A method and system for automatic segmentation of lung organ models based on CT images

Abstract

The present invention discloses a method for automatic lobulation of a lung organ model based on CT images, comprising the following steps: importing the original data ImageData of chest sequence CT images; extracting the lung parenchyma region by using a threshold segmentation algorithm, obtaining a binary image of the lung parenchyma and recording it as MASK1, and then performing three-dimensional reconstruction on the detected and segmented lung parenchyma region; calculating the maximum bounding box of the lung parenchyma according to whether there are binary points on the outermost boundary of the binary image of the lung parenchyma, performing enhancement calculation on the sheet structure, obtaining enhanced image data MASK2, further processing to obtain a binary image MASK3, extracting the set of all points of the left and right binary data and recording them as left lung point cloud S1 and right lung point cloud S2; using the left lung point cloud S1 as input data to perform left lung cutting and segmentation; using the right lung point cloud S2 as input data to perform right lung primary cutting and right lung secondary cutting in sequence. The present invention can reduce the interference of non-lung areas on the extraction of lung interstitial point clouds, improve the operation efficiency, and can significantly reduce the workload without manual intervention.

Description

Technical Field

[0001] The present invention relates to a medical 3D image visualization processing method, and in particular to a CT image-based lung organ model automatic lobing method and system. Background Art

[0002] Currently, lung cancer has become the world's leading cause of death, and the most effective way to improve the survival rate of lung cancer is early detection, early diagnosis and early treatment. The lungs are divided into five lobes: two on the left and three on the right. The lobes of the lungs are divided according to the direction of the blood vessels in the lungs. The lobes of the lungs are of great guiding significance for the diagnosis and treatment of lung cancer, and also have very important clinical value for the resection of cancerous sites and the planning of surgery in the later stage.

[0003] With the development of artificial intelligence technology, existing lung lobulation methods rely on a large number of manually outlined gold standards as training data. Please refer to the following existing literature:

[0004] [1] Yan Kai. Lung lobulation segmentation based on deep learning[D]. Xiamen University, 2019;

[0005] [2] Wu Yajun, Wu Yuanjiang, Huang Yansong, Guo Liyun. Automatic and semi-automatic lung segmentation method and system based on bronchus[P];

[0006] The disadvantages of this method are that training data requires a lot of manpower and material resources, and the performance of the model is correlated with the amount of data used for model training; at the same time, the trained model needs to be equipped with high-configuration computer hardware to run; and the cost of modifying and iterating the trained model is high. In addition, the traditional lung lobing method requires first obtaining the bronchial model and then extracting the centerline. These two operations require a lot of preparatory work, a large workload and manpower investment. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method and system for automatic lobing of lung organ models based on CT images, which can reduce the interference of non-lung areas on the extraction of lung interstitial point clouds, improve computing efficiency, and do not require manual intervention, thereby significantly reducing the workload.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions.

[0009] A method for automatic segmentation of lung organ models based on CT images comprises the following steps: step S1, importing original data ImageData of chest sequence CT images, and displaying them in three orthogonal views of cross section, sagittal plane and coronal plane; step S2, extracting lung parenchyma area by using threshold segmentation algorithm, obtaining binary image of lung parenchyma and recording it as MASK1, and then using the binary image of lung parenchyma as input, performing three-dimensional reconstruction on the lung parenchyma area segmented by detection, obtaining surface model of nodule and recording it as S; step S3, calculating the maximum bounding box of lung parenchyma according to whether there is binary point at the outermost boundary of binary image of lung parenchyma, extracting original data ImageData from the maximum bounding box, and using Hessian matrix enhancement algorithm to reconstruct the surface model of nodule. The method is used to enhance the sheet structure and obtain the enhanced image data MASK2. The voxels with a threshold value higher than 100 in the image data MASK2 are threshold segmented to obtain the binary image MASK3. The most connected operation is performed to eliminate fragmentary non-sheet structure pixels. Then, according to the left and right position relationship of the original data ImageData, the set of all points of the left binary data is extracted and recorded as the left lung point cloud S1, and the set of all points of the right binary data is extracted and recorded as the right lung point cloud S2; step S4, using the left lung point cloud S1 obtained in the step S3 as input data to perform left lung cutting and segmentation; step S5, using the right lung point cloud S2 obtained in the step S3 as input data to perform right lung primary cutting and right lung secondary cutting in sequence.

[0010] Preferably, in step S1, chest sequence CT images conforming to the DICOM 3.0 standard are used as the original data ImageData.

[0011] Preferably, the threshold range of the threshold segmentation algorithm used in step S2 is set to -1024 to -300. After obtaining the binary image of the lung parenchyma, the FlyingEdge three-dimensional reconstruction method is used to perform three-dimensional reconstruction on the detected and segmented lung parenchyma area.

[0012] Preferably, in step S3, after extracting the original data ImageData, the voxels with a value of 0 are set to -300 according to the voxel values ​​of the binary image.

[0013] Preferably, in step S3, the threshold value range of the image data MASK2 is 0-1024.

[0014] Preferably, in step S3, when performing the most connected operation, two connected sheet structures with the largest number of pixels are retained.

[0015] Preferably, in step S4, the left lung cutting and segmenting process includes: step S40, constructing a grid body: calculating the data bounding box of the left lung point cloud S1, obtaining a bounding box B1 according to the minimum and maximum values ​​in the three directions of the x, y and z axes, dividing the space in the space of the bounding box B1 in the direction of the three coordinate axes, and establishing a grid body G1 with a step length of s, wherein the grid body G1 contains a plurality of cubes of the same volume; step S41, point cloud division: for each point in the left lung point cloud S1, divide it according to the three-dimensional coordinate position of the point Assign to the cubes at the corresponding positions in the grid body G1, so that each cube of the grid body G1 contains 0, 1 or more points; step S42, point cloud filtering: set the maximum connection distance L, select any cube containing the point to start neighborhood search, if the cube with a radius of L in the three coordinate axis directions of a cube does not contain vertices, then stop searching, and record the vertex set obtained by this search as Ca; repeat the above operation for the remaining cubes in the grid body G1 to obtain N mutually unconnected vertex sets: Ca, Cb...CN, take the vertex set with the largest number of vertices in the above vertex sets as set C1, and perform step S43 as input data; step S43, interface extraction: in the grid body G1, clear the vertex values ​​in all cubes except the cube where the vertex in the set C1 is located, for the grid body G1, scan each cube in the XOZ plane from bottom to top along the Y axis direction, retain only the cube with the smallest Y value, clear the vertex values ​​in the remaining cubes, and then scan each cube in the XOY plane along the Z axis direction. Scan the cubes from bottom to top, and only keep the cube with the smallest Z value, clear the vertex values ​​in the remaining cubes, and record the remaining vertex set as D1; ​​step S44, cutting surface fitting: for the vertex set D1, use the bounding box of the left lung mesh model as the boundary to fit and generate the cutting surface P1; step S45, cutting: use the cutting surface P1 to cut the left lung mesh model to obtain two left lung lobe models, calculate the geometric center points of the two left lung lobe models respectively, and record the model with the higher z coordinate value as the left upper lung model, and the other one as the left lower lung model.

[0016] Preferably, in step S5, the process of performing the first-level cutting of the right lung includes: step S50, interface point cloud acquisition: taking the right lung point cloud S2 as input, according to the methods of steps S40 to S43, respectively obtain the maximum connected point set C2 and the vertex set D2 of the first-level interface of the right lung, remove the vertices simultaneously included in the vertex set D2 from the point set C2, and obtain the remaining vertex set D3; step S51, first-level cutting surface fitting: for the vertex set D2, with the bounding box of the right lung mesh model as the boundary, fit and generate the cutting surface P2; step S52, cutting: use the cutting surface P2 to cut the right lung mesh model to obtain two right lung lobe models, respectively calculate the geometric center points of the two right lung lobe models, and record the one with the higher z coordinate value as the right upper lobe model, and the other as the right lower lobe model.

[0017] Preferably, in step S5, the process of performing secondary cutting of the right lung includes: step S53, secondary cutting surface fitting: for the vertex set D3, the mesh bounding box of the right upper lobe model is used as the boundary, and the cutting surface P3 is fitted; step S54, cutting: using the cutting surface P3, the mesh of the right upper lobe model is cut, and the cut models are respectively the right upper lobe model and the right middle lobe model, and the geometric center points of the right upper lobe model and the right middle lobe model are respectively calculated, the one with the higher z coordinate value is the right upper lobe model, and the other one is the right middle lobe model.

[0018] A CT image-based automatic lobulation system for a lung organ model, the system being used to execute the above-mentioned automatic lobulation method for a lung organ model.

[0019] In the method and system for automatic lobing of lung organ model based on CT images disclosed in the present invention, after the original data ImageData of chest sequence CT images is imported into the system, the system directly extracts the interface point cloud from the CT image according to the image features of the lung interspace, and then extracts the first-level and second-level interfaces of the left lung and the right lung respectively through noise reduction, clustering and other operations, and cuts the left lung and the right lung respectively, so as to achieve the purpose of automatic lobing of the lung. In the specific lung lobing process, the lung parenchyma area is first segmented, which can reduce the interference of the non-lung area on the extraction of the lung interspace point cloud, thereby improving the operation efficiency; at the same time, the present invention is based on lung CT data for calculation, combined with the morphological features of the lung interspace, without manual intervention, and can effectively reduce the workload; furthermore, the present invention is based on the point cloud processing and grid cutting method for automatic lobing, and the intermediate data generated can be used for subsequent manual adjustment, so it is more flexible; in addition, the present invention automatically names the lobes according to the geometric model features after the lobing process, which greatly reduces the manual operation and better meets the application requirements of intelligent processing of CT medical images. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1The chest plain scan CT images are in compliance with the DICOM 3.0 standard;

[0021] Figure 2 This is the lung parenchyma segmentation result image;

[0022] Figure 3 The lung model image is obtained by three-dimensionally reconstructing the lung parenchyma binary image;

[0023] Figure 4 The original point cloud images of the left lung and the right lung extracted from the image data;

[0024] Figure 5 Graphical representation of the results of boundary point cloud filtering and interface generation;

[0025] Figure 6 This is a diagram of the lung lobes.

[0026] Figure 7 This is a flow chart of the automatic lobing method of the lung organ model based on CT images of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be described in more detail below with reference to the accompanying drawings and embodiments.

[0028] The present invention discloses a method for automatic segmentation of lung organ models based on CT images, see Figure 7 , which includes the following steps:

[0029] Step S1, importing the original chest CT image data ImageData, and displaying it in three orthogonal views: transverse, sagittal and coronal;

[0030] Step S2, using a threshold segmentation algorithm to extract the lung parenchyma area, obtaining a binary image of the lung parenchyma and recording it as MASK1, and then using the binary image of the lung parenchyma as input, performing three-dimensional reconstruction on the lung parenchyma area detected and segmented, obtaining a surface model of the nodule and recording it as S;

[0031] Step S3, calculate the maximum bounding box of the lung parenchyma according to whether there are binary points on the outermost boundary of the binary image of the lung parenchyma, extract the original data ImageData from the maximum bounding box, use the Hessian matrix enhancement algorithm to perform enhancement calculation on the sheet structure to obtain the enhanced image data MASK2, perform threshold segmentation on the voxels with a threshold value higher than 100 in the image data MASK2 to obtain the binary image MASK3, perform the most connected operation to eliminate fragmentary non-sheet structure pixels, and then extract the set of all points of the left binary data according to the left and right position relationship of the original data ImageData, record it as the left lung point cloud S1, and extract the set of all points of the right binary data, record it as the right lung point cloud S2;

[0032] Step S4, using the left lung point cloud S1 obtained in step S3 as input data to perform left lung segmentation;

[0033] Step S5, using the right lung point cloud S2 obtained in step S3 as input data, perform right lung primary cropping and right lung secondary cropping in sequence.

[0034] In the above method, after the original data ImageData of the chest sequence CT image is imported into the system, the system directly extracts the interface point cloud from the CT image based on the image features of the lung interspace, and then extracts the first-level and second-level interfaces of the left lung and the right lung respectively through noise reduction, clustering and other operations, and cuts the left lung and the right lung respectively, so as to achieve the purpose of automatic lung lobing. In the specific lung lobing process, the lung parenchyma area is first segmented, which can reduce the interference of the non-lung area on the extraction of the lung interspace point cloud, thereby improving the operation efficiency; at the same time, the present invention is based on lung CT data for calculation, combined with the morphological features of the lung interspace, without manual intervention, which can effectively reduce the workload; furthermore, the present invention is based on the point cloud processing and grid cutting method for automatic lobing, and the intermediate data generated can be used for subsequent manual adjustment, so it is more flexible; in addition, the present invention automatically names the lobes according to the geometric model features after the lobing process, which greatly reduces the manual operation and better meets the application requirements of intelligent processing of CT medical images.

[0035] Regarding the specific processing means of the above method, in step S1, the chest sequence CT image that conforms to the DICOM 3.0 standard is used as the original data ImageData. Among them, the chest plain scan CT image that conforms to the DICOM 3.0 standard is loaded and displayed. The imported image data can be viewed from three directions: the cross-sectional plane, the sagittal plane and the tubular plane. For the imported image, please refer to Figure 1 .

[0036] Furthermore, in step S2, the threshold range of the threshold segmentation algorithm is set to -1024 to -300. After obtaining the binary image of the lung parenchyma, the FlyingEdge 3D reconstruction method is used to perform 3D reconstruction on the detected and segmented lung parenchyma area. The mesh models of the left and right lungs obtained based on the FlyingEdge 3D reconstruction algorithm can be found in Figure 3 shown.

[0037] The specific processing means for step S3 include:

[0038] In the step S3, after extracting the original data ImageData, the voxels with a value of 0 are set to -300 according to the voxel values ​​of the binary image;

[0039] In the step S3, the threshold value range of the image data MASK2 is 0 to 1024;

[0040] In step S3, when performing the most connected operation, the two connected sheet structures with the largest number of pixels are retained.

[0041] In order to achieve accurate left lung cutting and segmentation, in step S4 of this embodiment, the left lung cutting and segmentation process includes:

[0042] Step S40, constructing a mesh body: calculating the data bounding box of the left lung point cloud S1, obtaining a bounding box B1 according to the minimum and maximum values ​​in the three directions of the x, y, and z axes, dividing the space in the space of the bounding box B1 in the directions of the three coordinate axes, and establishing a mesh body G1 with a step length of s, wherein the mesh body G1 contains a plurality of cubes of the same volume;

[0043] Step S41, point cloud division: for each point in the left lung point cloud S1, according to the three-dimensional coordinate position of the point, assign it to a cube at a corresponding position in the grid body G1, so that each cube in the grid body G1 contains 0, 1 or more points;

[0044] Step S42, point cloud filtering: set the maximum connection distance L, select any cube containing the point to start neighborhood search, if the cube with a radius of L in the three coordinate axis directions of a cube does not contain vertices, stop searching, and record the vertex set obtained by this search as Ca; repeat the above operation for the remaining cubes in the grid body G1 to obtain N unconnected vertex sets: Ca, Cb...CN, take the vertex set with the largest number of vertices in the above vertex sets as set C1, and use it as input data to execute step S43;

[0045] Step S43, interface extraction: in the grid body G1, clear the vertex values ​​in all cubes except the cube where the vertex in the set C1 is located, for the grid body G1, scan each cube in the XOZ plane from bottom to top along the Y axis direction, retain only the cube with the smallest Y value, clear the vertex values ​​in the remaining cubes, then scan each cube in the XOY plane from bottom to top along the Z axis direction, retain only the cube with the smallest Z value, clear the vertex values ​​in the remaining cubes, and record the remaining vertex set as D1;

[0046] Step S44, cutting surface fitting: for the vertex set D1, with the bounding box of the left lung mesh model as the boundary, fitting is performed to generate a cutting surface P1;

[0047] Step S45, cutting: Use the cutting plane P1 to cut the left lung mesh model to obtain two left lung lobe models, calculate the geometric center points of the two left lung lobe models respectively, and record the model with a higher z coordinate value as the left upper lung lobe model, and the other one as the left lower lung lobe model.

[0048] Based on the left lung segmentation, referring to the steps of constructing a mesh, dividing a point cloud, filtering a point cloud, and extracting an interface, in step S5, the process of performing the first-level segmentation of the right lung includes:

[0049] Step S50, interface point cloud acquisition: using the right lung point cloud S2 as input, according to the method of steps S40 to S43, respectively obtain the maximum connected point set C2 and the vertex set D2 of the first-level interface of the right lung, remove the vertices simultaneously included in the vertex set D2 from the point set C2, and obtain the remaining vertex set D3;

[0050] Step S51, first-level cutting surface fitting: for the vertex set D2, with the bounding box of the right lung mesh model as the boundary, fitting is performed to generate a cutting surface P2;

[0051] Step S52, cutting: Use the cutting plane P2 to cut the right lung mesh model to obtain two right lung lobe models, calculate the geometric center points of the two right lung lobe models respectively, record the one with a higher z coordinate value as the right upper lung lobe model, and the other one as the right lower lung lobe model.

[0052] Furthermore, in step S5, the process of performing the second-level cutting of the right lung includes:

[0053] Step S53, secondary cutting surface fitting: for the vertex set D3, with the mesh bounding box of the right upper lobe model as the boundary, fitting is performed to generate a cutting surface P3;

[0054] Step S54, cutting: using the cutting plane P3, the mesh of the right upper lobe model is cut, and the models obtained by cutting are respectively the right upper lobe model and the right middle lobe model. The geometric center points of the right upper lobe model and the right middle lobe model are calculated respectively. The one with the higher z coordinate value is the right upper lobe model, and the other one is the right middle lobe model.

[0055] In practical applications, the present invention can implement the above-mentioned automatic lobing method of the lung organ model through a lung organ model automatic lobing system based on CT images.

[0056] The present invention discloses a method for automatic lobing of a lung organ model based on CT images, which addresses the defects of existing machine learning and deep learning methods for lobing lung models, namely, the need to rely on a large amount of gold standard data for training, and the problem that traditional lung lobing methods rely on a series of complex pre-work such as bronchial segmentation and centerline extraction. A method based on medical image processing and grid processing is provided, and technical means such as threshold segmentation, surface fitting, and grid cutting are combined. The following embodiments can be referred to in practical applications.

[0057] Embodiment 1

[0058] The automatic lobulation method of the lung organ model in this embodiment includes the following steps:

[0059] Step 1, data preparation: import chest CT images that conform to the DICOM 3.0 standard into the program and display them in three orthogonal views: transverse, sagittal, and coronal. If the acquired data contains multiple stages, you need to stage the data first, and then select the appropriate stage data for processing. Define the input raw data as ImageData; see the imported image Figure 1 As shown;

[0060] Step 2, lung parenchyma segmentation: In order to eliminate the influence of other image areas on the lung lobing process, after importing the CT image data, the threshold segmentation algorithm (threshold range -1024 to -300) is used to extract the lung parenchyma area, and the binary image of the lung parenchyma is obtained, which is recorded as MASK1. Then, the binary image of the lung parenchyma is used as input, and the FlyingEdge 3D reconstruction method is used to perform 3D reconstruction on the lung parenchyma area detected and segmented to obtain the surface model of the nodule, which is recorded as S. The results of lung parenchyma segmentation can be found in Figure 2 As shown in the figure, the mesh models of the left and right lungs obtained based on the FlyingEdge 3D reconstruction algorithm can be found in Figure 3 As shown;

[0061] Step 3, interface point cloud extraction: Calculate the maximum bounding box of the lung parenchyma according to whether there are binary points on the outermost boundary of the acquired binary image, extract the data of the original image ImageData in this bounding box, further set the voxels with a value of 0 to -300 according to the value of the voxels of the binary image to eliminate the influence of these voxels on the subsequent calculations, use the Hessian matrix enhancement algorithm to enhance the sheet structure and obtain the enhanced image data MASK2 (threshold range is 0-1024), perform threshold segmentation on the voxels with a threshold higher than 100 for the MASK2 data, obtain the binary image MASK3, perform the most connected operation (the two connected sheet structures with the largest pixel points are retained), eliminate the fragmentary non-sheet structure pixels, and then extract the set of all points of the left binary data according to the left and right positional relationship of the ImageData image, record it as the point cloud S1 of the left lung, and extract the set of all points of the right binary data as the point cloud S2 of the right lung; for the original point cloud diagrams of the left and right lungs extracted from the image data, please refer to Figure 4 ;

[0062] Step 4, left lung segmentation: This step uses the point cloud S1 belonging to the left lung obtained in step 3 as input data, and specifically includes the following process:

[0063] 4.1 Constructing a mesh volume: For the input point cloud data S1, find the data bounding box of S1, that is, the bounding box B1 formed by the minimum and maximum values ​​in the three directions of the x, y, and z axes. In the bounding box B1 space, divide the space in the directions of the three coordinate axes and establish a mesh volume G1 with a step size of s, which contains cubes of the same volume;

[0064] 4.2 Point cloud partitioning: Assign each point in the point cloud data S1 to the corresponding cube of the grid body G1 according to its three-dimensional coordinate position. After this step is completed, each cube of the grid body G1 contains 0, 1 or more points;

[0065] 4.3 Point cloud filtering: Set the maximum connection distance L, select any cube containing a point to start neighborhood search. If no cube with a radius of L in the three coordinate axis directions of a cube contains vertices, stop searching. The vertex set obtained by this search is Ca. Repeat the above operation for the remaining cubes of G1, and you will get N unconnected vertex sets: Ca, Cb...CN. Take the vertex set with the largest number of vertices in the above vertex sets as set C1, and pass it as input data to the next operation;

[0066] 4.4 Interface extraction: In the mesh body G1, clear the vertex values ​​of all cubes except the cubes containing the vertices in the point set C1 obtained in step ③. For the mesh body G1, scan each cube in the XOZ plane from bottom to top along the Y-axis direction, retain only the cube with the smallest Y value, and discard the rest of the cubes (clear the vertices in them). Then scan each cube in the XOY plane from bottom to top along the Z-axis direction, retain only the cube with the smallest Z value, and discard the rest of the cubes. The remaining vertex set is D1;

[0067] 4.5 Cutting surface fitting: For the vertex set D1, the cutting surface P1 is fitted with the bounding box of the left lung mesh model as the boundary;

[0068] 4.6 Cutting: Use the cutting plane P1 to cut the left lung mesh model to obtain the left upper lobe and left lower lobe models. Calculate the geometric center points of the two models respectively. The one with the higher z coordinate value is the left upper lobe model, and the other is the left lower lobe model. For the results of boundary point cloud filtering and interface generation, please refer to Figure 5 As shown;

[0069] Step 5, right lung primary cutting: This step uses the right lung interface point cloud data S2 obtained in step 3 as input. Because the right lung is divided into three sections, it is necessary to first extract the primary cutting surface, and then extract the secondary cutting surface, which specifically includes the following steps:

[0070] 5.1 Obtaining the interface point cloud. Take the original point cloud data S2 of the right lung as input, refer to steps 4.1-4.4 in step 4, obtain the maximum connected point set C2 and the vertex set of the interface D2, remove the vertices in the point set C2 that are also included in the point set D2, and obtain the point set D3;

[0071] 5.2 First-level cutting surface fitting: For the vertex set D2, the cutting surface P2 is generated by fitting with the bounding box of the right lung mesh model as the boundary;

[0072] 5.3 Cutting: Use cutting plane P2 to cut the right lung mesh model. The cut model is the upper part of the right lung and the lower lobe of the right lung. Calculate the geometric center points of the two meshes respectively. The one with the higher z coordinate value is the upper part of the right lung, and the other one is the lower lobe of the right lung.

[0073] Step 6, right lung secondary cropping: Using the point set D2 obtained in step 5 as input, perform the following operations:

[0074] 6.1 Secondary cutting surface fitting: For the vertex set D3, the cutting surface P3 is generated by fitting with the bounding box of the upper part of the right lung mesh obtained by cutting in step (5) as the boundary;

[0075] 6.2 Cutting: Use cutting plane P3 to cut the mesh of the upper part of the right lung. The cut model is the right upper lobe and the right middle lobe. Calculate the geometric center points of the two meshes respectively. The one with the higher z coordinate value is the right upper lobe, and the other one is the right lower lobe. For the results of lung lobes, please refer to Figure 6 shown.

[0076] By executing steps 1 to 6 above, this embodiment realizes automatic lobing of left and right lung mesh models based on CT images, and obtains five models of left upper lobe, left lower lobe, right upper lobe, right middle lobe and right lower lobe through intelligent and automated analysis and processing, accurately completing the automatic lobing of lung organ models, providing a technical basis for computer-aided medical diagnosis and treatment and 3D visualization of medical images, and is particularly suitable for application in the field of CT medical image segmentation and three-dimensional reconstruction technology, and has good application prospects.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the technical scope of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for automatic segmentation of lung organ models based on CT images, characterized in that: The steps include: Step S1, importing the original chest CT image data ImageData, and displaying it in three orthogonal views: transverse, sagittal and coronal; Step S2, using a threshold segmentation algorithm to extract the lung parenchyma area, obtaining a binary image of the lung parenchyma and recording it as MASK1, and then using the binary image of the lung parenchyma as input, performing three-dimensional reconstruction on the lung parenchyma area detected and segmented, obtaining a surface model of the nodule and recording it as S; Step S3, calculate the maximum bounding box of the lung parenchyma according to whether there are binary points on the outermost boundary of the binary image of the lung parenchyma, extract the original data ImageData from the maximum bounding box, use the Hessian matrix enhancement algorithm to perform enhancement calculation on the sheet structure to obtain the enhanced image data MASK2, perform threshold segmentation on the voxels with a threshold value higher than 100 in the image data MASK2 to obtain the binary image MASK3, perform the most connected operation to eliminate fragmentary non-sheet structure pixels, and then extract the set of all points of the left binary data according to the left and right position relationship of the original data ImageData, record it as the left lung point cloud S1, and extract the set of all points of the right binary data, record it as the right lung point cloud S2; Step S4, using the left lung point cloud S1 obtained in step S3 as input data to perform left lung segmentation; Step S5, using the right lung point cloud S2 obtained in step S3 as input data, sequentially performing right lung primary cropping and right lung secondary cropping; In step S4, the left lung cutting and segmenting process includes: Step S40, constructing a mesh body: calculating the data bounding box of the left lung point cloud S1, obtaining a bounding box B1 according to the minimum and maximum values ​​in the three directions of the x, y, and z axes, dividing the space in the space of the bounding box B1 in the directions of the three coordinate axes, and establishing a mesh body G1 with a step length of s, wherein the mesh body G1 contains a plurality of cubes of the same volume; Step S41, point cloud division: for each point in the left lung point cloud S1, according to the three-dimensional coordinate position of the point, assign it to a cube at a corresponding position in the grid body G1, so that each cube in the grid body G1 contains 0, 1 or more points; Step S42, point cloud filtering: set the maximum connection distance L, select any cube containing the point to start neighborhood search, if the cube with a radius of L in the three coordinate axis directions of a cube does not contain vertices, stop searching, and record the vertex set obtained by this search as Ca; repeat the above operation for the remaining cubes in the grid body G1 to obtain N unconnected vertex sets: Ca, Cb...CN, take the vertex set with the largest number of vertices in the above vertex sets as set C1, and use it as input data to execute step S43; Step S43, interface extraction: in the grid body G1, clear the vertex values ​​in all cubes except the cube where the vertex in the set C1 is located, for the grid body G1, scan each cube in the XOZ plane from bottom to top along the Y axis direction, retain only the cube with the smallest Y value, clear the vertex values ​​in the remaining cubes, then scan each cube in the XOY plane from bottom to top along the Z axis direction, retain only the cube with the smallest Z value, clear the vertex values ​​in the remaining cubes, and record the remaining vertex set as D1; Step S44, cutting surface fitting: for the vertex set D1, with the bounding box of the left lung mesh model as the boundary, fitting is performed to generate a cutting surface P1; Step S45, cutting: Use the cutting plane P1 to cut the left lung mesh model to obtain two left lung lobe models, calculate the geometric center points of the two left lung lobe models respectively, and record the model with a higher z coordinate value as the left upper lung lobe model, and the other one as the left lower lung lobe model.

2. The method for automatic segmentation of lung organ models based on CT images according to claim 1, characterized in that: In the step S1, a chest CT image sequence conforming to the DICOM 3.0 standard is used as the original data ImageData.

3. The method for automatic segmentation of lung organ models based on CT images according to claim 1, characterized in that: The threshold range of the threshold segmentation algorithm used in step S2 is set to -1024 to -300. After obtaining the binary image of the lung parenchyma, the FlyingEdge three-dimensional reconstruction method is used to perform three-dimensional reconstruction on the detected and segmented lung parenchyma area.

4. The method for automatic segmentation of lung organ models based on CT images according to claim 1, characterized in that: In the step S3, after extracting the original data ImageData, the voxels with a value of 0 are set to -300 according to the voxel values ​​of the binary image.

5. The method for automatic segmentation of lung organ models based on CT images according to claim 1, characterized in that: In the step S3, the threshold value range of the image data MASK2 is 0-1024.

6. The method for automatic segmentation of lung organ models based on CT images according to claim 1, characterized in that: In step S3, when performing the most connected operation, the two connected sheet structures with the largest number of pixels are retained.

7. The method for automatic segmentation of lung organ models based on CT images according to claim 1, characterized in that: In step S5, the process of performing the first-level cutting of the right lung includes: Step S50, interface point cloud acquisition: using the right lung point cloud S2 as input, according to the method of steps S40 to S43, respectively obtain the maximum connected point set C2 and the vertex set D2 of the first-level interface of the right lung, remove the vertices simultaneously included in the vertex set D2 from the point set C2, and obtain the remaining vertex set D3; Step S51, first-level cutting surface fitting: for the vertex set D2, with the bounding box of the right lung mesh model as the boundary, fitting is performed to generate a cutting surface P2; Step S52, cutting: Use the cutting plane P2 to cut the right lung mesh model to obtain two right lung lobe models, calculate the geometric center points of the two right lung lobe models respectively, record the one with a higher z coordinate value as the right upper lung lobe model, and the other one as the right lower lung lobe model.

8. The method for automatic segmentation of lung organ models based on CT images according to claim 7, characterized in that: In step S5, the process of performing the second-level cutting of the right lung includes: Step S53, secondary cutting surface fitting: for the vertex set D3, with the mesh bounding box of the right upper lobe model as the boundary, fitting is performed to generate a cutting surface P3; Step S54, cutting: using the cutting plane P3, the mesh of the right upper lobe model is cut, and the models obtained by cutting are respectively the right upper lobe model and the right middle lobe model. The geometric center points of the right upper lobe model and the right middle lobe model are calculated respectively. The one with the higher z coordinate value is the right upper lobe model, and the other one is the right middle lobe model.

9. A CT image-based lung organ model automatic lobulation system, characterized in that: The system is used to execute the automatic lobulation method of a lung organ model according to any one of claims 1 to 8.

Images