Fracture repair and topology analysis method for blood vessel segmentation mask in liver CT image

Abstract

The invention provides a fracture repair and topology analysis method for a blood vessel segmentation mask in a liver CT image, and the method specifically comprises the steps: 1) carrying out the skeletonization of the liver blood vessel segmentation mask, and extracting a blood vessel skeleton, a skeleton end point, and a blood vessel radius rough value at a skeleton point; 2) solving a connected blood vessel center line (1) for the purpose of fracture repair without considering removal of wrong adhesion of hepatic veins and portal veins; 3) considering possible wrong adhesion for removing the hepatic vein and the portal vein, setting a stricter connection condition to obtain a broken blood vessel center line, and solving a broken blood vessel center line (2) in which the hepatic vein and the portal vein are obviously distinguished by taking classification as a purpose; and 4) combining the center lines (1) and (2), reconstructing missing blood vessel segments in the original blood vessel segmentation mask, and combining the reconstructed blood vessel segments into the original segmentation mask to obtain a final blood vessel mask and center line topological structure after fracture repair.

Description

Technical Field

[0001] The invention relates to an image analysis technology, belonging to the technical field of computer image processing, and in particular to a method for repairing the breakage and topological analysis of a blood vessel segmentation mask in a liver CT image. Background Art

[0002] In the field of medical image analysis, the processing and analysis of liver CT images has become an important research direction, especially in the diagnosis and treatment planning of liver lesions. The vascular structure in liver CT images is crucial anatomical information in diagnosis and treatment. The accurate segmentation and reconstruction of blood vessels and the extraction and classification of vascular centerlines are of great significance for subsequent lesion detection and surgical planning.

[0003] In recent years, the technology of blood vessel segmentation in CT images has been highly developed. The representative technology is deep learning technology typified by convolutional neural networks, which has been widely used in medical image analysis. These algorithms can quickly and accurately segment blood vessels in liver CT images, but in complex vascular morphology, blood vessels are prone to rupture or adhesion, and it is difficult to distinguish between the hepatic vein and the portal vein, which affects subsequent analysis and processing. Therefore, the mask obtained by the segmentation algorithm usually needs to be post-processed according to application requirements. Many post-processing methods usually focus on a single task, often only focusing on repairing the connectivity of blood vessels or simply classifying blood vessels, which is not enough to meet the needs of practical applications. Summary of the invention

[0004] In order to solve the problems of vascular breakage, different vascular adhesions and the classification of vascular masks in vascular segmentation masks of liver CT images, the present invention provides a method for fracture repair and topological analysis of vascular segmentation masks in liver CT images, which can simultaneously obtain the fracture repaired vascular mask and the centerline topological structure to meet the needs of vascular segmentation mask analysis.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for repairing and topologically analyzing a blood vessel segmentation mask in a liver CT image comprises the following steps:

[0007] Step 1: segment the blood vessels in the liver CT image using the mask S mask Skeletonize and get the skeleton point P skel and the skeleton endpoint P skel_end , and the rough value of the vessel radius R is estimated based on the maximum radial distance method rough ;

[0008] Step 2: Based on the blood vessel skeleton point P obtained in step 1 skel , Skeleton endpoint P skel_endand the vessel radius R rough Without considering the removal of the wrong adhesions of the hepatic vein and portal vein, the center line C of the adhesion vessel is directly formed from the skeleton points through depth-first search with the purpose of fracture repair. adhesion , and then C adhesion Connect the broken part and obtain the center line of the connected blood vessel①;

[0009] Step 3: Based on the blood vessel skeleton point P obtained in step 1 skel , Skeleton endpoint P skel_end and the rough value R of the blood vessel radius obtained in step 2 rough , consider removing possible erroneous adhesions between the hepatic vein and the portal vein, set more stringent connection conditions, add constraints such as tangent vector angle and branch length in the depth-first search process, and form the broken vascular centerline C from the skeleton point noadhesion ; To distinguish the hepatic vein from the portal vein, noadhesion Classification was performed to obtain the vascular centerline including the rupture and with the hepatic vein and portal vein clearly distinguished②;

[0010] Step 4: merge and cut the center line ① obtained in step 2 and the center line ② obtained in step 3 based on the vascular branch to obtain the merged vascular center line C merge That is, the vascular topology and the connecting line C connect ;

[0011] Step 5: Based on the blood vessel segmentation mask S mask and the connecting line C obtained in step 4 connect , the radial heuristic method of the centerline tangent vector is used to accurately calculate the radius of the blood vessels of the two branches that should be connected. exact , and R exact The mean value is taken as the radius of the connecting vessel R connect , reconstruct the missing blood vessel segment S in the original blood vessel segmentation mask connect , and S connect Merge into the original segmentation mask S mask The final complete blood vessel mask is obtained.

[0012] Furthermore, the step 2 specifically includes the following process:

[0013] Step 21) Skeleton endpoint P skel_end According to the rough value of blood vessel radius R rough Sort from large to small to get the sorted skeleton breakpoints, denoted as P skel_end_sorted ;

[0014] Step 22) P skel_end_sorted The first endpoint in is taken as the starting point, and the skeleton points in the 26 domains are connected in sequence using the depth-first search method, and P is updated. skel , Pskel_end_sorted When no skeleton point is found within the 26-neighborhood, it means that the center line C of the vascular tree tree Generation completed;

[0015] Step 23) Delete the center line C of the vascular tree tree The leaf branches with less than n branch nodes are used to obtain the corrected vascular tree centerline C. tree_correct ;

[0016] Step 24: Correct the center line of the vascular tree C tree_correct Join the collection of adherent vessel centerlines;

[0017] Step 25: Loop through the above steps 22) to 24 until P skel_end_sorted Empty, get the center line C of the adhered blood vessel adhesion ;

[0018] Step 26: Select a main vascular tree T according to the number of nodes contained in the vascular tree main , the rest are sub-vessel trees T sub ;

[0019] Step 27: Connection probability P of the sub-vessel tree branch endpoints and the main vessel tree nodes connect , solve the following equation:

[0020] P connect =1-(dis+curdif)

[0021] Where dis is the distance between all branch endpoints of the sub-vessel tree and all points on the main vessel tree, and curdif is the difference in curvature between all branch endpoints of the sub-vessel tree and all points on the main vessel tree.

[0022] Calculate the connection probability P between each branch endpoint of the sub-vascular tree and the node of the main vascular tree connect , and connect P connect The two endpoints corresponding to the maximum value are used to update the main vessel tree and the sub-vessel tree;

[0023] Step 28: The connection probability solution process in step 27 is calculated repeatedly until the sub-vessel tree is empty, and the connected vessel centerline ① is obtained.

[0024] Furthermore, the step 3 considers the adhesion of the hepatic vein and the portal vein, including the ruptured vascular centerline ② where the hepatic vein and the portal vein are clearly distinguished, and specifically includes the following process:

[0025] Step 31: Skeleton Endpoint P skel_end According to the rough value of blood vessel radius R rough Sort from large to small to get the sorted skeleton breakpoints, denoted as P skel_end_sorted ;

[0026] Step 32: P skel_end_sorted The first endpoint in is taken as the starting point, and the skeleton points in the 26 domains are connected in sequence using the depth-first search method. It is required that the cosine value of the angle between the vector between the skeleton point and the connection point and the vector of all points in the vascular branch where the skeleton point is located is less than a given threshold a, and P is updated. skel , P skel_end_sorted When no skeleton point is found within the 26-neighborhood, it means that the center line C of the vascular tree tree Generation completed;

[0027] Step 33: Delete the center line C of the vessel tree tree The leaf branches with less than n branch nodes in the middle obtain the corrected vascular tree centerline C tree_correct ; n is 3 or 5;

[0028] Step 34: Delete the corrected vascular tree centerline C tree_correct The non-leaf branches and their sub-branches whose number of branch nodes is less than m, and update P skel , P skel_end_sorted , and obtain the corrected vascular tree centerline C tree_rcorrect ;

[0029] Step 35: The corrected vascular tree centerline C tree_rcorrect Join the broken vessel centerline set;

[0030] Step 36: Repeat the above steps 32 to 35 until P skel_end_sorted Empty, finally obtain the broken blood vessel centerline C noadhesion ;

[0031] Step 37: Select two main vascular trees T according to the number of nodes contained in the vascular tree main1 , T main2 , respectively the hepatic vein and portal vein, and the rest are sub-vascular trees T sub ;

[0032] Step 38: The probability P of the sub-vessel tree belonging to the main vessel tree belong , solve the following equation:

[0033] P belong =1-(w1×dis+w2×curdif+C)

[0034] C=w3×(1-cosα)+w4×(1-cos(β+γ))

[0035] Where dis is the distance between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and its weight is recorded as w1, curdif is the difference in curvature between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and its weight is recorded as w2, C represents the directional similarity score, and the smaller the value, the more similar the directions of the two lines are. α is the angle between the direction vectors of all branch endpoints of the sub-vessel tree and the direction vectors of all points on the main vascular tree, β is the angle between the breakpoint connecting line and the branch where the breakpoint of the main vascular tree is located, γ is the angle between the breakpoint connecting line and the branch where the breakpoint of the sub-vessel tree is located, and w3 and w4 are weighting coefficients used to adjust the influence of different angle items on the total similarity score. Where w1, w2, w3, and w4 are determined according to actual conditions, such as: for the hepatic vein, w1 = 1.5, w2 = 0.01, w3 = 0.1, w4 = 1.0, for the portal vein, w1 = 2, w2 = 0.1, w3 = 0.1, w4 = 1.0.

[0036] Calculate the probability P of the sub-vessel tree belonging to each main vessel tree belong , and connect two groups of P belong The two endpoints corresponding to the maximum value in the middle update the corresponding main vessel tree and sub-vessel tree;

[0037] Step 39: The attribution probability solving process in step 38 is cyclically calculated until the sub-vessel tree is empty, and the vessel centerline ② containing the fracture and with the hepatic vein and portal vein clearly distinguished is obtained.

[0038] Furthermore, the merging in step 4 includes the following process:

[0039] According to the center line ②, determine which category each point of each branch in the center line ① belongs to. If more than half of the points on the center line of a branch belong to the portal vein, then the branch belongs to the portal vein. If more than half of the points belong to the hepatic vein, then the branch belongs to the hepatic vein. If the judgment result is that the branch belongs to neither the portal vein nor the hepatic vein, then eliminate the wrong vascular segment;

[0040] The hepatic vein and portal vein merge into the combined vascular centerline C merge , get the vascular topology and connection line C connect .

[0041] Furthermore, the reconstruction of the broken blood vessel segment and the mask merging in step 5 include the following processes:

[0042] Step 51: Use Euclidean distance transformation to obtain the distance from each point in the blood vessel to the nearest boundary;

[0043] Step 52: near each centerline point of the blood vessel, expand along the orthogonal direction of the tangent vector until reaching the boundary of the blood vessel;

[0044] Step 53: The median of the distances extended along multiple orthogonal directions is the radius of the blood vessel at that location;

[0045] Step 54: The average value of the vascular radius at all points on the center line of the two connected branches is taken as the vascular radius R at the connecting line connect ;

[0046] Step 55: Connect Line C connect and the vessel radius R at the connecting line connect , based on R connect The spherical structure is reconstructed using the binary expansion method to reconstruct the broken blood vessel segment S connect ;

[0047] Step 56: Merge S connect and the original segmentation mask to obtain the final complete blood vessel mask.

[0048] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0049] The present invention is different from the conventional method of only performing connectivity repair or centerline classification. The corresponding connection conditions are set according to different purposes, and the connection centerline and the classification centerline are obtained from the skeleton points respectively. The connected vascular centerline and the classified vascular centerline are then merged, and finally reconstructed and visualized. The extraction of vascular centerlines, the repair of broken vascular segments and the topological analysis of blood vessels can be realized at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic flow chart of a method for break repair and topological analysis of a blood vessel segmentation mask in a liver CT image provided by the present invention.

[0051] Figure 2 A schematic diagram of the results of a method for repairing the breakage and topological analysis of the blood vessel segmentation mask in a liver CT image provided by the present invention. (a) The original segmentation mask and the corresponding centerline, the circle is the broken blood vessel, and the white centerline is the unclassified centerline before processing; (b) is the image after the method of the present invention is applied, the breakage at the circle has been repaired according to the algorithm, and the centerline distinguishes the hepatic vein (blue) and the portal vein (green). DETAILED DESCRIPTION

[0052] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0053] This embodiment provides a method for repairing the breakage and topological analysis of the blood vessel segmentation mask in the liver CT image, and its process is as follows: Figure 1 As shown, the following steps are included:

[0054] Step 101: segment the blood vessels in the liver CT image using a mask S mask Skeletonize and get the skeleton point P skel and the skeleton endpoint P skel_end , and the rough value of the vessel radius R is estimated based on the maximum radial distance method rough .

[0055] In this step, the input is the segmentation mask of the blood vessels in the liver CT image based on deep learning, and the skeleton extraction method based on Euler features is used to extract the blood vessel skeleton points P skel and the skeleton endpoint P skel_end , where the number of loops is set to 20. Then, the vascular boundary is calculated based on the vascular segmentation mask. Then, the distance R from the vascular skeleton node to the nearest non-zero coordinate point on the vascular boundary is calculated on the XY, YZ, and XZ planes respectively. XY , R YZ , R XZ Finally, R XY , R YZ , R XZ The maximum value is taken as the rough value R of the blood vessel radius at the skeleton point rough .

[0056] Step 102: based on the blood vessel skeleton point P obtained in step 101 skel , skeleton endpoint P skel_end and the vessel radius R rough Without considering the removal of the wrong adhesions of the hepatic vein and portal vein, the center line C of the adhesion vessel is directly formed from the skeleton points through depth-first search with the purpose of fracture repair. adhesion , and then C adhesion Connect the broken part and obtain the center line of the connected blood vessel①.

[0057] In this step, the specific implementation plan is as follows:

[0058] 1) Skeleton endpoint P skel_end According to the rough value of blood vessel radius R rough Sort from large to small to get the sorted skeleton breakpoints, denoted as P skel_end_sorted ;

[0059] 2) P skel_end_sorted The first endpoint in is taken as the starting point, and the skeleton points in the 26 domains are connected in sequence using the depth-first search method, and P is updated. skel , P skel_end_sorted When no skeleton point is found within the 26-neighborhood, it means that the center line C of the vascular treetree Generation completed;

[0060] 3) Delete the center line C of the vascular tree tree The leaf branches with less than n branch nodes are used to obtain the corrected vascular tree centerline C. tree_correct ;

[0061] 4) The corrected vascular tree centerline C tree_correct Join the collection of adherent vessel centerlines;

[0062] 5) Repeat the above steps 2) to 4) until P skel_end_sorted Empty, get the center line C of the adhered blood vessel adhesion ;

[0063] 6) According to the number of nodes contained in the vascular tree, select a main vascular tree T main , the rest are sub-vessel trees T sub ;

[0064] 7) Connection probability P between the endpoints of the sub-vessel tree branches and the nodes of the main vessel tree connect , solve the following equation:

[0065] P connect =1-(dis+curdif)

[0066] Where dis is the distance between all branch endpoints of the sub-vessel tree and all points on the main vessel tree, and curdif is the difference in curvature between all branch endpoints of the sub-vessel tree and all points on the main vessel tree.

[0067] Calculate the connection probability P between each branch endpoint of the sub-vascular tree and the node of the main vascular tree connect , and connect P connect The two endpoints corresponding to the maximum value are used to update the main vessel tree and the sub-vessel tree;

[0068] 8) The connection probability solution process in step 7) is calculated repeatedly until the sub-vessel tree is empty, and the connected vessel centerline ① is obtained.

[0069] Step 103: based on the blood vessel skeleton point P obtained in step 101 skel , skeleton endpoint P skel_end and the rough value R of the blood vessel radius obtained in step 102 rough , consider removing possible erroneous adhesions between the hepatic vein and the portal vein, set more stringent connection conditions, add constraints such as tangent vector angle and branch length in the depth-first search process, and form the broken vascular centerline C from the skeleton point noadhesion ; To distinguish the hepatic vein from the portal vein, noadhesion After classification, the vascular centerline containing the rupture and clearly distinguishing the hepatic vein and portal vein was obtained②.

[0070] In this step, the specific implementation plan is as follows:

[0071] 1) Skeleton endpoint P skel_end According to the rough value of blood vessel radius R rough Sort from large to small to get the sorted skeleton breakpoints, denoted as P skel_end_sorted ;

[0072] 2) P skel_end_sorted The first endpoint in is taken as the starting point, and the skeleton points in the 26 domains are connected in sequence using the depth-first search method. It is required that the cosine value of the angle between the vector between the skeleton point and the connection point and the vector of all points in the vascular branch where the skeleton point is located is less than a given threshold a, and P is updated. skel , P skel_end_sorted When no skeleton point is found within the 26-neighborhood, it means that the center line C of the vascular tree tree Generation completed;

[0073] 3) Delete the center line C of the vascular tree tree The leaf branches with less than n branch nodes in the middle obtain the corrected vascular tree centerline C tree_correct ;

[0074] 4) Delete the corrected vascular tree centerline C tree_correct The non-leaf branches and their sub-branches whose number of branch nodes is less than m, and update P skel , P skel_end_sorted , and obtain the corrected vascular tree centerline C tree_rcorrect ;

[0075] 5) The corrected vascular tree centerline C tree_rcorrect Join the broken vessel centerline set;

[0076] 6) Repeat the above steps 2) to 5) until P skel_end_sorted Empty, finally obtain the broken blood vessel centerline C noadhesion ;

[0077] 7) According to the number of nodes contained in the vascular tree, select two main vascular trees T main1 , T main2 , respectively the hepatic vein and portal vein, and the rest are sub-vascular trees T sub ;

[0078] 8) The probability P of a sub-vessel tree belonging to the main vessel tree belong , solve the following equation:

[0079] P belong =1-(w1×dis+w2×curdif+C)

[0080] C=w3×(1-cosα)+w4×(1-cos(β+γ))

[0081] Where dis is the distance between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and its weight is recorded as w1, curdif is the difference in curvature between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and its weight is recorded as w2, C represents the directional similarity score, and the smaller the value, the more similar the directions of the two lines are. α is the angle between the direction vectors of all branch endpoints of the sub-vessel tree and the direction vectors of all points on the main vascular tree, β is the angle between the breakpoint connecting line and the branch where the breakpoint of the main vascular tree is located, γ is the angle between the breakpoint connecting line and the branch where the breakpoint of the sub-vessel tree is located, and w3 and w4 are weighting coefficients used to adjust the influence of different angle items on the total similarity score. Where w1, w2, w3, and w4 are determined according to actual conditions, such as: for the hepatic vein, w1 = 1.5, w2 = 0.01, w3 = 0.1, w4 = 1.0, for the portal vein, w1 = 2, w2 = 0.1, w3 = 0.1, w4 = 1.0.

[0082] Calculate the probability P of the sub-vessel tree belonging to each main vessel tree belong , and connect two groups of P belong The two endpoints corresponding to the maximum value in the middle update the corresponding main vessel tree and sub-vessel tree;

[0083] 9) The attribution probability solving process in step 8) is calculated repeatedly until the sub-vessel tree is empty, and the vascular centerline ② containing the fracture and with the hepatic vein and portal vein clearly distinguished is obtained.

[0084] Step 104, the center line ① obtained in step 102 and the center line ② obtained in step 103 are merged and cut in units of blood vessel branches to obtain a merged blood vessel center line C merge That is, the vascular topology and the connecting line C connect .

[0085] In this step, the specific implementation plan is as follows:

[0086] According to the center line ②, determine which category each point of each branch in the center line ① belongs to. If more than half of the points on the center line of a branch belong to the portal vein, then the branch belongs to the portal vein. If more than half of the points belong to the hepatic vein, then the branch belongs to the hepatic vein. If the judgment result is that the branch belongs to neither the portal vein nor the hepatic vein, then eliminate the wrong vascular segment;

[0087] The hepatic vein and portal vein merge into the combined vascular centerline C merge , get the vascular topology and connection line C connect .

[0088] Step 105: based on the blood vessel segmentation mask S mask and the connecting line C obtained in step 104 connect , the radial heuristic method of the centerline tangent vector is used to accurately calculate the radius of the blood vessels of the two branches that should be connected. exact , and R exact The mean value is taken as the radius of the connecting vessel R connect , reconstruct the missing blood vessel segment S in the original blood vessel segmentation mask connect , and S connect Merge into the original segmentation mask S mask The final complete blood vessel mask is obtained.

[0089] In this step, the specific implementation plan is as follows:

[0090] 1) Use Euclidean distance transform to obtain the distance from each point in the blood vessel to the nearest boundary;

[0091] 2) Near each centerline point of the blood vessel, expand along the orthogonal direction of the tangent vector until reaching the boundary of the blood vessel;

[0092] 3) The median of the distances extended along multiple orthogonal directions is the radius of the blood vessel at that location;

[0093] 4) The average value of the vascular radius at all points on the center line of the two connecting branches is taken as the vascular radius R at the connecting line. connect ;

[0094] 5) According to the connection line C connect and the vessel radius R at the connecting line connect , based on R connect The spherical structure of the original liver CT size was reconstructed using the binary expansion method to reconstruct the broken vessel segment S connect ;

[0095] 6) Use logical OR operation to merge the blood vessel segmentation mask S mask and vessel segment S connect , and obtain the final complete blood vessel mask S result ;

[0096] 7) Finally, the merged vascular mask is reconstructed based on the application of the surface reconstruction algorithm; using geometric reconstruction, multiple discrete points are connected to form a continuous line to reconstruct the vascular topology.

[0097] like Figure 2 As shown, (a) the original segmentation mask and the corresponding centerline, the circle is the broken blood vessel, and the white centerline is the unclassified centerline before processing; (b) is the image after the method of the present invention is applied, the break at the circle has been repaired according to the algorithm, and the centerline distinguishes the hepatic vein (blue) and the portal vein (green).

[0098] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above technical features.

Claims

1. A method for repairing and topological analysis of blood vessel segmentation masks in liver CT images, characterized in that: The steps include: Step 1: segment the blood vessels in the liver CT image using the mask S mask Skeletonize and get the skeleton point P skel and the skeleton endpoint P skel_end , and the rough value of the vessel radius R is estimated based on the maximum radial distance method rough ; Step 2: Based on the blood vessel skeleton point P obtained in step 1 skel , Skeleton endpoint P skel_end and the vessel radius R rough Without considering the removal of the wrong adhesions of the hepatic vein and portal vein, the center line C of the adhesion vessel is directly formed from the skeleton points through depth-first search with the purpose of fracture repair. adhesion , and then C adhesion Connect the broken part and obtain the center line of the connected blood vessel①; Step 3: Based on the blood vessel skeleton point P obtained in step 1 skel , Skeleton endpoint P skel_end and the rough value R of the blood vessel radius obtained in step 2 rough , consider removing possible erroneous adhesions between the hepatic vein and the portal vein, set more stringent connection conditions, add constraints such as tangent vector angle and branch length in the depth-first search process, and form the broken vascular centerline C from the skeleton point noadhesion ; To distinguish the hepatic vein from the portal vein, noadhesion Classification was performed to obtain the vascular centerline including the rupture and with the hepatic vein and portal vein clearly distinguished②; Step 4: merge and segment the centerline ① obtained in step 2 and the centerline ② obtained in step 3 based on the vascular branches to obtain the merged vascular centerline topology structure C merge And connecting line C connect ; Step 5: Based on the blood vessel segmentation mask S mask and the connecting line C obtained in step 4 connect , the radial heuristic method of the centerline tangent vector is used to accurately calculate the radius of the blood vessels of the two branches that should be connected. exact , and R exact The mean value is taken as the radius of the connecting vessel R connect , reconstruct the missing blood vessel segment S in the original blood vessel segmentation mask connect , and S connect Merge into the original segmentation mask S mask The final complete blood vessel mask is obtained.

2. The method for repairing the breakage and topological analysis of the blood vessel segmentation mask in the liver CT image according to claim 1, characterized in that: The step 2 specifically includes the following process: Step 21: Skeleton Endpoint P skel_end According to the rough value of blood vessel radius R rough Sort from large to small to get the sorted skeleton breakpoints, denoted as P skel_end_sorted ; Step 22: P skel_end_sorted The first endpoint in is taken as the starting point, and the skeleton points in the 26 domains are connected in sequence using the depth-first search method, and P is updated. skel , P skel_end_sorted When no skeleton point is found within the 26-neighborhood, it means that the center line C of the vascular tree tree Generation completed; Step 23: Delete the center line C of the vessel tree tree The leaf branches with less than n branch nodes are used to obtain the corrected vascular tree centerline C. tree_correct , n is 3 or 5; Step 24: Correct the center line of the vascular tree C tree_correct Join the collection of adherent vessel centerlines; Step 25: Repeat steps 22 to 24 until P skel_end_sorted Empty, get the center line C of the adhered blood vessel adhesion ; Step 26: Select a main vascular tree T according to the number of nodes contained in the vascular tree main , the rest are sub-vessel trees T sub ; Step 27: Connection probability P of the sub-vessel tree branch endpoints and the main vessel tree nodes connect , solve the following equation: P connect =1-(dis+curdif) Where dis is the distance between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and curdif is the difference in curvature between all branch endpoints of the sub-vessel tree and all points on the main vascular tree; Calculate the connection probability P between each branch endpoint of the sub-vascular tree and the node of the main vascular tree connect , and connect P connect The two endpoints corresponding to the maximum value are used to update the main vessel tree and the sub-vessel tree; 28: The connection probability solution process in step 27 is calculated repeatedly until the sub-vessel tree is empty, and the connected vessel centerline ① is obtained.

3. The method for repairing the breakage and topological analysis of the blood vessel segmentation mask in the liver CT image according to claim 1, characterized in that: Step 3 considers the adhesion of the hepatic vein and portal vein, including the ruptured vascular centerline② where the hepatic vein and portal vein are clearly distinguished, and specifically includes the following process: Step 31: Skeleton Endpoint P skel_end According to the rough value of blood vessel radius R rough Sort from large to small to get the sorted skeleton breakpoints, denoted as P skel_end_sorted ; Step 32: P skel_end_sorted The first endpoint in is taken as the starting point, and the skeleton points in the 26 domains are connected in sequence using the depth-first search method. It is required that the cosine value of the angle between the vector between the skeleton point and the connection point and the vector of all points in the vascular branch where the skeleton point is located is less than a given threshold a, and P is updated. skel , P skel_end_sorted When no skeleton point is found within the 26-neighborhood, it means that the center line C of the vascular tree tree Generation completed; Step 33: Delete the center line C of the vessel tree tree The leaf branches with less than n branch nodes in the middle obtain the corrected vascular tree centerline C tree_correct ; Step 34: Delete the corrected vessel tree centerline C tree_correct The non-leaf branches and their sub-branches whose number of branch nodes is less than m, and update P skel , P skel_end_sorted , and obtain the corrected vascular tree centerline C tree_rcorrect ; Step 35: The corrected vascular tree centerline C tree_rcorrect Join the broken vessel centerline set; Step 36: Repeat the above steps 32 to 35 until P skel_end_sorted Empty, finally obtain the broken blood vessel centerline C noadhesion ; Step 37: Select two main vascular trees T according to the number of nodes contained in the vascular tree main1 、T main2 , respectively the hepatic vein and portal vein, and the rest are sub-vascular trees T sub ; Step 38: The probability P of the sub-vessel tree belonging to the main vessel tree belong , solve the following equation: P belong =1-(w1×dis+w2×curdif+C) C=w3×(1-cosα)+w4×(1-cos(β+γ)) Where dis is the distance between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and its weight is w1. Curdif is the difference in curvature between all branch endpoints of the sub-vessel tree and all points on the main vascular tree, and its weight is w2. C represents the directional similarity score. The smaller the value, the more similar the directions of the two lines are. α is the angle between the direction vectors of all branch endpoints of the sub-vessel tree and the direction vectors of all points on the main vascular tree. β is the angle between the breakpoint connecting line and the branch where the breakpoint of the main vascular tree is located. α is the angle between the breakpoint connecting line and the branch where the breakpoint of the sub-vessel tree is located. w3 and w4 are weighting coefficients used to adjust the influence of different angle items on the total similarity score. For the hepatic vein, w1 = 1.5, w2 = 0.01, w3 = 0.1, w4 = 1.

0. For the portal vein, w1 = 2, w2 = 0.1, w3 = 0.1, w4 = 1.

0. Calculate the probability P of the sub-vessel tree belonging to each main vessel tree belong , and connect two groups of P belong The two endpoints corresponding to the maximum value in the middle update the corresponding main vessel tree and sub-vessel tree; Step 39: The attribution probability solving process in step 38 is cyclically calculated until the sub-vessel tree is empty, and the vessel centerline ② containing the fracture and with the hepatic vein and portal vein clearly distinguished is obtained.

4. The method for repairing the breakage and topological analysis of the blood vessel segmentation mask in the liver CT image according to claim 1, characterized in that: The merging method in step 4 is as follows: According to the center line ②, determine which category each point of each branch in the center line ① belongs to. If more than half of the points on the center line of a branch belong to the portal vein, then the branch belongs to the portal vein. If more than half of the points belong to the hepatic vein, then the branch belongs to the hepatic vein. If the judgment result is that the branch belongs to neither the portal vein nor the hepatic vein, then eliminate the wrong vascular segment; The hepatic vein and portal vein merge into the combined vascular centerline C merge , get the vascular topology and connection line C connect .

5. The method for repairing the breakage and topological analysis of the blood vessel segmentation mask in the liver CT image according to claim 1, characterized in that: The step 5 specifically includes the following process: Step 51: Use Euclidean distance transformation to obtain the distance from each point in the blood vessel to the nearest boundary; Step 52: near each centerline point of the blood vessel, expand along the orthogonal direction of the tangent vector until reaching the boundary of the blood vessel; Step 53: The median of the distances extended along multiple orthogonal directions is the radius of the blood vessel at that location; Step 54: The average value of the vascular radius at all points on the center line of the two connected branches is taken as the vascular radius R at the connecting line connect ; Step 55: Connect Line C connect and the blood vessel radius R at the connecting line connect , based on R connect The spherical structure is reconstructed using the binary expansion method to reconstruct the broken blood vessel segment S connect ; Step 56: Merge S connect and the original segmentation mask to obtain the final complete blood vessel mask.

Images