A blood vessel naming method, system, terminal and storage medium based on center line extraction
By generating vascular models through 3D segmentation and centerline extraction, and combining them with point cloud segmentation networks, the problem of low accuracy in existing vascular naming is solved, achieving more efficient and accurate vascular naming.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN INST OF ARTIFICIAL INTELLIGENCE & ROBOTICS FOR SOC
- Filing Date
- 2026-02-11
- Publication Date
- 2026-07-21
AI Technical Summary
Existing vessel naming technologies suffer from low accuracy, and the point cloud information generated by existing methods is either redundant or insufficient, failing to support subsequent vessel naming tasks.
A 3D segmentation method is used to obtain a tubular model, extract the centerline, and generate a vascular tube model. The coordinate information of the vascular label and the tangent direction information of the centerline point are used as additional features of the point cloud and input into the point cloud segmentation network to determine the category of each point and map it to the center point of the vascular tube for naming.
It improves the accuracy of vessel naming by preserving the topological structure of the centerline, thereby increasing the efficiency of centerline annotation and the accuracy of vessel naming.
Smart Images
Figure CN121709167B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical imaging technology, and in particular to a method, system, terminal, and storage medium for naming blood vessels based on centerline extraction. Background Technology
[0002] Vascular naming refers to the process of using computer technology to separate and label specific blood vessels from the background in three-dimensional medical images (such as X-ray tomography, magnetic resonance imaging, etc.), then extracting the central path of the blood vessel from it using automatic or semi-automatic techniques, and semantically annotating different branches of the extracted central path of the blood vessel according to anatomical standards; accurate naming of blood vessels can provide a foundation for subsequent blood vessel matching and three-dimensional reconstruction of blood vessels.
[0003] Existing vessel naming techniques mainly generate point clouds directly from vessel segmentation data, and then use the generated point clouds for classification and naming. However, since the morphology of vessel segmentation data varies in different image data, the generated point cloud information contains a lot of redundant or invalid data. Furthermore, the point cloud generated by vessel segmentation does not retain comprehensive information about the centerline topology, resulting in low accuracy of the final vessel naming effect.
[0004] Existing vessel naming technology 2 first obtains vessel segmentation, then derives the centerline based on the segmentation, and then directly extracts the point cloud information from the centerline. Because the centerline is very thin, the amount of effective information that can be extracted is limited, resulting in insufficient point cloud information to support subsequent vessel naming tasks.
[0005] Therefore, existing technologies still need improvement. Summary of the Invention
[0006] The technical problem to be solved by the present invention is to provide a method, system, terminal and storage medium for blood vessel naming based on centerline extraction, in order to solve the problem of low accuracy in blood vessel naming of existing methods based on centerline extraction.
[0007] The technical solution adopted by this invention to solve the technical problem is as follows:
[0008] In a first aspect, the present invention provides a method for naming blood vessels based on centerline extraction, comprising:
[0009] The original medical images were acquired, and a tubular model was obtained using a three-dimensional segmentation method. The centerline of the tubular model was then extracted.
[0010] A vascular network model is generated based on the centerline. The vascular network model is then converted into vascular label data. The coordinate information of the vascular labels and the tangent direction information of the centerline points are used as additional features of the point cloud to obtain the point cloud data of the vascular network.
[0011] The point cloud data is input into a point cloud segmentation network to obtain the category of each point;
[0012] Each point in the point cloud data is mapped to the nearest blood vessel center point, and the category of each center line segment is determined based on the mapping result, thus obtaining the blood vessel name corresponding to each center line segment.
[0013] In one implementation, acquiring the original medical image, obtaining a tubular model using a three-dimensional segmentation method, and extracting the centerline of the tubular model includes:
[0014] The original medical image is acquired; wherein the region of interest in the original medical image is the tubular anatomical structure region;
[0015] The original medical images are preprocessed, and the tubular objects in the medical images are segmented using traditional segmentation methods or deep learning segmentation methods to obtain the tubular object model.
[0016] The centerline of the tubular model is extracted using a skeleton refinement method or a distance transformation method. The extracted centerline is then preprocessed and adjusted to obtain the adjusted centerline.
[0017] In one implementation, the preprocessing and adjustment of the extracted centerline to obtain the adjusted centerline includes:
[0018] The extracted centerline is smoothed and resampled at equal intervals, and the tangent vector and normal vector of each sampling point are calculated to obtain the preprocessed centerline.
[0019] For the sampling points in the preprocessed centerline, the tubular model is segmented in the direction of the tangent vector perpendicular to the center point to obtain the lumen outline;
[0020] Calculate the centroid position of the lumen profile, and move the corresponding center point to the centroid position to obtain a new centerline;
[0021] The new centerline is smoothed and resampled at equal intervals to obtain the adjusted centerline.
[0022] In one implementation, the process of generating a vascular network model based on a centerline, converting the vascular network model into vascular label data, and using the coordinate information of the vascular labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain point cloud data of the vascular network includes:
[0023] An initial circle point is obtained based on the preset diameter and the normal vector direction of each sampling point in the center line. The initial circle point is then rotated counterclockwise around the corresponding tangent vector to obtain several circle points.
[0024] Connect two adjacent circular points using triangular facets to obtain a blood vessel with a preset diameter;
[0025] Generate the vascular conduit model based on all the vascular conduits;
[0026] The vascular channel model is labeled to obtain the vascular label data; wherein, the value 1 in the vascular label data is the intravascular label and the value 0 is the extravascular label;
[0027] The coordinate information of the value 1 in the blood vessel label data is extracted, and the tangent direction information of the corresponding center line point of the point cloud is extracted. The coordinate information and the tangent direction information are used as additional features of the corresponding point cloud to obtain the point cloud data of the blood vessel.
[0028] In one implementation, the step of labeling the vascular network model to obtain the vascular label data includes:
[0029] A new 3D image is created based on the extent of all blood vessels and a preset voxel interval; wherein the area of the 3D image covers all blood vessels.
[0030] Traverse the points in the three-dimensional image and determine whether the current point is located in the corresponding blood vessel.
[0031] If it is located in the corresponding blood vessel, then assign a value of 1 to the current point;
[0032] If the current point is not located in the corresponding blood vessel, then assign a value of 0.
[0033] In one implementation, inputting the point cloud data into a point cloud segmentation network to obtain the category of each point includes:
[0034] The coordinate information and tangent direction information of the point cloud data are input into a pre-trained point cloud segmentation network, and the category information of each point in the point cloud is output.
[0035] In one implementation, mapping each point of the point cloud data to the nearest vessel center point and determining the category of each centerline segment based on the mapping result to obtain the vessel name corresponding to each centerline segment includes:
[0036] Find the nearest blood vessel centerline point for each point in the point cloud data, and map each point to the nearest blood vessel centerline point;
[0037] Summarize the point cloud data mapped to each center point on each centerline segment to obtain the number of point clouds for each type of blood vessel in each centerline segment.
[0038] The category with the most point cloud counts on each centerline segment is identified, and the corresponding blood vessel name is obtained based on the identified categories.
[0039] Secondly, the present invention provides a vessel naming system based on centerline extraction, comprising:
[0040] The segmentation and extraction module is used to acquire the original medical images, obtain the tubular model using a three-dimensional segmentation method, and extract the centerline of the tubular model.
[0041] The point cloud data generation module is used to generate a blood vessel and pipeline model based on the centerline, convert the blood vessel and pipeline model into blood vessel label data, and use the coordinate information of the blood vessel labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain the point cloud data of the blood vessel and pipeline.
[0042] The point cloud data classification module is used to input the point cloud data into the point cloud segmentation network to obtain the category of each point;
[0043] The vessel naming module is used to map each point of the point cloud data to the nearest vessel center point, and determine the category of each center line segment based on the mapping result, thereby obtaining the vessel name corresponding to each center line segment.
[0044] Thirdly, the present invention provides a terminal, comprising: a processor and a memory, the memory storing a vessel naming program based on centerline extraction, wherein the vessel naming program based on centerline extraction, when executed by the processor, is used to implement the operation of the vessel naming method based on centerline extraction as described in the first aspect.
[0045] Fourthly, the present invention also provides a computer-readable storage medium storing a centerline-based vessel naming program, which, when executed by a processor, is used to implement the operation of the centerline-based vessel naming method as described in the first aspect.
[0046] The present invention, by employing the above technical solution, has the following effects:
[0047] This invention employs a 3D segmentation method to segment original medical images, extracting the centerline from the obtained tubular model and generating a vascular network model based on the centerline. Furthermore, by converting the vascular network model into vascular label data, the coordinate information of the vascular labels and the tangent direction information of the centerline points are used as additional features of the point cloud. The category of each point is determined based on the point cloud data of the vascular network. Finally, by mapping each point in the point cloud data to the nearest vascular center point and determining the category of each centerline segment based on the mapping result, the vascular name corresponding to each centerline segment can be obtained. This invention generates a point cloud based on the centerline, preserving the topological structure of the centerline. Centerline annotation can be performed using only the tubular topology, improving the efficiency of centerline annotation and the accuracy of vascular naming. Attached Figure Description
[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.
[0049] Figure 1 This is a flowchart of the blood vessel naming method based on centerline extraction in this invention.
[0050] Figure 2 This is a schematic diagram of the tubular model in this invention.
[0051] Figure 3 This is a schematic diagram of the center line extracted in this invention.
[0052] Figure 4 This is a schematic diagram of the blood vessel model generated in this invention.
[0053] Figure 5 This is a schematic diagram illustrating the conversion of vascular pipeline models into vascular label data in this invention.
[0054] Figure 6 This is a point cloud schematic diagram of the blood vessel channels generated in this invention.
[0055] Figure 7 This is a schematic diagram of the point cloud categories after point cloud classification of blood vessels in this invention.
[0056] Figure 8 This is a schematic diagram illustrating the naming of blood vessel channels in this invention.
[0057] Figure 9 This is a functional schematic diagram of the terminal in one implementation of the present invention.
[0058] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0059] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0060] Exemplary methods
[0061] Existing vessel naming techniques mainly generate point clouds directly from vessel segmentation data, and then use the generated point clouds for classification and naming. However, since the morphology of vessel segmentation data varies in different image data, the generated point cloud information contains a lot of redundant or invalid data. Furthermore, the point cloud generated by vessel segmentation does not retain comprehensive information about the centerline topology, resulting in low accuracy of the final vessel naming effect.
[0062] Existing vessel naming technology 2 first obtains vessel segmentation, then derives the centerline based on the segmentation, and then directly extracts the point cloud information from the centerline. Because the centerline is very thin, the amount of effective information that can be extracted is limited, resulting in insufficient point cloud information to support subsequent vessel naming tasks.
[0063] To address the above technical problems, this invention provides a method for naming blood vessels based on centerline extraction, comprising: acquiring original medical images; obtaining a tubular model using a 3D segmentation method; and extracting the centerline from the tubular model; generating a blood vessel model based on the centerline; converting the blood vessel model into blood vessel label data; and using the coordinate information of the blood vessel labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain point cloud data of the blood vessel; inputting the point cloud data into a point cloud segmentation network to obtain the category of each point; mapping each point in the point cloud data to the nearest blood vessel centerline point, and determining the category of each centerline segment based on the mapping result to obtain the blood vessel name corresponding to each centerline segment. This invention can improve the accuracy of blood vessel naming.
[0064] like Figure 1 As shown, this embodiment of the invention provides a method for naming blood vessels based on centerline extraction, including the following steps:
[0065] Step S100: Acquire the original medical image, obtain the tubular model using a three-dimensional segmentation method, and extract the centerline of the tubular model.
[0066] In this embodiment, the original medical image is first segmented using a 3D segmentation method to obtain a tubular model. Then, based on the segmentation results, a skeleton refinement method or other methods are used to extract the centerline of the blood vessels, so that the blood vessel model can be obtained by subsequent modeling based on this centerline.
[0067] Specifically, in one implementation of this embodiment, step S100 includes the following steps:
[0068] Step S101: Acquire the original medical image; wherein the region of interest in the original medical image is the tubular anatomical structure region;
[0069] Step S102: Perform data preprocessing on the original medical image, and segment the tubular object in the medical image using traditional segmentation methods or deep learning segmentation methods to obtain the tubular object model.
[0070] In this embodiment, during the 3D segmentation process, raw medical images from CT (Computed Tomography) and MRI (Magnetic Resonance Imaging) scans are first acquired; the image capture area represents the location of the tubular anatomical structure of interest. Then, the raw medical images undergo preprocessing such as filtering, denoising, and interpolation to obtain preprocessed medical images. Finally, the tubular structure model (e.g., [missing information]) is segmented from the preprocessed medical images using traditional segmentation methods or deep learning-based segmentation methods. Figure 2 (As shown).
[0071] As an example, the traditional segmentation methods adopted in this embodiment include: threshold-based segmentation methods, region-growing-based segmentation methods, level set-based segmentation methods, and tracking-based segmentation methods. Deep learning-based segmentation methods mainly include the UNet (a convolutional neural network designed specifically for image segmentation tasks) series of methods.
[0072] Step S103: The centerline of the tubular model is extracted using a skeleton refinement method or a distance transformation method, and the extracted centerline is preprocessed and adjusted to obtain the adjusted centerline.
[0073] In this embodiment, after segmenting the tubular model, the centerline of the blood vessel is extracted from it using a skeleton refinement method or a distance transformation method; then, the centerline is preprocessed and adjusted to meet the requirements of subsequent blood vessel modeling.
[0074] In this embodiment, the skeleton refinement method is used as an example to extract the centerline of the tubular model, and the extracted centerline is preprocessed and adjusted. The specific process is as follows:
[0075] 1) Skeleton refinement methods:
[0076] Skeleton thinning is an iterative algorithm that peels away pixels layer by layer from the boundary inwards until only a single-pixel-wide centerline skeleton remains. The steps of skeleton thinning are as follows:
[0077] a. Traverse the voxels in the tubular model, use the 6-neighborhood or 26-neighborhood method to determine whether the voxel is a deletable boundary point, and then retain or delete the point according to the determination result;
[0078] b. Continue iterating and refining the process until no more voxels can be deleted in a complete iteration, thus forming the centerline.
[0079] 2) Establishment of centerline topology:
[0080] After refining the skeleton to obtain the centerline, a centerline topology is established to label the centerline. The specific process is as follows:
[0081] 2.1. Establishing the centerline diagram:
[0082] By traversing each center point, we can determine whether there are other points in the 26-neighborhood of the center point. If there are, we can connect the point to the current center point as an edge, thus establishing the entire skeletonized centerline into a centerline graph structure.
[0083] Once the centerline diagram is created, blood vessels can be marked on the centerline.
[0084] 2.2. Interactive centerline annotation:
[0085] The interactive centerline annotation method primarily involves loading the initial centerline and the initial tubular model. Interactively, users select the start and end points of the initial centerline for a specific tubular structure and assign it a specific name, thus completing the annotation of that tubular structure. The specific annotation steps are as follows:
[0086] a. Based on the tubular segmentation results, the MarchingCubes algorithm (a voxel-level 3D reconstruction method) is used to obtain the tubular model; then, the centerline map and the tubular model are loaded for display.
[0087] b. Using an interactive method, sequentially click on the starting point, n path points, and ending point of the initial centerline of a tubular structure to mark it. When clicking on each point on the initial centerline, calculate the shortest path between that point and the previous point using Dijkstra's algorithm, and retain the shortest path as the centerline path between these two points. Repeat this process until the entire centerline is marked and the centerline path is generated (where n can be 0 points).
[0088] c. Provide a list of names of tubular structures to be labeled via a drop-down list, and then select a specific name of the tubular structure from the drop-down list to complete the naming of the current centerline path, thereby completing the labeling of the centerline of a tubular structure.
[0089] d. The remaining tubular centerlines can be marked in the same way.
[0090] In this embodiment, after establishing the centerline topology, the centerline segments can be extracted, and then each segment of the centerline can be preprocessed. The centerline preprocessing involves smoothing and resampling the centerline and generating some centerline attribute information to prepare for subsequent steps.
[0091] In one implementation of this embodiment, step S103 includes the following steps:
[0092] Step S103a: The extracted centerline is smoothed and resampled at equal intervals, and the tangent vector and normal vector of each sampling point are calculated to obtain the preprocessed centerline.
[0093] In this embodiment, the step of establishing the centerline topology as described above further includes:
[0094] 3) Centerline preprocessing, the specific preprocessing steps are as follows:
[0095] a. Smooth the centerline; this process can use any of the following methods: Gaussian smoothing filter, sliding window filter, or spline interpolation.
[0096] b. Resample the centerline at equal intervals to generate a series of equally spaced center points. These equally spaced center points can avoid model distortion caused by uneven point distribution.
[0097] c. Calculate the difference vector between adjacent points on the center line to obtain the tangent vector of that center line point.
[0098] d. Given the normal vector of the first point on the centerline, the binormal vector of the first point can be calculated; based on the tangent vector, normal vector, binormal vector of the previous point and the tangent vector of the next point, the normal vector and binormal vector of the next point can be calculated.
[0099] In one implementation of this embodiment, step S103 further includes the following steps:
[0100] Step S103b: For the sampling points in the preprocessed centerline, the tubular model is segmented in the direction of the tangent vector perpendicular to the center point to obtain the lumen outline;
[0101] Step S103c: Calculate the centroid position of the lumen profile, and move the corresponding center point to the centroid position to obtain a new centerline;
[0102] Step S103d: The new centerline is smoothed and resampled at equal intervals to obtain the adjusted centerline.
[0103] In this embodiment, the pre-processed centerline is automatically centered to ensure more accurate positioning. The specific centerline adjustment steps are as follows:
[0104] First, for the equally spaced resampling center points, the tubular model is segmented along the tangent direction perpendicular to the center point to obtain the contour of the cavity. This process specifically includes: defining a cutting plane based on the center point and the tangent direction; and then, using the cutting plane as a cutting tool, calculating the intersection with the triangular facet model of the cavity. This process iterates through each triangular facet in the mesh, determining whether the facet intersects with the cutting plane. For intersecting triangular facets, the precise coordinates of the intersection points of the plane and the two sides of the triangle are calculated, and these intersection points are connected to form a closed intersection line contour located on the surface of the cavity.
[0105] Then, the centroid position of the lumen profile is calculated, and the center point is moved to that position as the new centerline; because the centroid position of this profile better represents the center of the lumen, avoiding the problem of the original center point being offset.
[0106] Afterwards, the new centerline is smoothed and sampled at equal intervals to obtain the adjusted centerline.
[0107] Finally, the centerline of the remaining tubular structure can be generated in the same way.
[0108] This embodiment extracts the centerline of the tubular model, and after preprocessing and adjusting the extracted centerline, the final centerline is as follows: Figure 3 As shown.
[0109] like Figure 1 As shown, this embodiment of the invention provides a method for naming blood vessels based on centerline extraction, including the following steps:
[0110] Step S200: Generate a vascular pipeline model based on the centerline, convert the vascular pipeline model into vascular label data, and use the coordinate information of the vascular labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain the point cloud data of the vascular pipeline.
[0111] In this embodiment, point cloud data is generated from the extracted vascular centerline. This process mainly involves generating a vascular channel model based on the centerline and converting the vascular channel model into label data (i.e., tag data). The label inside the vascular cavity is 1, and the label outside the vascular cavity is 0. Then, the coordinate information of the vascular cavities labeled 1 is extracted, and the tangent direction information of the corresponding centerline point in the point cloud is extracted as additional features of that point cloud point, thereby obtaining supplementary point cloud data of the vascular channel.
[0112] Specifically, in one implementation of this embodiment, step S200 includes the following steps:
[0113] Step S201: Obtain initial circle points based on the preset diameter and the normal vector direction of each sampling point in the center line, and rotate the initial circle points counterclockwise around the corresponding tangent vector to obtain several circle points;
[0114] Step S202: Connect two adjacent circular points using triangular facets to obtain a blood vessel with a preset diameter;
[0115] Step S203: Generate the vascular channel model based on all the vascular channels.
[0116] In this embodiment, a blood vessel with a specified diameter is established based on the centerline to generate a blood vessel model. The specific establishment steps are as follows:
[0117] a. Obtain initial circle points based on the normal vector direction of the specified diameter and center point. Then, use these circle points to rotate counterclockwise by 10 to 350 degrees around the tangent vector to obtain a complete circle of points. Each center point can yield these 36 circle points.
[0118] b. Connecting two adjacent circular points with triangular facets will create a pipe.
[0119] c. Perform the same processing on all centerlines to obtain multiple vascular channels, thus completing the construction of the vascular channel model.
[0120] In this embodiment, a blood vessel of a specified diameter is established based on the above modeling method, and the final blood vessel model is as follows. Figure 4 As shown.
[0121] Specifically, in one implementation of this embodiment, step S200 further includes the following steps:
[0122] Step S204: Label the blood vessel model to obtain the blood vessel label data; wherein, the value 1 in the blood vessel label data is the inner label of the blood vessel, and the value 0 is the outer label of the blood vessel.
[0123] In one implementation of this embodiment, step S204 includes the following steps:
[0124] Step S204a: Construct a new three-dimensional image based on the extent of all blood vessels and a preset voxel interval; wherein the area of the three-dimensional image covers all blood vessels.
[0125] Step S204b: Traverse the points of the three-dimensional image and determine whether the current point is located in the corresponding blood vessel.
[0126] Step S204c: If the point is located in the corresponding blood vessel, assign a value of 1 to the current point.
[0127] In step S204d, if the point is not located in the corresponding blood vessel, then assign a value of 0 to the current point.
[0128] In this embodiment, the specific steps for converting the vascular network model into label data are as follows:
[0129] a. Create a new 3D image based on the extent of all blood vessels and the specified voxel spacing. Set the origin parameter (used to define the origin of the image coordinate system, representing the starting coordinates of the image data in the RAS coordinate system) and the spacing parameter (used to describe the resolution of the image, representing the spacing between voxels in each dimension of the image) of the 3D image so that the 3D image area just covers the extent of all blood vessels.
[0130] b. Iterate through the points in the 3D image and determine whether a point is located inside the pipe. If it is inside the pipe, assign it a value of 1; otherwise, assign it a value of 0.
[0131] This judgment method can be based on the parity test of ray projection, that is, emitting a ray from the point along any coordinate axis, and calculating the number of intersections between the ray and the triangular facet of the pipe. If the number is odd, it means that the pipe is inside; if the number is even, it means that the pipe is outside.
[0132] In this embodiment, the vascular channel model is converted into label data based on the above method, and the resulting structural diagram is as follows. Figure 5 As shown.
[0133] Specifically, in one implementation of this embodiment, step S200 further includes the following steps:
[0134] Step S205: Extract the coordinate information of the value 1 in the blood vessel label data, and extract the tangent direction information of the corresponding center line point of the point cloud. Use the coordinate information and the tangent direction information as additional features of the corresponding point cloud to obtain the point cloud data of the blood vessel.
[0135] In this embodiment, after converting the vascular network model into label data, the coordinate information of the vascular label with a value of 1 is extracted, and the tangent direction information of the centerline point of the corresponding point cloud is also extracted. This coordinate information and tangent direction information are used as additional features of the point cloud; that is, the points with a value of 1 in the 3D image are traversed, and their coordinate information and the corresponding tangent direction are extracted as additional features of the point cloud. As an example, after the above point cloud generation steps, the obtained vascular network point cloud data in this embodiment is as follows: Figure 6 As shown.
[0136] In this embodiment, a point cloud is generated based on the centerline, which not only preserves the topological structure of the centerline but also introduces rich point cloud information, ensuring the accuracy of subsequent point cloud classification and blood vessel naming.
[0137] like Figure 1 As shown, this embodiment of the invention provides a method for naming blood vessels based on centerline extraction, including the following steps:
[0138] Step S300: Input the point cloud data into the point cloud segmentation network to obtain the category of each point.
[0139] In this embodiment, the generated point cloud data is input into a trained point cloud segmentation network to obtain the category of each point, thereby achieving point cloud data segmentation.
[0140] Specifically, in one implementation of this embodiment, step S300 includes the following steps:
[0141] Step S301: Input the coordinate information and tangent direction information of the point cloud data into the pre-trained point cloud segmentation network, and output the category information of each point in the point cloud.
[0142] In this embodiment, point cloud segmentation can be divided into three parts: point cloud segmentation data annotation, point cloud segmentation training, and point cloud segmentation inference.
[0143] 1) Point cloud segmentation data annotation:
[0144] a. Centerlines can be generated by manually selecting points and labeling the named centerline data. This will ultimately yield the specified N centerline segments.
[0145] b. Generate point clouds for each centerline separately to obtain point cloud feature data containing named labels.
[0146] 2) Point cloud segmentation training:
[0147] The coordinates, coordinate direction vectors, and other information of the point cloud annotation data containing named labels are input into the point cloud segmentation network for calculation. The calculation results are compared with the point cloud annotation classification results and backpropagation is used to adjust the parameters, thus completing the training of the point cloud segmentation network.
[0148] 3) Point cloud segmentation reasoning:
[0149] The point cloud data of the blood vessels obtained in step S200 (including features such as coordinate information and tangent direction information) is directly input into the trained point cloud segmentation network, and the output is the category of each point in the point cloud.
[0150] As an example, in this embodiment, the point cloud data of blood vessels obtained in step S200 is classified using a trained point cloud segmentation network, and the resulting point cloud categories are as follows: Figure 7 As shown.
[0151] like Figure 1 As shown, this embodiment of the invention provides a method for naming blood vessels based on centerline extraction, including the following steps:
[0152] Step S400: Map each point of the point cloud data to the nearest blood vessel center point, and determine the category of each segment of the center line based on the mapping result, thereby obtaining the blood vessel name corresponding to each segment of the center line.
[0153] In this embodiment, each point in the classified point cloud data is mapped to the nearest vessel center point. By extracting the point cloud data corresponding to each segment of the vessel centerline network, the category with the most points in that segment is counted, and this category is used as the name of that segment of the centerline. This completes the naming of each segment of the centerline.
[0154] Specifically, in one implementation of this embodiment, step S400 includes the following steps:
[0155] Step S401: Find the nearest blood vessel centerline point for each point in the point cloud data, and map each point to the nearest blood vessel centerline point;
[0156] Step S402: Summarize the point cloud data mapped to each center point on each segment of the centerline to obtain the number of point clouds for each type of blood vessel in each segment of the centerline.
[0157] Step S403: Count the category with the most point cloud on each centerline segment, and name the blood vessel corresponding to each centerline segment based on the counted category.
[0158] In this embodiment, the nearest vessel centerline point is obtained for each point in the point cloud data, thus mapping the point cloud point to the nearest vessel centerline point. Since the extracted centerline underwent equidistant resampling in step S103, segmented centerline results are obtained. For each centerline segment, the point cloud data mapped to each centerline point is summarized to obtain the number of points in each vessel point cloud category within that segment. Finally, the category with the most points in that segment is counted, and this category is used as the name of the centerline segment. This completes the naming of each centerline segment.
[0159] As an example, in this embodiment, based on the above-described blood vessel naming method, the named blood vessel model is as follows: Figure 8 As shown.
[0160] The method provided in this embodiment only requires the topology of the tubular structure to annotate the centerline, avoiding fully manual annotation and improving the efficiency of centerline annotation. Furthermore, the point cloud generated based on the centerline not only preserves the topology of the centerline but also introduces rich point cloud information. The method provided in this embodiment can improve the accuracy of blood vessel naming.
[0161] This embodiment achieves the following technical effects through the above technical solution:
[0162] This embodiment employs a 3D segmentation method to segment the original medical image. The centerline of the obtained tubular model can be extracted, and a vascular network model can be generated based on the centerline. Furthermore, by converting the vascular network model into vascular label data, the coordinate information of the vascular labels and the tangent direction information of the centerline points can be used as additional features of the point cloud. The category of each point is determined based on the point cloud data of the vascular network. Finally, by mapping each point in the point cloud data to the nearest vascular center point, and determining the category of each segment of the centerline based on the mapping results, the vascular name corresponding to each segment of the centerline can be obtained. This embodiment generates a point cloud based on the centerline, preserving the topological structure of the centerline. Centerline annotation can be performed using only the topological structure of the tubular structure, improving the efficiency of centerline annotation and the accuracy of vascular naming.
[0163] Exemplary device
[0164] Based on the above embodiments, the present invention also provides a vessel naming system based on centerline extraction, comprising:
[0165] The segmentation and extraction module is used to acquire the original medical images, obtain the tubular model using a three-dimensional segmentation method, and extract the centerline of the tubular model.
[0166] The point cloud data generation module is used to generate a blood vessel and pipeline model based on the centerline, convert the blood vessel and pipeline model into blood vessel label data, and use the coordinate information of the blood vessel labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain the point cloud data of the blood vessel and pipeline.
[0167] The point cloud data classification module is used to input the point cloud data into the point cloud segmentation network to obtain the category of each point;
[0168] The vessel naming module is used to map each point of the point cloud data to the nearest vessel center point, and determine the category of each center line segment based on the mapping result, thereby obtaining the vessel name corresponding to each center line segment.
[0169] This embodiment achieves the following technical effects through the above technical solution:
[0170] This embodiment employs a 3D segmentation method to segment the original medical image. The centerline of the obtained tubular model can be extracted, and a vascular network model can be generated based on the centerline. Furthermore, by converting the vascular network model into vascular label data, the coordinate information of the vascular labels and the tangent direction information of the centerline points can be used as additional features of the point cloud. The category of each point is determined based on the point cloud data of the vascular network. Finally, by mapping each point in the point cloud data to the nearest vascular center point, and determining the category of each segment of the centerline based on the mapping results, the vascular name corresponding to each segment of the centerline can be obtained. This embodiment generates a point cloud based on the centerline, preserving the topological structure of the centerline. Centerline annotation can be performed using only the topological structure of the tubular structure, improving the efficiency of centerline annotation and the accuracy of vascular naming.
[0171] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 9 As shown.
[0172] The terminal includes: a processor, a memory, an interface, a display screen, and a communication module connected via a system bus; wherein, the processor of the terminal provides computing and control capabilities; the memory of the terminal includes a computer-readable storage medium and internal memory; the computer-readable storage medium stores an operating system and computer programs; the internal memory provides an environment for the operation of the operating system and computer programs in the computer-readable storage medium; the interface is used to connect to external devices; the display screen is used to display relevant information; and the communication module is used to communicate with a cloud server or other devices.
[0173] When executed by the processor, this computer program is used to implement a blood vessel naming method based on centerline extraction.
[0174] It will be understood by those skilled in the art that Figure 9The schematic diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0175] In one embodiment, a terminal is provided, comprising: a processor and a memory, the memory storing a centerline-based vessel naming program, which, when executed by the processor, is used to implement the operations of the centerline-based vessel naming method described above.
[0176] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a centerline-based vessel naming program, which, when executed by a processor, is used to implement the operations of the centerline-based vessel naming method described above.
[0177] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, database, or other media used in the embodiments provided by this invention can include both non-volatile and volatile memory.
[0178] In summary, this invention provides a method, system, terminal, and storage medium for vessel naming based on centerline extraction, comprising: acquiring original medical images; obtaining a tubular model using a 3D segmentation method; extracting the centerline from the tubular model; generating a vascular pipeline model based on the centerline; converting the vascular pipeline model into vascular label data; and using the coordinate information of the vascular labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain point cloud data of the vascular pipeline; inputting the point cloud data into a point cloud segmentation network to obtain the category of each point; mapping each point of the point cloud data to the nearest vascular center point, and determining the category of each centerline segment based on the mapping result to obtain the vessel name corresponding to each centerline segment. This invention can improve the accuracy of vessel naming.
[0179] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.
Claims
1. A method for naming blood vessels based on centerline extraction, characterized in that, include: The original medical images were acquired, and a tubular model was obtained using a three-dimensional segmentation method. The centerline of the tubular model was then extracted. A vascular network model is generated based on a centerline. This model is then converted into vascular label data. The coordinate information of the vascular labels and the tangent direction information of the centerline points are used as additional features of the point cloud to obtain the point cloud data of the vascular network. The process includes: obtaining initial circular points based on a preset diameter and the normal vector direction of each sampling point in the centerline; rotating these initial circular points counterclockwise around their corresponding tangent vectors to obtain several circular points; connecting adjacent circular points using triangular facets to obtain a vascular network of the preset diameter; generating the vascular network model based on all the vascular networks; labeling the vascular network model to obtain the vascular label data; wherein, a value of 1 in the vascular label data represents an inner vascular label, and a value of 0 represents an outer vascular label; extracting the coordinate information of the value 1 in the vascular label data, and extracting the tangent direction information of the corresponding centerline points in the point cloud; using the coordinate information and the tangent direction information as additional features of the corresponding point cloud to obtain the point cloud data of the vascular network. The point cloud data is input into a point cloud segmentation network to obtain the category of each point; Each point in the point cloud data is mapped to the nearest blood vessel center point, and the category of each center line segment is determined based on the mapping result, thus obtaining the blood vessel name corresponding to each center line segment.
2. The vessel naming method based on centerline extraction according to claim 1, characterized in that, The process of acquiring the original medical image, obtaining a tubular model using a 3D segmentation method, and extracting the centerline of the tubular model includes: The original medical image is acquired; wherein the region of interest in the original medical image is the tubular anatomical structure region; The original medical images are preprocessed, and the tubular objects in the medical images are segmented using traditional segmentation methods or deep learning segmentation methods to obtain the tubular object model. The centerline of the tubular model is extracted using a skeleton refinement method or a distance transformation method. The extracted centerline is then preprocessed and adjusted to obtain the adjusted centerline.
3. The vessel naming method based on centerline extraction according to claim 2, characterized in that, The process of preprocessing and adjusting the extracted centerline to obtain the adjusted centerline includes: The extracted centerline is smoothed and resampled at equal intervals, and the tangent vector and normal vector of each sampling point are calculated to obtain the preprocessed centerline. For the sampling points in the preprocessed centerline, the tubular model is segmented in the direction of the tangent vector perpendicular to the center point to obtain the lumen outline; Calculate the centroid position of the lumen profile, and move the corresponding center point to the centroid position to obtain a new centerline; The new centerline is smoothed and resampled at equal intervals to obtain the adjusted centerline.
4. The vessel naming method based on centerline extraction according to claim 1, characterized in that, The process of annotating the vascular network model to obtain the vascular label data includes: A new 3D image is created based on the extent of all blood vessels and a preset voxel interval; wherein the area of the 3D image covers all blood vessels. Traverse the points in the three-dimensional image and determine whether the current point is located in the corresponding blood vessel. If it is located in the corresponding blood vessel, then assign a value of 1 to the current point; If the current point is not located in the corresponding blood vessel, then assign a value of 0.
5. The vessel naming method based on centerline extraction according to claim 1, characterized in that, The step of inputting the point cloud data into a point cloud segmentation network to obtain the category of each point includes: The coordinate information and tangent direction information of the point cloud data are input into a pre-trained point cloud segmentation network, and the category information of each point in the point cloud is output.
6. The vessel naming method based on centerline extraction according to claim 1, characterized in that, The process of mapping each point in the point cloud data to the nearest vessel center point, determining the category of each centerline segment based on the mapping result, and obtaining the vessel name corresponding to each centerline segment includes: Find the nearest blood vessel centerline point for each point in the point cloud data, and map each point to the nearest blood vessel centerline point; Summarize the point cloud data mapped to each center point on each centerline segment to obtain the number of point clouds for each type of blood vessel in each centerline segment. The category with the most point cloud counts on each centerline segment is identified, and the corresponding blood vessel name is obtained based on the identified categories.
7. A vessel naming system based on centerline extraction, used to implement the vessel naming method based on centerline extraction as described in any one of claims 1-6, characterized in that, include: The segmentation and extraction module is used to acquire the original medical images, obtain the tubular model using a three-dimensional segmentation method, and extract the centerline of the tubular model. The point cloud data generation module is used to generate a blood vessel and pipeline model based on the centerline, convert the blood vessel and pipeline model into blood vessel label data, and use the coordinate information of the blood vessel labels and the tangent direction information of the centerline points as additional features of the point cloud to obtain the point cloud data of the blood vessel and pipeline. The point cloud data classification module is used to input the point cloud data into the point cloud segmentation network to obtain the category of each point; The vessel naming module is used to map each point of the point cloud data to the nearest vessel center point, and determine the category of each center line segment based on the mapping result, thereby obtaining the vessel name corresponding to each center line segment.
8. A terminal, characterized in that, include: The processor and memory, the memory storing a centerline-based vessel naming program, which, when executed by the processor, is used to implement the operation of the centerline-based vessel naming method as described in any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a centerline-based vessel naming program, which, when executed by a processor, is used to implement the operation of the centerline-based vessel naming method as described in any one of claims 1-6.