radius features by first constructing search key-value pairs and then retrieving historical mapping data from the storage module. For segments with existing matching data, the existing radius feature set is directly retrieved, eliminating the need for repeated morphological erosion iteration calculations. For segments without matching data, the sampling point-by-sampling radius detection process is re-executed, balancing overall computational efficiency and computational integrity. This effectively reduces redundant computation and improves the overall generation speed of the vessel tree inscribed circle radius set.

[0062] In one possible implementation, step S204 filters all vascular segments based on the duplicate matching relationship of multiple geometric hash signatures to obtain multiple first target segments, and uses a differential judgment rule to complete the deduplication filtering: The geometric hash signatures corresponding to all blood vessel segments are compared and matched pairwise. For geometric hash signatures that are unique and have independent spatial morphological features, the corresponding blood vessel segments are directly designated as the first target segments. For multiple sets of vascular segments with duplicate signature matching relationships, the first-in retention principle is followed, and the vascular segment generated first in the duplicate signature is retained and determined as the first target segment, while redundant duplicate data is removed. The remaining blood vessel segments that are repeatedly matched in the same geometric hash signature group and whose generation time is later are uniformly classified as non-target segments and are not retained to participate in subsequent mask generation operations.

[0063] This screening rule can retain all blood vessel segments with unique features, and can also perform temporal selection to retain and remove redundant segments with spatial overlap and feature redundancy, effectively avoiding subsequent problems such as local blood vessel mask overlap, redundant calculation, and disordered topology, and ensuring the uniqueness and structural rationality of the first target segment after screening.

[0064] See Figure 6 , Figure 6 The flowchart illustrates the implementation of a method for extracting local blood vessel masks in segmented blood vessels, as provided in this application embodiment. Accordingly, step S204, which involves obtaining the local blood vessel mask corresponding to each first target segment based on the computational task package corresponding to each first target segment, can be implemented through steps S601-S606. S601: Each first target segment is sequentially treated as a segment to be processed.

[0065] To ensure the orderly completion of local mask extraction for all first target segments and avoid omissions, duplications, or disordered processing, and to guarantee a standardized, controllable, and logically rigorous mask extraction process, all first target segments after screening can be traversed and scheduled in a predetermined order: each first target segment is selected in turn as the current segment to be processed, and each segment is processed in a single-segment focused manner, entering subsequent computational stages such as mask construction, bounding box generation, and connected component filtering. This ensures that each first target segment can complete the mask extraction process completely, achieving standardized processing of all segments without omissions or errors.

[0066] S602: Calculate the physical radius buffer of the segment to be processed, and construct the local bounding box of the segment to be processed based on the physical radius buffer of the segment to be processed.

[0067] To precisely define the computational range of the segment to be processed, reduce interference from irrelevant regions, and ensure the accuracy and rationality of local mask extraction while avoiding waste of computational resources, after determining the segment to be processed, the physical radius buffer of the segment can be calculated first, and then a local bounding box can be constructed based on this, laying the foundation for the accurate generation of the subsequent local mask.

[0068] Specifically, the physical radius buffer of the segment to be processed is first calculated using a preset formula. Buffer Subsequently, using this physical radius buffer as a reference, and combining it with the centerline points of all the segments to be processed... x, y, z Three-dimensional coordinates ( , , The local bounding box is calculated using a preset formula. x, y, z The maximum and minimum coordinates in three directions are specified, while the coordinate range is limited to not exceeding the width of the original 3D image. W ),long( H ) and number of layers ( D This ensures that the local bounding box can completely encompass the entire area of ​​the segment to be processed, while also accurately defining the calculation range. This reduces the redundancy of subsequent calculations, allowing subsequent mask generation and region filtering to focus only on the bounding box, thus improving computational efficiency and accuracy.

[0069] In one possible implementation, the formula for calculating the physical radius buffer is as follows: , Buffer Buffer for physical radius; The average diameter of the first target segment; ， ， The first target segment is in x, y, z Spacing of direction.

[0070] In one possible implementation, the formula for constructing the local bounding box is as follows: ; ; ; ; ; ; in, , , These are the x-axis coordinates, y-axis coordinates, and z-axis coordinates of all centerline points in the first target segment. W, H, D These represent the width, length, and number of layers of the original three-dimensional image of the target vascular tree structure, respectively.

[0071] S603: Determine the initial mask for the segment to be processed based on the local bounding box and computational task package of the segment to be processed.

[0072] In order to accurately locate the mask generation range of the segment to be processed and quickly generate an initial mask that fits the blood vessel morphology based on the core parameters in the computation task package, laying the foundation for subsequent mask optimization, after constructing the local bounding box of the segment to be processed, the initial mask can be determined together with the pre-configured computation task package.

[0073] S604: Take the Boolean intersection of the initial mask and the original vascular base map of the segment to be processed to obtain M connected regions.

[0074] To eliminate invalid regions in the initial mask that do not conform to the actual vascular anatomy and filter out interference noise, ensuring that the final generated local vascular mask fits the real vascular morphology and is practical, after obtaining the initial mask of the segment to be processed, Boolean intersection operation is used to fuse and match the initial mask with the original vascular base image of the segment to be processed, and to filter out the effective regions that both conform to the outline of the initial mask and exist in the original vascular image, finally obtaining several independent connected regions (i.e., M connected regions, where M is a positive integer).

[0075] The core logic of Boolean intersection operation is to retain regions that coexist and overlap in the initial mask and the original vascular image, and to remove false regions in the initial mask that are outside the range of the original vascular image or do not match the real vascular structure (such as non-vascular tissues or background noise mistakenly included in the initial mask). At the same time, it also excludes regions in the original vascular image that do not belong to the current segment to be processed, ensuring that the finally obtained connected regions are all effective regions that fit the actual vascular structure and have actual physiological significance. This provides a basis for subsequent selection of the optimal mask and avoidance of redundancy and errors, ensuring the accuracy of the mask and laying the foundation for subsequent selection and optimization of connected regions.

[0076] It should be noted that in this application, the largest connected component is defined as the connected component with the largest spatial volume or the largest number of voxels.

[0077] S605: If M equals 1, then the connected region is used as the local vascular mask for the segment to be processed.

[0078] To directly determine a standard and compliant local vascular mask when only a single, complete, and effective connected region exists, eliminating the need for additional screening and judgment steps, while ensuring the integrity, continuity, and anatomical accuracy of the mask, when the number of connected regions M obtained after performing a Boolean intersection between the initial mask and the original vascular base map equals 1, it indicates that the overall contour of the connected region and the vascular orientation completely fit the current segment to be processed, with no separation fragments, no isolated noise interference, continuous morphology, and clear boundaries, accurately representing the real vascular lumen structure of the segment. At this point, there is no need for redundant screening and domain selection processing; the single connected region can be directly used as the final local vascular mask for the segment to be processed. This simplifies the calculation process, improves processing efficiency, and ensures that the output mask highly matches the actual vascular anatomy.

[0079] S606: If M is greater than 1, then the largest connected region in the M connected regions is used as the local vascular mask of the segment to be processed.

[0080] To accurately remove noise fragments, small pseudo-branches, and isolated irrelevant regions when multiple scattered connected regions appear, and to retain the most complete and largest vascular trunk region, ensuring the authenticity and structural integrity of the local vascular mask, when the number of connected regions M obtained after taking the Boolean intersection of the initial mask and the original vascular base map is greater than 1, it indicates the existence of multiple independent connected regions, which include both the main vascular trunk region and invalid regions such as image noise, small pseudo-branches, and edge fragments. In this case, following the principle of prioritizing the main trunk and selecting the largest region, the M connected regions are sorted according to the voxel size or spatial area occupied, and the largest connected region is selected as the core vascular region. Other regions that are too small or belong to interference fragments are discarded. This avoids the mask splitting and contour distortion problems caused by multiple connected regions, and obtains a standard local vascular mask that fits the actual vascular trunk direction and lumen morphology.

[0081] By passing through steps S601-S606, the calculation range is reduced by using local bounding boxes, the boundaries are constrained by Boolean intersections, and noise is removed by using the maximum connected component. This allows for the accurate extraction of local vascular masks that are free of overflow, burrs, and have complete structures, ensuring that the mask strictly conforms to the real vascular anatomy.

[0082] In one possible implementation, the radii in the set of physical radii of the inscribed circle correspond to the sampling points of the center line; each sampling line segment corresponds to multiple blood vessel segments.

[0083] See Figure 7 , Figure 7 This is a flowchart illustrating the implementation of an initial mask generation method provided in this application embodiment. Accordingly, step S603, which determines the initial mask of the segment to be processed based on the local bounding box of the segment to be processed and the computation task package, can be implemented through steps S701-S704: S701: Convert all voxel coordinate points in the original image of the segmented computation task package to be processed, which are covered by the local bounding box, into physical coordinates.

[0084] To eliminate computational errors caused by voxel size differences and unify spatial measurement standards, ensuring that subsequent Euclidean distance calculations have a true physical scale rather than being limited to voxel scales, and guaranteeing the accuracy of radius determination and mask contour generation, the voxel coordinates of all voxel points within the local bounding box and the original image contained in the computation task package can be converted from image voxel coordinates to actual physical coordinates one by one. Since the voxel spacing in medical images is often inconsistent in the x, y, and z directions, and the spatial reference corresponding to each voxel unit is also different, directly using integer voxel coordinates for distance calculations will result in scale distortion. Therefore, based on the original image parameters, each voxel point within the bounding box can be mapped from row and column layer integer coordinates to true spatial coordinates, achieving standardized conversion from voxel coordinates to physical 3D coordinates. This ensures that subsequent distance calculations and radius threshold comparisons are based on real millimeters or spatial units, guaranteeing global scale consistency and rigorous, unbiased calculations, laying a unified foundation for subsequent Euclidean distance calculations and vascular region screening.

[0085] S702: For each physical coordinate, calculate the Euclidean distance from that physical coordinate to the corresponding second target sampling point to obtain the physical distance corresponding to that physical coordinate.

[0086] To accurately determine the distance of each spatial physical coordinate relative to the vessel centerline, and to measure the spatial relationship between the point and the vessel centerline using real physical scales, providing a reliable basis for subsequent radius threshold comparisons, all voxel points are uniformly converted to physical coordinates. Then, each physical coordinate is matched one by one with its corresponding centerline sampling point. Specifically, using all centerline sampling points on the sampling line segment to be processed as a reference, the sampling point closest to the current physical coordinate spatial position is selected as the second target sampling point. Finally, the three-dimensional Euclidean distance between this physical coordinate and the second target sampling point is calculated to obtain a physical distance with a real spatial scale.

[0087] By matching the centerline sampling point with the nearest point, each physical coordinate can be bound to the most suitable center point of the blood vessel. Then, the spatial interval between the two is quantified by Euclidean distance. This method can strictly follow the overall direction of the blood vessel and the local morphological characteristics, avoiding the judgment deviation caused by using a fixed reference point. It provides accurate and quantified distance parameters to support the subsequent screening of blood vessel regions based on the inscribed circle radius threshold.

[0088] S703: Determine the physical radius of the inscribed circle corresponding to the second target sampling point from the set of physical radii of the inscribed circles of the segmented calculation task package to be processed, and use it as the radius threshold.

[0089] To establish a quantitative standard for determining the physical distance and the boundary of the blood vessel lumen, and to ensure that the target coordinates selected in the subsequent screening accurately match the actual lumen thickness at each position of the segment to be processed, thus avoiding the problem of the boundary being too wide or too narrow, after determining the second target sampling point corresponding to each physical coordinate, the physical radius of the inscribed circle corresponding to the second target sampling point can be accurately matched and extracted based on the set of inscribed circle physical radii stored in the task package of the segment to be processed, and used as the radius threshold of the current physical coordinate.

[0090] Using this radius as a threshold, "point-by-point adaptive" boundary determination can be achieved. That is, different physical coordinates correspond to different radius thresholds, which perfectly match the thickness variation of the blood vessel lumen along the center line. This provides accurate and realistic quantitative basis for subsequent judgment of whether the physical coordinates belong to the blood vessel lumen area, ensuring that the outline of the initial mask is highly consistent with the actual shape of the blood vessel.

[0091] It should be noted that there is a one-to-one correspondence between the set of physical radii of the inscribed circle and the sampling points of the center line. Each sampling point of the center line (including the second target sampling point) has a unique corresponding physical radius of the inscribed circle. This radius can accurately represent the actual thickness of the blood vessel lumen at the corresponding sampling point location.

[0092] S704: Determine the physical coordinates corresponding to physical distances less than the corresponding radius threshold as target coordinates, and perform voxelization on all target coordinates to obtain the initial mask.

[0093] To ensure that the initial mask accurately fits the actual vascular morphology of the segment to be processed, eliminates invalid interference areas, clarifies the effective range of the blood vessels, and provides a standardized basis for subsequent global fusion and modeling, the previously determined radius threshold can be used as the core judgment criterion to filter and integrate the physical coordinates, and finally generate an initial mask that meets the requirements.

[0094] Specifically, first, the radius threshold corresponding to each physical coordinate (i.e., the radius of the inscribed circle corresponding to the second target sampling point) is determined. Each physical coordinate is compared with its corresponding radius threshold, and physical coordinates with a physical distance less than the corresponding radius threshold are selected. These coordinates that meet the conditions are determined as target coordinates. The area where these coordinates are located belongs to the effective range of the blood vessel lumen and can accurately reflect the actual direction and size of the blood vessel lumen. Then, all the selected target coordinates are summarized, and the coordinates are standardized through voxelization. After voxelization, they are integrated into a continuous region, thus forming the initial mask.

[0095] This process ensures that the initial mask can completely cover the main area of ​​the blood vessel, while effectively eliminating invalid points that exceed the radius threshold and avoiding interference from irrelevant areas. This ensures the accuracy and practicality of the initial mask and lays a solid foundation for subsequent steps such as connected component screening and global fusion.

[0096] By using steps S701-S704 to achieve distance discrimination and radius constraint based on the local bounding box and the nearest centerline sampling point, an initial mask that fits the actual diameter and direction of the blood vessel can be generated quickly and accurately, providing a reliable foundation for subsequent local vascular mask optimization.

[0097] See Figure 8 , Figure 8 The flowchart illustrates the implementation of a three-dimensional mask fusion and stitching method for a vascular tree provided in this application embodiment. Accordingly, step S205, based on all local vascular masks and the original structural parameters and morphological features of the aortic structure, constructs a three-dimensional mask for the target vascular tree structure, which can be achieved through steps S801-S804. S801: Merge all local vascular masks to obtain a global fusion mask.

[0098] To integrate the scattered local vascular masks extracted independently from each vascular segment into a whole spatial structure, eliminate the fragmented and independent state between segments, realize the spatial alignment and regional connectivity of branches at all levels, and provide a complete base for subsequent optimization of the overall vascular tree structure and correction of the aortic region, all the selected local vascular masks can be registered and merged block by block according to their corresponding spatial coordinate positions.

[0099] Specifically, while maintaining the original contour, boundary and voxel properties of each local mask, the masks are superimposed and fused according to the spatial position of the three-dimensional image. Adjacent segments are seamlessly connected and overlapping areas are merged and unified to form a global fusion mask that covers the entire range of blood vessel branches and has a unified spatial position.

[0100] This approach balances the local accuracy of a single segmented mask with the topological continuity of the overall vascular tree, avoiding structural discontinuities and positional shifts caused by segmented processing. It lays a complete and reliable global data foundation for subsequent removal of redundant aortic regions and reconstruction of the standard aortic mask.

[0101] S802: Remove redundant mask voxel regions belonging to the aorta from the global fusion mask to obtain the vascular tree base mask.

[0102] To eliminate redundant voxels, excessive bulges, and false spread areas in the aortic region caused by multi-segment superposition, boundary expansion, and branch overlap in the global fusion mask, and to avoid aortic contour hypertrophy, boundary distortion, and disordered adhesion with branch masks, the aortic region can be accurately located after all local vascular masks are merged to obtain the global fusion mask. This allows for the identification and removal of mask regions containing non-real anatomical structures such as redundant voxels, invalid bulges, repeated superpositions, and pseudo-connections at the edges.

[0103] While preserving the complete morphology of the effective mask for each level of coronary artery branch, only the redundant areas of the aorta are cleaned up and reduced, and unreasonable mask voxels are removed. Finally, a basic vascular tree mask with a clean structure, regular boundaries, and clear topological relationships is obtained, laying a pure and interference-free foundation for subsequent replacement and splicing with the standard aortic morphology.

[0104] S803: Using the original structural parameters and morphological features of the aortic structure, extract the aortic mask segment in the corresponding region of the aorta.

[0105] To overcome the problems of redundant distortion and boundary deformities in the aortic region in the original global fusion mask, this method reconstructs a complete aortic mask with standard contours, realistic morphology, and regular boundaries, based on the inherent original structural parameters and anatomical morphological features of the aorta itself. It can retrieve the original structural parameters and inherent morphological features corresponding to the aortic structure, including key information such as aortic diameter, course curvature, origin and termination positions, slice thickness distribution, and anatomical contour curvature. Using these prior parameters and morphological features as constraints, the corresponding spatial region of the aorta is accurately located in the 3D image. According to its true physiological structural trend and lumen scale, the method accurately segments and extracts aortic mask segments with standard morphology, smooth edges, and conformity to the actual anatomical structure.

[0106] This method does not rely on the superposition results of previous local masks, but performs independent and precise reconstruction based on the original morphology of the aorta. It can fundamentally avoid the bulging, distortion, and redundancy defects caused by previous segmentation and merging, and provide high-precision standard aortic segments for subsequent replacement, splicing, and construction of a complete and regular vascular tree 3D mask.

[0107] S804: The aortic mask segment is spliced ​​with the vascular tree base mask at corresponding positions to obtain the complete vascular tree mask of the target vascular tree structure.

[0108] To achieve seamless integration, topological coherence, and anatomical uniformity of the aortic trunk and its branches in three-dimensional space, and to correct the redundancy and distortion in the original mask of the aortic region while preserving the true details of each sub-branch, the reconstructed aortic mask fragment and the base vascular tree mask (after removing redundancy) are precisely aligned and stitched together according to three-dimensional spatial coordinates. Using the spatial anatomical position as a benchmark, the original incomplete and distorted aortic region is replaced with a well-formed, standard-boundary aortic mask fragment. This fragment is then seamlessly integrated with the coronary artery branches in the base vascular tree mask, with smoothed edges to eliminate stitching gaps, tomographic misalignments, and abrupt boundaries. This ensures a natural connection and structural integrity between the aortic trunk and its branches, ultimately forming a topologically continuous, anatomically standard, and smoothly contoured three-dimensional mask of the target vascular tree structure, meeting the accuracy requirements for subsequent vascular measurement, structural analysis, and 3D modeling.

[0109] Steps S801-S804 first fuse the local vascular masks to form a global basic framework, then remove redundant areas of the aorta, accurately extract aortic mask segments, and finally complete the splicing and integration. This effectively avoids redundant artifacts and boundary misalignments caused by local mask fusion, while ensuring the anatomical integrity and accuracy of the aortic structure. Ultimately, a three-dimensional mask of the target vascular tree with a clear topological structure, smooth connection of each vascular segment, and conformity to the real anatomical morphology is constructed, ensuring the overall accuracy and integrity of the mask.

[0110] In one possible implementation, the method further includes: For blood vessel segments that meet the preset screening requirements, priority labels are marked on their corresponding calculation task packages one by one.

[0111] The preset screening requirements are specifically designed to identify vascular segments that have significant reference value or key assessment value for clinical diagnosis. By adding priority labels to the computational task packages for these high-value vascular segments, priority processing, priority scheduling, and priority computation can be achieved in subsequent task scheduling, mask reconstruction, and feature analysis processes. This ensures the processing accuracy and response efficiency of key clinical vascular segments, facilitating subsequent targeted assessment of vascular lesions, anatomical structure analysis, and clinical diagnostic reference.

[0112] In one possible implementation, step S203 involves parallel computation of a unique geometric hash signature corresponding to each blood vessel segment based on the centerline coordinate point set in each computation task package, specifically including: Parallel deordering and standardization processing is first performed on the voxel coordinate point set in the computation task package marked with priority labels, and then parallel deordering and standardization processing is performed on the voxel coordinate point set in the computation task package without priority labels. Finally, a unique geometric hash signature corresponding to each blood vessel segment is generated and obtained.

[0113] See Figure 9 As shown, Figure 9 This is a schematic diagram of a 3D mask construction device for a vascular tree provided in an embodiment of this application. Figure 9 As shown, the vascular tree 3D mask construction device includes: Information acquisition unit 901 is used to acquire multiple centerlines in the target vascular tree structure, as well as segment information corresponding to each centerline; the target vascular tree structure includes the aorta, multiple main trunk vessels, and multiple branch vessels. The main trunk vessels originate from the aorta and extend outward, while the branch vessels branch off from the main trunk vessels; the starting point of the centerline is the root of the main trunk vessel to which the branch vessel belongs, and the ending point of the centerline is the terminal end of the branch vessel; each centerline corresponds to multiple vascular segments; the segment information includes the segment index of each vascular segment, the coordinates of the segment's starting point, the coordinates of the segment's ending point, and the coordinates of all centerline points from the segment's starting point to its ending point; The sampling unit 902 is used to sample each centerline to obtain the sampling line segment corresponding to each centerline; each sampling line segment is composed of multiple centerline sampling points, and each blood vessel segment corresponds to multiple centerline sampling points; The radius detection unit 903 is used to obtain the set of physical radii of the inscribed circles of each blood vessel segment on each sampling line segment based on morphological corrosion detection. The task package generation unit 904 is used to generate a calculation task package for each blood vessel segment. The calculation task package includes the segment information of the blood vessel segment, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the set of centerline coordinate points of the centerline where the blood vessel segment is located, and the set of voxel coordinate points of the original image corresponding to the blood vessel segment. Hash calculation unit 905 is used to calculate the unique geometric hash signature corresponding to each blood vessel segment in parallel based on the centerline coordinate point set in each calculation task package; The segmentation filtering unit 906 is used to filter all vascular segments based on the duplicate matching relationship of multiple geometric hash signatures to obtain multiple first target segments; The local mask generation unit 907 is used to obtain the local blood vessel mask corresponding to each first target segment based on the computation task package corresponding to each first target segment. The three-dimensional mask construction unit 908 is used to construct a three-dimensional mask of the target vascular tree structure based on all local vascular masks and the original structural parameters and morphological features of the aortic structure.

[0114] In one possible implementation, the sampling unit 902 is specifically used for: Identify the total number of discrete spatial coordinate points that constitute each centerline; For centerlines whose total number of coordinates is greater than or equal to the number threshold, sample the corresponding sampled line segments according to the first step length. For centerlines whose total number of coordinates is less than the number threshold, sample the corresponding sampled line segments according to the second step size; Wherein, the length of the first step is greater than the length of the second step.

[0115] In one possible implementation, the radius detection unit 903 is specifically used for: Each blood vessel segment is sequentially used as the second target segment, and each centerline sampling point corresponding to the second target segment is sequentially used as the first target sampling point. A local binary image is obtained by extracting a local image centered on the first target sampling point; The local image is iteratively eroded using spherical structural elements with progressively increasing radii until the first target sampling point is completely covered by the spherical structural elements. The radius of the spherical structural elements at the time of the last iteration of erosion is taken as the physical radius of the inscribed circle of the first target sampling point. The physical radii of the inscribed circles of all the first target sampling points on the second target segment are integrated to obtain the set of physical radii of the inscribed circles corresponding to the second target segment.

[0116] In one possible implementation, the device further includes: The feature statistics unit is used to count the number of sampling points of the centerline sampling points corresponding to each blood vessel segment, and to integrate them based on the physical radius of the inscribed circle of each centerline sampling point corresponding to each blood vessel segment to obtain a radius feature set. The index building unit is used to construct segment index key-value pairs by using the number of sampling points corresponding to each blood vessel segment as the key and the segment index corresponding to each blood vessel segment as the value. The mapping storage unit is used to establish a mapping relationship between segmented index key-value pairs and the corresponding radius feature sets, and to store the mapping relationship in the storage module.

[0117] In one possible implementation, the device further includes: The search key-value pair construction unit is used to construct a search key-value pair for each second target segment, using the number of sampling points corresponding to the second target segment as the key and the segment index corresponding to the second target segment as the value. The cached matching query unit is used to query the segmented index key-value pairs that match the search key-value pairs in the storage module; The feature reuse unit is used to, if a segmented index key-value pair matching the search key-value pair is found in the storage module, take the segmented index key-value pair as the target key-value pair and obtain the set of radius features that have a mapping relationship with the target key-value pair as the set of physical radii of the inscribed circle of the second target segment corresponding to the search key-value pair; The detection execution unit is used to execute the step of taking each centerline sampling point corresponding to the second target segment as the first target sampling point and subsequent steps if no segment index key-value pair matching the search key-value pair is found in the storage module, until the set of physical radii of the inscribed circle corresponding to the second target segment is obtained.

[0118] In one possible implementation, the segmented filtering unit 906 is specifically used for: The vascular segment corresponding to the geometric hash signature that does not repeat with other geometric hash signatures among multiple geometric hash signatures is determined as the first target segment; The blood vessel segment that is duplicated with other geometric hash signatures and was generated first among multiple geometric hash signatures is identified as the first target segment; Vessel segments that are duplicates of other geometric hash signatures and are not the first generated segment are identified as non-target segments.

[0119] In one possible implementation, the local mask generation unit 907 is specifically used for: Each of the first target segments is sequentially treated as segments to be processed. Calculate the physical radius buffer of the segment to be processed, and construct the local bounding box of the segment to be processed based on the physical radius buffer of the segment to be processed; The initial mask for the segment to be processed is determined based on the local bounding box of the segment to be processed and the computation task package. The Boolean intersection of the initial mask and the original vascular map of the segment to be processed yields M connected regions; M is a positive integer. If M equals 1, then the connected region is used as the local vascular mask of the segment to be processed; if M is greater than 1, then the largest connected region among the M connected regions is used as the local vascular mask of the segment to be processed.

[0120] In one possible implementation, the radii in the set of physical radii of the inscribed circle correspond to the sampling points of the center line; each sampling line segment corresponds to multiple blood vessel segments.

[0121] In one possible implementation, determining the initial mask for the segment to be processed based on the local bounding box and computational task package includes: Convert all voxel coordinate points in the original image of the segmented computation task package to be processed, which are covered by the local bounding box, into physical coordinates. For each physical coordinate, calculate the Euclidean distance from that physical coordinate to the corresponding second target sampling point to obtain the physical distance corresponding to that physical coordinate; the second target sampling point is the centerline sampling point on the sampling line segment corresponding to the segment to be processed that is closest to the physical coordinate; The physical radius of the inscribed circle corresponding to the second target sampling point is determined from the set of physical radii of the inscribed circles of the segmented calculation task package to be processed, and used as the radius threshold. The physical coordinates corresponding to physical distances less than the corresponding radius threshold are determined as target coordinates, and all target coordinates are voxelized to obtain the initial mask.

[0122] In one possible implementation, the three-dimensional mask construction unit 908 is specifically used for: All local vascular masks are merged to obtain a globally fused mask; The redundant mask voxel regions belonging to the aorta in the global fusion mask are removed to obtain the vascular tree base mask. Using the original structural parameters and morphological features of the aortic structure, the aortic mask segment corresponding to the aortic region is extracted; The aortic mask fragment is spliced ​​with the vascular tree base mask at corresponding positions to obtain the complete vascular tree mask of the target vascular tree structure.

[0123] In one possible implementation, the device further includes: Priority labeling unit is used to label the calculation task package corresponding to the blood vessel segment that meets the preset screening requirements with priority labels.

[0124] In one possible implementation, the hash calculation unit 905 is specifically used for: First, the voxel coordinate point set in the computation task package labeled with priority is processed in parallel to deorder and standardize. Then, the voxel coordinate point set in the computation task package without priority label is processed in parallel to deorder and standardize, so as to obtain the unique geometric hash signature corresponding to each blood vessel segment.

[0125] In addition, this application embodiment also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the vascular tree 3D mask construction method as described above.

[0126] This application's embodiments standardize the hierarchical structure of the vascular tree, unify the rules for acquiring centerline and segment information, and combine adaptive centerline sampling and morphological erosion detection to accurately obtain the physical radius of the inscribed circle of each vascular segment. This constructs a vascular diameter estimation system that closely matches the actual anatomical morphology, effectively overcoming the problems of radius estimation distortion, misjudgment at stenosis, and breakage at bifurcation points in traditional methods. It can realistically reproduce the changes in vascular lumen thickness while completely preserving lesions and anatomical details, providing precise geometric constraints for 3D mask construction. Simultaneously, using standardized computational task package encapsulation and geometric hash signature deduplication as the core, parallel computation and intelligent filtering are implemented for each vascular segment, abandoning the traditional serial traversal mode, significantly reducing the overhead of repeated reading and writing of 3D voxels, and significantly improving the construction efficiency of complex vascular trees. Furthermore, through local mask generation, Boolean intersection constraints, and aortic structure fusion repair, label expansion overflow, discrete spurs, and artifact interference are suppressed from the source, avoiding encroachment on adjacent non-target tissues and ensuring clear vascular annotation boundaries and complete and continuous anatomical structures.

[0127] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0128] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing a three-dimensional mask for a vascular tree, characterized in that, The method includes: Obtain multiple centerlines from the target vascular tree structure, along with segment information corresponding to each centerline. The target vascular tree structure includes the aorta, multiple main trunk vessels, and multiple branch vessels. The main trunk vessels originate from the aorta and extend outwards, while the branch vessels branch off from the main trunk vessels. The starting point of each centerline is the root of the main trunk vessel to which the branch vessel belongs, and the ending point of each centerline is the terminal end of that branch vessel. Each centerline corresponds to multiple vascular segments. The segment information includes the segment index, the coordinates of the segment's starting point, the coordinates of the segment's ending point, and the coordinates of all centerline points from the segment's starting point to its ending point. Each centerline is sampled to obtain the corresponding sampling line segment, and the set of physical radii of the inscribed circle corresponding to each blood vessel segment on each sampling line segment is obtained based on morphological corrosion detection; each sampling line segment is composed of multiple centerline sampling points, and each blood vessel segment corresponds to multiple centerline sampling points; A computational task package is generated for each blood vessel segment, and a unique geometric hash signature corresponding to each blood vessel segment is calculated in parallel based on the centerline coordinate point set in each computational task package. The computational task package includes the segment information of the blood vessel segment, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the set of centerline coordinate points of the centerline where the blood vessel segment is located, and the set of voxel coordinate points of the original image corresponding to the blood vessel segment. Multiple first target segments are obtained by filtering all blood vessel segments based on the repeated matching relationship of multiple geometric hash signatures, and the local blood vessel mask corresponding to each first target segment is obtained based on the computation task package corresponding to each first target segment. Based on all local vascular masks, as well as the original structural parameters and morphological features of the aortic structure, a three-dimensional mask of the target vascular tree structure is constructed.

2. The method of claim 1, wherein, The step of sampling each centerline to obtain the corresponding sampling line segment includes: Identify the total number of discrete spatial coordinate points that constitute each centerline; For centerlines whose total number of coordinates is greater than or equal to the number threshold, sample the corresponding sampled line segments according to the first step length. For centerlines whose total number of coordinates is less than the number threshold, sample the corresponding sampled line segments according to the second step size; Wherein, the length of the first step is greater than the length of the second step.

3. The method according to claim 1, characterized in that, The set of physical radii of the inscribed circles corresponding to each blood vessel segment on each sampling line segment, obtained based on morphological corrosion detection, includes: Each blood vessel segment is sequentially used as the second target segment, and each centerline sampling point corresponding to the second target segment is sequentially used as the first target sampling point. A local binary image is obtained by extracting a local image centered on the first target sampling point; The local image is iteratively eroded using spherical structural elements with progressively increasing radii until the first target sampling point is completely covered by the spherical structural elements. The radius of the spherical structural elements at the time of the last iteration of erosion is taken as the physical radius of the inscribed circle of the first target sampling point. The physical radii of the inscribed circles of all the first target sampling points on the second target segment are integrated to obtain the set of physical radii of the inscribed circles corresponding to the second target segment.

4. The method according to claim 3, characterized in that, The method further includes: The number of sampling points for the centerline sampling points corresponding to each blood vessel segment is counted, and the radius feature set is obtained by integrating the inscribed circles of each centerline sampling point corresponding to each blood vessel segment. Construct segment index key-value pairs using the number of sampling points corresponding to each blood vessel segment as the key and the segment index corresponding to each blood vessel segment as the value. Establish a mapping relationship between the segmented index key-value pairs and the corresponding radius feature sets, and store the mapping relationship in the storage module.

5. The method according to claim 3, characterized in that, Before sequentially using the centerline sampling points corresponding to the second target segment as the first target sampling points, the method further includes: For each second target segment, the number of sampling points corresponding to the second target segment is used as the key, and the segment index corresponding to the second target segment is used as the value to construct a search key-value pair; In the storage module, query the segmented index key-value pairs that match the search key-value pairs; If a segmented index key-value pair matching the search key-value pair is found in the storage module, then the segmented index key-value pair is taken as the target key-value pair, and the set of radius features that have a mapping relationship with the target key-value pair is obtained as the set of physical radii of the inscribed circle of the second target segment corresponding to the search key-value pair. If no matching segment index key-value pair is found in the storage module, the step of taking each centerline sampling point corresponding to the second target segment as the first target sampling point and subsequent steps is executed until the set of physical radii of the inscribed circle corresponding to the second target segment is obtained.

6. The method according to claim 1, characterized in that, The process of filtering all vascular segments based on repeated matching relationships using multiple geometric hash signatures yields multiple first target segments, including: The vascular segment corresponding to the geometric hash signature that does not repeat with other geometric hash signatures among multiple geometric hash signatures is determined as the first target segment; The blood vessel segment that is duplicated with other geometric hash signatures and is generated first among multiple geometric hash signatures is identified as the first target segment. Vessel segments that are duplicates of other geometric hash signatures and are not the first generated segment are identified as non-target segments.

7. The method according to claim 1, characterized in that, The step of obtaining the local vascular mask corresponding to each first target segment based on the computational task package corresponding to each first target segment includes: Each of the first target segments is sequentially treated as segments to be processed. Calculate the physical radius buffer of the segment to be processed, and construct the local bounding box of the segment to be processed based on the physical radius buffer of the segment to be processed; The initial mask for the segment to be processed is determined based on the local bounding box of the segment to be processed and the computation task package. The Boolean intersection of the initial mask and the original vascular base map of the segment to be processed yields M connected regions; M is a positive integer. If M equals 1, then the connected region is used as the local vascular mask of the segment to be processed; if M is greater than 1, then the largest connected region among the M connected regions is used as the local vascular mask of the segment to be processed.

8. The method according to claim 7, characterized in that, The radii in the set of physical radii of the inscribed circle correspond to the sampling points on the centerline; each sampling line segment corresponds to multiple blood vessel segments. The determination of the initial mask for the segment to be processed based on the local bounding box and computational task package includes: Convert all voxel coordinate points in the original image of the segmented computation task package to be processed, which are covered by the local bounding box, into physical coordinates. For each physical coordinate, calculate the Euclidean distance from that physical coordinate to the corresponding second target sampling point to obtain the physical distance corresponding to that physical coordinate; the second target sampling point is the centerline sampling point on the sampling line segment corresponding to the segment to be processed that is closest to the physical coordinate; The physical radius of the inscribed circle corresponding to the second target sampling point is determined from the set of physical radii of the inscribed circles of the segmented calculation task package to be processed, and used as the radius threshold. The physical coordinates corresponding to physical distances less than the corresponding radius threshold are determined as target coordinates, and all target coordinates are voxelized to obtain the initial mask.

9. The method according to claim 1, characterized in that, The construction of a three-dimensional mask for the target vascular tree structure based on all local vascular masks, as well as the original structural parameters and morphological features of the aortic structure, includes: All local vascular masks are merged to obtain a globally fused mask; The redundant mask voxel regions belonging to the aorta in the global fusion mask are removed to obtain the vascular tree base mask. Using the original structural parameters and morphological features of the aortic structure, the aortic mask segment corresponding to the aortic region is extracted; The aortic mask fragment is spliced ​​with the vascular tree base mask at corresponding positions to obtain the complete vascular tree mask of the target vascular tree structure.

10. The method according to claim 1, characterized in that, The method further includes: Priority labels are marked on the calculation task packages corresponding to the blood vessel segments that meet the preset screening requirements; The parallel computation of unique geometric hash signatures corresponding to each blood vessel segment based on the centerline coordinate point set in each computation task package includes: First, the voxel coordinate point set in the computation task package labeled with priority is processed in parallel to deorder and standardize. Then, the voxel coordinate point set in the computation task package without priority label is processed in parallel to deorder and standardize, so as to obtain the unique geometric hash signature corresponding to each blood vessel segment.

11. A three-dimensional mask construction device for vascular trees, characterized in that, The device includes: An information acquisition unit is used to acquire multiple centerlines in a target vascular tree structure, as well as segment information corresponding to each centerline. The target vascular tree structure includes the aorta, multiple main trunk vessels, and multiple branch vessels. The main trunk vessels originate from the aorta and extend outward, while the branch vessels branch off from the main trunk vessels. The starting point of the centerline is the root of the main trunk to which the branch vessel belongs, and the ending point of the centerline is the terminal end of the branch vessel. Each centerline corresponds to multiple vascular segments. The segment information includes the segment index, the coordinates of the segment's starting point, the coordinates of the segment's ending point, and the coordinates of all centerline points from the segment's starting point to its ending point. The sampling unit is used to sample each centerline to obtain the sampling line segment corresponding to each centerline; each sampling line segment is composed of multiple centerline sampling points, and each segment of the blood vessel corresponds to multiple centerline sampling points; The radius detection unit is used to obtain the set of physical radii of the inscribed circles of each blood vessel segment on each sampling line segment based on morphological corrosion detection. The task package generation unit is used to generate a calculation task package for each blood vessel segment. The calculation task package includes the segment information of the blood vessel segment, the total number of centerline sampling points, the set of physical radii of the inscribed circle, the set of centerline coordinate points of the centerline where the blood vessel segment is located, and the set of voxel coordinate points of the original image corresponding to the blood vessel segment. The hash calculation unit is used to calculate the unique geometric hash signature corresponding to each blood vessel segment in parallel based on the centerline coordinate point set in each calculation task package. The segmentation filtering unit is used to filter all vascular segments based on the duplicate matching relationship of multiple geometric hash signatures to obtain multiple first target segments; The local mask generation unit is used to obtain the local vascular mask corresponding to each first target segment based on the computational task package corresponding to each first target segment. A three-dimensional mask construction unit is used to construct a three-dimensional mask of the target vascular tree structure based on all local vascular masks, as well as the original structural parameters and morphological features of the aortic structure.

12. A three-dimensional mask construction device for vascular trees, characterized in that, include: A memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the vascular tree three-dimensional mask construction method as described in any one of claims 1-10.

Images