Tubular object dividing method, tubular object dividing device, and storage medium

Abstract

The tubular object segmentation method of the present application includes: obtaining a medical image containing a tubular object; outputting a tubular object probability map and a network segmentation result including a tubular object trunk and free branches, and obtaining a center line and an end point of each of the tubular object trunk and the free branches; thresholding the tubular object probability map, taking a region in the tubular object probability map within a prescribed first threshold range as a tubular object, extracting a second trunk with the largest volume from the obtained result, and obtaining a center line of the second trunk and an end point on the center line, and obtaining a candidate branch by subtracting the tubular object trunk from the second trunk; determining whether a distance between each end point of each branch of the candidate branch and a corresponding end point on the tubular object trunk is less than a prescribed second threshold, and if the distance is less than the second threshold, connecting the candidate branch to the corresponding end point of the tubular object trunk and displaying the candidate branch.

Description

Technical Field

[0001] This invention relates to a tubular segmentation method, a tubular segmentation device, and a computer-readable storage medium for segmenting tubular objects in medical images based on deep learning. Background Technology

[0002] In clinical medical research and applications, the three-dimensional reconstruction of tubular structures such as blood vessels based on medical images acquired by medical imaging devices can assist doctors in quickly locating organs, tissues, tumors, and other structures. Therefore, rapid and accurate segmentation of tubular structures is an important task in medical image processing.

[0003] With the continuous development of computer technology, the existing algorithms used for segmenting tubular structures such as blood vessels generally include: deep learning-related algorithms that modify the loss function or network structure; algorithms that utilize deep learning and tracking, using CNNs for initial segmentation to obtain the center line of the initial segmentation, and combining the initial position and tracking direction to detect distal blood vessels; and algorithms that perform deep learning to generate candidate tubular structures, using Hessian matrices or probabilistic graphs to generate candidate tubular structures, and using connected component matching to improve the breakage of candidate tubular structures or enrich the ends of tubular structures.

[0004] However, the aforementioned algorithms still have some drawbacks. For example, while algorithms that modify the loss function or network structure based on deep learning improve the connectivity of tubular structures to some extent, they cannot effectively address issues such as foreground-background imbalance in tubular segmentation (e.g., when the tubular structure is particularly thin) or the low probability of weak tubular structures at the far end in the network output probability map. Although the tracking-based algorithms can extend the distance of tubular structures and are suitable for scenarios with few branches, thicker tubular structures, and significant contrast, they cannot effectively handle situations with many small branches at the far end, varying branch directions, or missing tubular structures that did not originate from the centerline. The aforementioned algorithms for generating candidate tubular structures have drawbacks when using the Hessian matrix, including computational time consumption, unsuitability for applications with high time performance requirements, and the potential for noise introduction from incorrect tubular structures extracted by the Hessian matrix when image contrast is poor. However, when using a probabilistic map, since the probabilistic map is derived from the model output, it possesses richer connectivity information than the model's default result at a lower threshold. Therefore, compared to the previous two methods, it can more accurately find candidate tubular structures in the image, while omitting the tubular structure extraction process and running faster. On the other hand, the probabilistic map contains more uncertain information from the model at a lower threshold. If the main trunk part (i.e., the largest connected region) of the tubular structure under this low threshold is used directly, while improving the connectivity of blood vessels and the richness of the terminals, irrelevant thickened parts of blood vessels and noise are introduced.

[0005] Therefore, when using the existing deep learning-based algorithms to segment tubular structures, if the tubular structures are widely distributed and of uneven thickness, breakage can easily occur at particularly thin locations. Furthermore, certain diseases or poor image enhancement can lead to inconsistent image intensity across different regions of the tubular structures, also resulting in breakage. Summary of the Invention

[0006] In view of this situation, the object of the present invention is to provide a deep learning-based tubular segmentation method, tubular segmentation device, and storage medium, which improves the connectivity of the tubular segmentation results by utilizing post-processing, and by designing multiple thresholds in the post-processing, selectively connects the broken areas of the tubular structure based on the centerline and end points of the tubular structure, thereby reducing the breakage phenomenon or extending the end portion, while removing irrelevant noise.

[0007] To achieve the above objectives, one of the technical solutions of the present invention is a method for segmenting tubular objects, comprising:

[0008] The first step is to acquire medical images containing the tubular structure;

[0009] The second step is to output the tubular probability map for the tubular data and the network segmentation results including the main body of the tubular data and the free branches, and to obtain the center line and the end point on the center line of the main body of the tubular data and the free branches. The main body of the tubular data is the connected region with the largest volume or the most branches in the segmentation results. The free branches are the tubular parts that are connected regions other than the main body of the tubular data and are not connected to the main body of the tubular data.

[0010] The third step involves thresholding the tubular probability map, treating regions within a predetermined first threshold range as tubular structures. From this thresholded result, the connected component with the largest volume or the most branches (i.e., the second main branch) is extracted. The centerline and endpoints of the second main branch are then obtained. Candidate branches are obtained by subtracting the tubular main branch from the second main branch, and the endpoints of these candidate branches, excluding the endpoints of the second main branch, are obtained using the centerlines of these candidate branches.

[0011] The fourth step is to determine whether the distance between the end point of each candidate branch and the corresponding end point on the main body of the tubular structure is less than a predetermined second threshold. If it is less than the second threshold, the candidate branch is connected to the corresponding end point on the main body of the tubular structure and displayed.

[0012] Another technical solution of the present invention is a method for segmenting tubular objects, comprising:

[0013] The first step is to acquire medical images containing the tubular structure;

[0014] The second step is to output the tubular probability map for the tubular data and the network segmentation results including the main body of the tubular data and the free branches, and to obtain the center line and the end point on the center line of the main body of the tubular data and the free branches. The main body of the tubular data is the connected region with the largest volume or the most branches in the segmentation results. The free branches are the tubular parts that are connected regions other than the main body of the tubular data and are not connected to the main body of the tubular data.

[0015] The third step involves thresholding the tubular probability map, treating regions within a predetermined first threshold range as tubular structures. From this thresholded result, the connected component with the largest volume or the most branches (i.e., the second main branch) is extracted. The centerline and endpoints of the second main branch are then obtained. Candidate branches are obtained by subtracting the tubular main branch from the second main branch, and the endpoints of these candidate branches, excluding the endpoints of the second main branch, are obtained using the centerlines of these candidate branches.

[0016] The fourth step is to determine the corresponding end point in the candidate branches that corresponds to the end point in the free branch. Then, determine whether the distance between the corresponding end point and the corresponding end point on the main body of the tubular structure is less than the second threshold. If it is less than the second threshold, connect the candidate branch to which the corresponding end point belongs to the corresponding end point on the main body of the tubular structure and display it.

[0017] Another technical solution of the present invention is a tubular object splitting device, comprising:

[0018] The image acquisition unit acquires medical images containing the tubular object;

[0019] Processing Department; and

[0020] Display section,

[0021] The processing unit performs the above-described tubular object segmentation method.

[0022] Another technical solution of the present invention is a computer-readable storage medium that stores a program for enabling a computer to execute the above-described tubular segmentation method.

[0023] As described above, the tubular segmentation method, tubular segmentation device, and storage medium according to the present invention, by utilizing the model output probability map, can contain more connectivity information at a lower probability value. On the other hand, they can also effectively eliminate noise caused by the probability map. Attached Figure Description

[0024] Figure 1 (a) is a schematic diagram showing the network segmentation results of Example 1. Figure 1 (b) is a schematic diagram showing the main vascular trunk. Figure 1 (c) is a schematic diagram representing the free branch within it.

[0025] Figure 2 This is a schematic diagram illustrating the process of judging and connecting candidate branches.

[0026] Figure 3 This is a schematic diagram illustrating the process of judging and connecting candidate branches as a variation.

[0027] Figure 4 (a) is a schematic diagram showing the network segmentation results of Example 2. Figure 5 (b) is a schematic diagram showing the main trachea. Figure 4 (c) is a schematic diagram showing the centerline and terminal point of the main trachea. Figure 4 (d) is a schematic diagram showing the center line and end point of the free branch.

[0028] Figure 5 (a) is a schematic diagram representing the second main stem. Figure 5 (b) is a schematic diagram showing the centerline of the second main stem. Figure 5 (c) is a schematic diagram representing the candidate branches. Figure 5 (d) is a schematic diagram representing the centerline and endpoint of the candidate branch.

[0029] Figure 6 is a schematic diagram illustrating the process of judging and connecting candidate branches. Detailed Implementation

[0030] First, the tubular structures of the present invention refer to blood vessels (including arteries, veins, etc.) and / or trachea, etc.

[0031] Hereinafter, several embodiments for the specific implementation of the present invention will be described in detail based on the accompanying drawings.

[0032] (Example 1)

[0033] The following is for reference Figures 1-4 Example 1, an embodiment of the present invention, will be described below. In Example 1, a blood vessel will be used as an example of a tubular object.

[0034] The tubular (blood vessel) segmentation method involved in Example 1 includes the following steps.

[0035] The first step is to acquire medical images (usually CT images) that include blood vessels.

[0036] The second step outputs a vascular probability map and network segmentation results, including the main vessel trunk and free branches, based on the vascular data. It also obtains the centerlines and endpoints of the centerlines for the main vessel trunk and free branches. The probability map can be output by the model, including a foreground probability map and a background probability map, representing the probability of each pixel belonging to the foreground or background, respectively. The main vessel trunk (… Figure 1 (b) is the segmentation result ( Figure 1 The connected component with the largest volume (or the most branches, which can be determined by the length of the centerline) in (a) is the free branch. Figure 1 (c) represents the vascular portion in the segmentation result that is a connected region outside the main vascular trunk and not connected to it. Additionally, network segmentation refers to a known method for segmenting medical images using neural network models in deep learning.

[0037] The third step involves thresholding the blood vessel probability map (here, the foreground image). Regions within a predetermined first threshold range in the blood vessel probability map are identified as blood vessels, and the resulting value is the probability. Figure 2The connected component with the largest volume (or the most branches, which can be determined by the centerline length) is extracted from the value-enhanced image; this is the second main branch. The centerline of the second main branch and the terminal points on that centerline are obtained. Candidate branches are obtained by subtracting the main vascular branch from the second main branch. The centerline of the candidate branch is obtained by taking the intersection of the centerlines of the candidate branch and the second main branch. Then, the other terminal points of the candidate branch besides the terminal point of the second main branch are obtained using the centerline of the candidate branch. Finally, the candidate branch and its other terminal points besides the terminal point of the second main branch are obtained. For example, a first threshold of 0.1 can be set to adjust the probability. Figure 2 Value-based processing is used to enrich connectivity information. Furthermore, regarding the acquisition methods for the aforementioned centerlines and endpoints, centerlines can be extracted only for the main vessel trunk, free branches, and the second main vessel trunk, and their respective endpoints can be obtained using these centerlines. For candidate branches, their centerlines are obtained by taking the intersection of their centerlines with the centerlines of the second main vessel trunk, and these centerlines are then used to obtain the endpoints of the candidate branches. The endpoints of the main vessel trunk and candidate branches mentioned above are used not only for matching as described later, but also for removing noise that may be generated in the probability map foreground image after binarization with a smaller threshold.

[0038] The fourth step, such as Figure 2 As shown, the distance between the terminal point of each candidate branch and the corresponding terminal point on the main vessel is determined to be sufficiently small, for example, whether it is less than a predetermined second threshold. If it is less than the second threshold, the candidate branch is connected to the corresponding terminal point on the main vessel and displayed. This allows for the acquisition of richer vessel segmentation results.

[0039] In the fourth step, the judgment conditions can be further relaxed. That is, if there is no corresponding end point on the main trunk of the blood vessel, and the nearest distance between the end point of the candidate branch and the main trunk of the blood vessel is less than a predetermined third threshold (the third threshold may or may not be equal to the second threshold), then the candidate branch to which the end point belongs is connected to the nearest part of the main trunk of the blood vessel and displayed. This allows for richer segmentation results of the blood vessels.

[0040] As a variation, it can also be like... Figure 3 As shown, in the fourth step, when identifying the corresponding end point from the candidate branches that corresponds to the end point in the free branch, it is determined whether the distance of that corresponding end point from the corresponding end point on the main branch of the blood vessel is less than a second threshold. If it is less than the second threshold, the candidate branch to which that corresponding end point belongs is connected to the corresponding end point on the main branch of the blood vessel and displayed. This makes the segmentation results of the blood vessel more accurate.

[0041] To enrich the segmentation results of blood vessels, in the above modified example, it can be further set that if there is no corresponding terminal point on the main trunk of the blood vessel, and the distance between the corresponding terminal point and the nearest point on the main trunk of the blood vessel is small enough, for example, less than a predetermined third threshold, then the candidate branch to which the corresponding terminal point belongs is connected to the nearest part of the main trunk of the blood vessel and displayed.

[0042] Additionally, in the fourth step, larger branches among the free branches that do not correspond to candidate branches can also be displayed. Here, for example, free branches whose connected components are greater than a certain threshold, or whose distance between the two farthest points on the branch is greater than a certain threshold, can be considered as the aforementioned larger branches.

[0043] After the fourth step, it is preferable to have a fifth step for noise removal: using the end point of the candidate branch and the corresponding end point on the main trunk of the blood vessel, or using the center line of the candidate branch, remove the part of the candidate branch that connects to the main trunk of the blood vessel that causes the main trunk of the blood vessel to thicken, thereby removing noise and displaying it.

[0044] Specifically, in the fifth step, after removing noise using the endpoints of the candidate branches and the corresponding endpoints on the main vessel trunk, the endpoints on the main vessel trunk and the candidate branches are subtracted to obtain a vector. This vector, along with the endpoints on the main vessel trunk, yields the corresponding normal plane. Only the connected components on one side of the candidate branch endpoints are retained in the normal plane. For example, this could be obtained using the coordinates p of the matched endpoints on the main vessel trunk. t and the coordinates of the endpoints of the candidate branches p c Obtain vector Traverse each voxel point p in the candidate branch i , and vector The starting point p t Constructing vectors And filter with p t Located in vector Voxel points on the same side of the normal plane, and when At that time, remove voxel points p from the connection results. i Thus, the voxel points that constitute noise are removed. In the case of noise removal using the centerline of the candidate branch, the shortest distance between each voxel point of the vessel to be denoised and a point on the centerline of its corresponding candidate branch is calculated. If this shortest distance is greater than a predetermined fifth threshold, the voxel point is removed from the connection results. Alternatively, the fifth step can be omitted if noise removal is not required.

[0045] After the fifth step, preferably, there is an additional interactive step: if the user clicks on a candidate branch connected to the main branch of the blood vessel in the displayed result image, the candidate branch is deleted, or noise reduction is performed on the candidate branch. Specifically, if the user determines that a candidate branch connected to the main branch of the blood vessel is noise, or the user does not want to connect to that part, the user can click on that part, and the system will delete that part through connected component analysis. Alternatively, if the user wants to retain a candidate branch connected to the main branch of the blood vessel but wants to remove the noise it generates, the system can improve the noise in several ways and provide results for the user to choose from:

[0046] The fifth threshold used when removing noise using the centerline of the candidate branch is modified to a smaller value in order to remove more noise, or the fifth threshold is increased in order to make the connected blood vessels match the blood vessels in the image better if the connected blood vessels are considered to be too thin and do not match the blood vessels in the image.

[0047] If the previous fifth step used the terminal points of the candidate branches and the corresponding terminal points on the main vessel trunk for denoising, then switch to using the centerline of the candidate branches and re-denoising; or if the centerline of the candidate branches was used for denoising previously, switch to using the terminal points of the candidate branches and the corresponding terminal points on the main vessel trunk and re-denoising. In other words, switch the denoising method.

[0048] Apply different first thresholds to the probability graph and re-binarize it, then repeat the subsequent steps.

[0049] Here, if a larger branch is also displayed as described above, the following interactive steps can be implemented: If the user clicks on the larger branch in the displayed result image, the system will decrease the first threshold to re-determine whether the larger branch can be connected to the main branch of the blood vessel. More specifically, the system can calculate and provide the maximum first threshold that allows the larger branch to be connected to the main branch of the blood vessel (i.e., the second main branch can contain the larger branch), or it can provide connection results under multiple different first thresholds, including the maximum first threshold, for the user to choose from. Alternatively, if it determines that no matter how small the threshold is, the larger branch cannot be connected to the main branch of the blood vessel, the system can prompt the user that it cannot be connected.

[0050] Alternatively, the following interactive steps can be set:

[0051] If a user clicks on a blank area in the displayed result image, it may indicate that the user believes the blood vessel in that area should be connected or extended. In this case, the system reduces the aforementioned first threshold to reassess whether a branch connecting to the main branch of the blood vessel exists in the blank area. More specifically, the system can calculate and provide the maximum first threshold at which a branch connecting to the main branch of the blood vessel can appear in the blank area (i.e., the second main branch can contain a branch located in the blank area). Alternatively, it can provide connection results at multiple different first thresholds, including the maximum first threshold, for the user to choose from. Furthermore, if the system determines that no matter how small the threshold, a branch connecting to the main branch of the blood vessel cannot appear in the blank area, it can prompt the user that a connection is impossible.

[0052] Through the above embodiment 1, compared with the results of deep learning segmentation using existing technology shown on the left, it is possible to reduce fragmentation, extend the terminal portion, and remove irrelevant noise.

[0053] Next, refer to Figure 4 6. Another embodiment of the present invention, namely Example 2, will be described. In Example 2, a trachea will be used as an example of a tubular object. Additionally, descriptions of contents or figures identical to those in Example 1 may sometimes be omitted.

[0054] The tubular (trachea) segmentation method involved in Example 2 includes the following steps.

[0055] The first step is to acquire medical images containing the trachea;

[0056] The second step, as follows Figure 4 As shown, the output is a tracheal probability map of the tracheal data and a network segmentation result including the main trachea and free branches, and the centerline and end points on the centerline of the main trachea and free branches are obtained respectively. Figure 4 (c) and (d)), the main tracheal trunk ( Figure 4 (b) is the segmentation result ( Figure 4 The largest (or most branched, which can be determined by the length of the centerline) connected domain in (a) is the free branch that is a tracheal part that is not connected to the main trachea, except for the main trachea.

[0057] The third step involves thresholding the tracheal probability map, identifying regions within a predetermined first threshold range as the trachea, and determining the probability from this threshold. Figure 2 Extract the connected component with the largest volume (or the most branches, which can be determined by the length of the centerline) from the value-enhanced graph; this is the second main branch. Figure 5 (a)), and obtain the center line of the second main cadre ( Figure 5(b) and the endpoints on the centerline obtained based on the centerline, by using the second main stem ( Figure 5 (a) minus the main trachea ( Figure 4 (b) to obtain candidate branches ( Figure 5 (c) By taking the intersection of the centerlines of the candidate branch and the second main branch, the centerline of the candidate branch is obtained, and the other end points of the candidate branch other than the end point of the second main branch are obtained using the centerline of the candidate branch. Figure 5 (d)). In addition, the method of obtaining the first threshold, the center line and the end point is the same as in Example 1, and will not be repeated here.

[0058] The fourth step, as shown in Figure 6(a), determines whether the distance from the terminal point of each candidate branch to the corresponding terminal point on the main trachea is less than a predetermined second threshold. If it is less than the second threshold, the candidate branch is connected to the corresponding terminal point on the main trachea and displayed. Here, since the center lines and terminal points of the main trachea and free branches were obtained in the second step, and the center lines and terminal points of the candidate branches were obtained in the third step, as shown in Figure 6, these center lines and terminal points can be directly used for matching. This results in a richer segmentation of the trachea.

[0059] Alternatively, as in a variation of Example 1, for example, in the fourth step, if there is no corresponding endpoint on the main trachea to which the endpoint of a candidate branch is located, and if the nearest distance between the endpoint of the candidate branch and the main trachea is less than a predetermined third threshold (the third threshold may or may not be equal to the second threshold), then the candidate branch to which the endpoint belongs is connected to the nearest part of the main trachea and displayed (Figure 6(b)). This allows for a richer segmentation of the trachea.

[0060] As a variation, in the fourth step, when identifying the corresponding end point from the candidate branches that corresponds to the end point in the free branch, it is determined whether the distance of that corresponding end point from the corresponding end point on the main trachea is less than a second threshold. If it is less than the second threshold, the candidate branch to which that corresponding end point belongs is connected to the corresponding end point on the main trachea and displayed (Figure 6(c)). This can make the tracheal segmentation results more accurate.

[0061] To enrich the segmentation results of the trachea, in the above modified example, it can be further set that if there is no corresponding terminal point on the main trachea, and the distance between the corresponding terminal point and the nearest point on the main trachea is small enough, for example less than a predetermined third threshold, then the candidate branch to which the corresponding terminal point belongs is connected to the nearest part of the main trachea and displayed (Figure 6(d)).

[0062] While the aforementioned methods improve the connectivity and richness of the trachea, they may also increase the probability of introducing noise. Users can adjust their approach based on the model segmentation results and probabilities. Figure 2 Depending on the specific circumstances, such as valueization, choose to use one of them.

[0063] Alternatively, in the fourth step, as in Example 1, larger branches in the free branches that do not correspond to candidate branches can also be displayed.

[0064] The fifth step and the steps for interaction are the same as in Example 1, so they will not be described again.

[0065] Through the above embodiment 2, compared with the results of deep learning segmentation shown on the left, it is possible to connect the terminal or entire broken trachea, extend the entire trachea, and remove irrelevant noise.

[0066] In summary, according to the present invention, since the model output probability graph is utilized, more connectivity information can be included at a smaller probability value. On the other hand, noise introduced by the probability graph can also be effectively eliminated.

[0067] Furthermore, Examples 1 and 2 described above are merely illustrative and not intended to limit the invention. For instance, Example 1 can be applied to the trachea, and Example 2 to blood vessels, etc. However, in cases where the tubular structure branches off from the main trunk rather than from the extremities, or where the candidate trachea is relatively thick, it is difficult to select a suitable cross-section for the broken branch to remove noise superimposed on the main tubular structure. Therefore, the method of Example 2, which sets the fifth threshold and filters which voxel points of the candidate branch can be added to the segmentation result, is preferred.

[0068] Alternatively, the present invention can also be implemented as an apparatus using the above-described tubular segmentation method. For example, such an apparatus may include: an image acquisition unit for acquiring a medical image containing the tubular object; a processing unit; and a display unit, wherein the processing unit selects to perform the method described in the above embodiments. Alternatively, it can also be implemented as a program for causing a computer to execute the above-described method, or as a storage medium storing the program.

[0069] While several embodiments of the present invention have been described above, these embodiments are provided as examples and are not intended to limit the scope of the invention. These new embodiments can be implemented in various other ways, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included within the scope and spirit of the invention, and are also included within the scope of the invention as described in the claims and its equivalents.

Claims

1. A method for segmenting a tubular object, characterized in that, include: The first step is to acquire medical images containing the tubular structure; The second step is to output the tubular probability map for the tubular data and the network segmentation results including the main body of the tubular data and the free branches, and to obtain the center line and the end point on the center line of the main body of the tubular data and the free branches respectively. The main body of the tubular data is the connected region with the largest volume or the most branches in the segmentation results. The free branches are the tubular parts that are connected regions other than the main body of the tubular data and are not connected to the main body of the tubular data. The third step is to threshold the tubular probability map, taking the regions in the tubular probability map that fall within a specified first threshold range as tubular objects, extracting the connected domain with the largest volume or the most branches, i.e., the second main branch, from the obtained result, and obtaining the center line of the second main branch and the terminal point on the center line. Candidate branches are obtained by subtracting the tubular main branch from the second main branch, and the other terminal points of the candidate branches other than the terminal point of the second main branch are obtained using the center line of the candidate branches. as well as The fourth step is to determine whether the distance between the end point of each candidate branch and the corresponding end point on the main body of the tubular structure is less than a predetermined second threshold. If it is less than the second threshold, the candidate branch is connected to the corresponding end point on the main body of the tubular structure and displayed.

2. The method for segmenting tubular objects as described in claim 1, characterized in that, In the fourth step, if there is no end point on the main body of the tubular structure corresponding to the end point of the candidate branch, and if the nearest distance between the end point of the candidate branch and the main body of the tubular structure is less than a predetermined third threshold, then the candidate branch to which the end point belongs is connected to the nearest part of the main body of the tubular structure and displayed.

3. A method for segmenting a tubular object, characterized in that, include: The first step is to acquire medical images containing the tubular structure; The second step is to output the tubular probability map for the tubular data and the network segmentation results including the main body of the tubular data and the free branches, and to obtain the center line and the end point on the center line of the main body of the tubular data and the free branches respectively. The main body of the tubular data is the connected region with the largest volume or the most branches in the segmentation results. The free branches are the tubular parts that are connected regions other than the main body of the tubular data and are not connected to the main body of the tubular data. The third step is to threshold the tubular probability map, taking the region in the tubular probability map that falls within a specified first threshold range as a tubular object, extracting the connected region with the largest volume or the most branches, i.e., the second main branch, from the obtained result, and obtaining the center line of the second main branch and the terminal point on the center line. The candidate branch is obtained by subtracting the tubular main branch from the second main branch, and the other terminal points of the candidate branch other than the terminal point of the second main branch are obtained using the center line of the candidate branch. as well as The fourth step is to determine the corresponding end point in the candidate branches that corresponds to the end point in the free branch. Then, determine whether the distance between the corresponding end point and the corresponding end point on the main body of the tubular structure is less than the second threshold. If it is less than the second threshold, connect the candidate branch to which the corresponding end point belongs to the corresponding end point on the main body of the tubular structure and display it.

4. The method for segmenting tubular objects as described in claim 3, characterized in that, In the fourth step, if there is no corresponding end point on the main body of the tubular structure, and the nearest distance between the corresponding end point and the main body of the tubular structure is less than a predetermined third threshold, then the candidate branch to which the corresponding end point belongs is connected to the nearest part of the main body of the tubular structure and displayed.

5. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, Center lines are extracted from the main body of the tubular structure, the free branch, and the second main body, and their respective end points are obtained using the center lines. The center line of the candidate branch is obtained by taking the intersection of the center lines of the candidate branch and the second main body, and the end point of the candidate branch is obtained using the center line of the candidate branch.

6. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, In the fourth step, the determination is made using the centerline and end point of the main body of the tubular structure, the free branches, and the candidate branches.

7. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, In the fourth step, the larger branches in the free branches that do not correspond to the candidate branches are also displayed.

8. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, Also includes: The fifth step involves using the end point of the candidate branch and the corresponding end point on the main body of the tubular structure, or using the center line of the candidate branch, to remove the portion of the candidate branch that causes the main body of the tubular structure to thicken, thereby removing noise and displaying the result.

9. The method for segmenting tubular objects as described in claim 8, characterized in that, In the fifth step, after removing noise using the end points of the candidate branches and the corresponding end points on the main body of the tubular structure, the vector is obtained by subtracting the end points on the main body of the tubular structure from the end points of the candidate branches. The corresponding normal plane is obtained by using this vector and the end points on the main body of the tubular structure, and only the connected components on the side of the end points of the candidate branches of the normal plane are retained. When using the centerline of the candidate branch to remove noise, the shortest distance between each voxel point of the tubular structure to be denoised and the point on the centerline of the corresponding candidate branch is calculated. If the shortest distance is greater than a predetermined fifth threshold, the voxel point is removed from the connection result.

10. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, It also includes the following steps: If a user clicks on a candidate branch connected to the main trunk of the tubular structure in the displayed result image, that candidate branch is deleted.

11. The method for segmenting tubular objects as described in claim 9, characterized in that, It also includes the following steps: If the user clicks on a candidate branch connected to the main trunk of the tubular structure in the displayed result image, the fifth threshold is changed and the branch is re-denoised and then displayed.

12. The method for segmenting tubular objects as described in claim 8, characterized in that, It also includes the following steps: If the user clicks on a branch connected to the main body of the tubular structure in the displayed result image, then the fifth step will switch between using the end point of the candidate branch and the corresponding end point on the main body of the tubular structure, and will also switch to using the center line of the candidate branch, and will be re-denoised and displayed.

13. The method for segmenting tubular objects as described in claim 8, characterized in that, It also includes the following steps: If the user clicks on a branch connected to the main trunk of the tubular structure in the displayed result image, the first threshold in the third step is modified, and the steps starting from the third step are repeated.

14. The method for segmenting tubular objects as described in claim 7, characterized in that, It also includes the following steps: If the user clicks on the larger branch in the displayed result image, the first threshold is reduced to re-evaluate whether the larger branch can be connected to the main body of the tubular structure.

15. The method for segmenting a tubular object as described in claim 14, characterized in that, It also includes the following steps: The display shows the largest first threshold that allows the larger branch to be connected to the main body of the tubular structure.

16. The method for segmenting a tubular object as described in claim 15, characterized in that, It also includes the following steps: Calculate and display the connection results under multiple different first thresholds, including the largest first threshold.

17. The method for segmenting tubular objects as described in claim 14, characterized in that, It also includes the following steps: If it is determined that the larger branch cannot be connected to the main body of the tubular structure by reducing the first threshold, a message indicating that connection is not possible is displayed.

18. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, It also includes the following steps: If the user clicks on a blank area of ​​the displayed result image, the first threshold is reduced to re-determine whether there is a branch connected to the main trunk of the tubular structure in that blank area.

19. The method for segmenting a tubular object as described in claim 18, characterized in that, It also includes the following steps: Calculate and display the largest first threshold at which a branch connecting to the main body of the tubular structure can appear in the blank space.

20. The method for segmenting a tubular object as described in claim 19, characterized in that, It also includes the following steps: Displays connection results under multiple different first thresholds, including the largest first threshold.

21. The method for segmenting tubular objects as described in claim 18, characterized in that, It also includes the following steps: If it is determined that a branch connected to the main body of the tubular structure cannot be found in the blank space by reducing the first threshold, a message indicating that connection is not possible is displayed.

22. The method for segmenting a tubular object as described in any one of claims 1 to 4, characterized in that, The tubular structure is a blood vessel and / or trachea.

23. A tubular object dividing device, characterized in that, include: The image acquisition unit acquires medical images containing the tubular object; Processing Department; as well as Display section, The processing unit performs the tubular segmentation method according to any one of claims 1 to 22.

24. A computer-readable storage medium storing a program for causing a computer to perform the tubular segmentation method according to any one of claims 1 to 22.

Images