Difference set derived 2D mediastinal image lymph node partition segmentation method

Abstract

The application discloses a 2D mediastinal image lymph node partition segmentation method based on difference set derivation, belongs to the medical image segmentation problem, is usually used for computer-aided diagnosis, and relates to the fields of digital medical treatment and artificial intelligence and the like. The application is proposed to solve the problems of mediastinal lymph node partition positioning, identification and classification. Technical points: the mediastinal region in a lung CT image comprises a mediastinal organ and a lymph node region, the method first segments out the mediastinal region and the mediastinal organ by using a segmentation model, and then obtains the lymph node region by difference set operation. Then, the sub-regions are divided by using a shortest path planning method of the mediastinal organ, and finally, the sub-region identification is realized by using a sub-region edge classification model, so that the lymph node partition result is obtained. The application is suitable for the process of automatically positioning and identifying the mediastinal lymph node partition.

Description

Technical Field

[0001] This invention relates to the field of computer-aided diagnostic image segmentation, specifically to a method for segmenting lymph nodes in 2D mediastinal images based on difference set derivation. Technical Background

[0002] In different stages of lung cancer, metastasis and deterioration can occur in different mediastinal lymph nodes, forming malignant mediastinal lymph nodes. Therefore, determining the location of malignant mediastinal lymph nodes, i.e., mediastinal lymph node division, is of great clinical significance.

[0003] The international TNM staging system for lung cancer divides the mediastinal lymph nodes into 14 regions. Current techniques primarily rely on physicians reviewing medical images to determine the location of malignant mediastinal lymph nodes. However, this method is time-consuming and its accuracy is limited by the physician's expertise. Furthermore, the morphology of different lymph node regions varies significantly across different CT mediastinal image layers, with different morphologies existing within the same lymph node region. Moreover, there are no clear boundaries between different lymph node regions, leading to overlap or blurred morphology. Therefore, this paper proposes a difference-based 2D mediastinal image lymph node segmentation method. This method segments the clearly identifiable mediastinal region in the 2D mediastinal image as the complete set of pixels and the mediastinal organs as a subset of pixels. Subtraction between pixels yields the lymph node regions. An undirected graph is constructed to generate the shortest paths between mediastinal organs and from mediastinal organs to the edges of mediastinal regions. Finally, a region growing method and a sub-region edge classification model are used to obtain the mediastinal lymph node segmentation results, effectively solving the problems of difficult localization, identification, and classification of mediastinal lymph node regions. Summary of the Invention

[0004] This invention proposes a 2D mediastinal image lymph node partitioning method derived from difference set derivation to solve the problems of difficulty in locating, identifying, and classifying mediastinal lymph nodes.

[0005] The aforementioned objectives are primarily achieved through the following technical solutions:

[0006] The mediastinal region in lung CT images includes mediastinal organs and lymph node regions. The method described first uses a segmentation model to segment the mediastinal region and mediastinal organs, then uses difference operations to obtain the lymph node regions. Next, a mediastinal organ shortest path planning method is used to divide the region into sub-regions. Finally, a sub-region edge classification model is used to identify the sub-regions and obtain the lymph node partitioning results. The method includes the following steps:

[0007] S1. Segmenting the mediastinal region and mediastinal organs: In any 2D mediastinal image, the mediastinal region R is segmented using different segmentation models. core and each type of mediastinal organ R(c i ), and obtain Rcore and R(c i Edge information E core and E info i ;

[0008] (1) Construct a longitudinal septum region segmentation model and train the model parameters using the longitudinal septum region segmentation dataset;

[0009] (2) Using a mediastinal region segmentation model, extract the mediastinal region R from the lung mediastinal image. core ;

[0010] (3) Using the edge detection algorithm, calculate R core Edge information E core We obtain the coordinate and category information of the edge points, which are represented as follows:

[0011] E core ={((x,y),c(x,y))|(x,y)∈R core} (1)

[0012] Where (x, y) represents the coordinate information of the edge point of the longitudinal diaphragm region, and c(x, y) = 0 represents the category information of the edge point of the longitudinal diaphragm region;

[0013] (4) Construct a mediastinal organ segmentation model and train it using a mediastinal organ segmentation dataset;

[0014] (5) Use the mediastinal organ segmentation model to segment each mediastinal organ R(c) i ), and use the edge detection algorithm in S13 to calculate R(c i The edge point information constitutes set E. info i :

[0015] E info i ={((x, y), c i (x, y))|(x, y)∈R(c i (2)

[0016] Where (x, y) represents the coordinate information of the edge point of the mediastinal organ, c i (x, y)∈{1, 2, ..., n} represents the number of the mediastinal organ category;

[0017] (6) R(c i The pixel value inside is set to c. i The value of (x, y) represents the organ that has been separated;

[0018] S2. Obtaining the lymph node region: The mediastinal region R... corePixels are represented as the complete set of pixels, and each type of mediastinal organ R(c i If each pixel is a subset of pixels, then the lymph node region D can be derived using the set relationships between pixels, as expressed by:

[0019]

[0020] Equation (3) represents a difference operation method. It is the union of the sets of all mediastinal organ pixels. R represents core The unsegmented pixel region, i.e., the lymph node region;

[0021] S3. Divide the lymph node regions: Obtain the shortest path d between mediastinal organs. ij And the shortest path d from the mediastinal organs to the edge of the mediastinal region i0 The lymph node region is divided using the shortest path method, generating multiple connected lymph node sub-regions. The specific calculation method is as follows:

[0022] (1) Calculate the mediastinal organ R(c) i The centroid coordinates P(c) i The calculation formula is:

[0023]

[0024] Where, |R(c i )| represents the total number of pixels for each type of mediastinal organ.

[0025] (2) Construction of an undirected graph of organs: Construct an undirected weighted graph G = (V, E), where V represents the graph nodes, i.e., mediastinal organs, and E is the edge connecting two mediastinal organs, with the weight of the edge being the distance d(c) between the centroids of the two mediastinal organs. i c j The calculation formula is:

[0026]

[0027] Each line represents d(c) i c j ) path d ij The path is plotted in the longitudinal diaphragm region image;

[0028] (3) Calculate the shortest distance d(c) from each mediastinal organ to the edge of the mediastinal region. i ,c0), and will represent d(c i The path d of c0) i0 Plotted in the diaphragmatic region image:

[0029] d(c i c0)=min||P(ci )-m|| (6)

[0030] Wherein, P(c i Let P(c) represent the centroid coordinates of the i-th organ, m represent any point on the edge of the mediastinal region, and ||P(c) represent the centroid coordinates of the i-th organ. i )-m|| represents the Euclidean distance from the centroid of the mediastinal organ to point m;

[0031] (4) Delete the shortest path through the mediastinal organ: For any shortest path, if the coordinate point on it is contained inside the segmented mediastinal organ, then delete the path.

[0032] S4. Divide lymph node sub-regions: using d ij d i0 The edges of the mediastinal organs and the mediastinal region serve as the boundary lines of the lymph node region D. Within D, a region generation method is used to generate sub-regions of lymph nodes D. i The specific calculation method is as follows:

[0033] (1) From d ij d i0 The boundaries of mediastinal organs and mediastinal regions constitute the boundary set B. all The partition of D is determined by the total boundary set B. all Sure:

[0034]

[0035] D i Indicates that by B all The divided lymph node subregions, D i The regions are connected and have boundaries. Belongs to B all ;

[0036] (2) Select a seed point p from within each connected lymph node subregion. i The selection rules are as follows:

[0037]

[0038] Seed point is located at D i Inside and not in B all superior;

[0039] (3) Using seed point p i Starting from the region, the connected region D is generated using the region growing method. i until boundary B all until;

[0040] (4) D i The internal pixel value is set to c p (x, y):

[0041] c p (x, y) ≠ 0, c i (x, y), c j (x, y) (9)

[0042] c i (x, y) represents R(c i ) Internal and edge pixel values, c j (x, y) represents the already segmented D. j (j < i) are the pixel values ​​inside;

[0043] S5. Identify lymph node regions: Analyze each D i Edge components, using sub-region edge classification models for D i Classify them, each category D i The identification result is the lymph node partitioning result of the 2D mediastinal image.

[0044] Invention Effects

[0045] This invention proposes a 2D mediastinal image lymph node segmentation method derived from difference set derivation, used for automatic localization and identification of mediastinal lymph node regions. The mediastinal region in lung CT images includes mediastinal organs and lymph node regions. The method first uses a segmentation model to segment the mediastinal region and mediastinal organs, then obtains the lymph node regions through difference set operations. Next, it uses a mediastinal organ shortest path planning method to divide sub-regions, and finally uses a sub-region edge classification model to identify the sub-regions, obtaining the lymph node partitioning results.

[0046] This invention proposes a 2D mediastinal image lymph node partitioning method based on difference set derivation. The main idea is to indirectly deduce lymph node regions by dividing the entire set of pixels and subsets of pixels. By dividing the lymph node region and segmenting the lymph node sub-regions, and using a sub-region edge classification model to achieve sub-region identification, the mediastinal lymph node partitioning results can be obtained, improving segmentation efficiency and detection accuracy, and effectively solving the problems of difficult localization, identification, and classification of mediastinal lymph node partitions. Attached Figure Description

[0047] Figure 1 This is a flowchart of the 2D mediastinal image lymph node partitioning method derived from difference set in an example of the present invention;

[0048] Figure 2 This is a flowchart illustrating the division of lymph node regions in an example of the present invention;

[0049] Figure 3 This is a flowchart illustrating the generation process of lymph node sub-regions in an example of the present invention. Detailed Implementation Plan Specific implementation method one:

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, a 2D mediastinal image lymph node segmentation method derived from difference set derivation is characterized by the fact that the mediastinal region in a lung CT image includes mediastinal organs and lymph node regions. The method first segments the mediastinal region and mediastinal organs using a segmentation model, then obtains the lymph node regions through difference set operations. Next, it uses a mediastinal organ shortest path planning method to divide sub-regions, and finally uses a sub-region edge classification model to identify the sub-regions, obtaining the lymph node segmentation results. The method includes the following steps:

[0053] S1. Segmenting the mediastinal region and mediastinal organs: In any 2D mediastinal image, the mediastinal region R is segmented using different segmentation models. core and each type of mediastinal organ R(c i ), and obtain R core and R(c i Edge information E core and E info i ;

[0054] S2. Obtaining the lymph node region: The mediastinal region R... core Pixels are represented as the complete set of pixels, and each type of mediastinal organ R(c i If each pixel is a subset of pixels, then the lymph node region D can be derived from the set relationship between pixels.

[0055] S3. Divide the lymph node regions: Obtain the shortest path d between mediastinal organs. ij And the shortest path d from the mediastinal organs to the edge of the mediastinal region i0 The shortest path is used to divide the lymph node region, generating multiple connected lymph node sub-regions, which are then used to divide the lymph node region.

[0056] S4. Divide lymph node sub-regions: using d ij d i0 The edges of the mediastinal organs and the mediastinal region serve as the boundary lines of the lymph node region D. Within D, a region generation method is used to generate sub-regions of lymph nodes D. i ;

[0057] S5. Identify lymph node regions: Analyze each Di Edge components, using sub-region edge classification models for D i Classify them, each category D i The identification result is the lymph node partitioning result of the 2D mediastinal image.

[0058] The embodiments of the present invention will be described in detail below:

[0059] The specific implementation of this invention is as follows.

[0060] S1. Segmenting the mediastinal region and mediastinal organs: In any 2D mediastinal image, the mediastinal region R is segmented using different segmentation models. core and each type of mediastinal organ R(c i ), and obtain R core and R(c i Edge information E core and E info i ;

[0061] (1) Construct the U-Net model for segmenting the core region of the longitudinal septum and train the model parameters using the longitudinal septum region segmentation dataset;

[0062] (2) Using the mediastinal core region segmentation model U-Net, the mediastinal core region R is extracted from the lung mediastinal image. core ;

[0063] (3) Using the Sobel edge detection algorithm, calculate R. core Edge information E core The coordinates and category information of the edge points are obtained, and represented as follows:

[0064] E core ={((x,y),c(x,y))|(x,y)∈R core} (1)

[0065] Where (x, y) represents the coordinate information of the edge point of the longitudinal diaphragm region, and c(x, y) = 0 represents the category information of the edge point of the longitudinal diaphragm region;

[0066] (4) Construct the TransUnet model for mediastinal organ segmentation and train it using the mediastinal organ segmentation dataset.

[0067] (5) Use the mediastinal organ segmentation model to segment each mediastinal organ R(c) i ), and use the edge detection algorithm in S13 to calculate R(c i The edge point information constitutes set E. info i :

[0068] Einfo i ={((x, y), c i (x, y))|(x, y)∈R(c i (2)

[0069] Where (x, y) represents the coordinate information of the edge point of the mediastinal organ, c i (x, y)∈{1,2,…,n} represents the number of the mediastinal organ category. In this example, n takes the value of 10, which means that 10 classic mediastinal organs are divided.

[0070] (6) R(c i The pixel value inside is set to c. i The value of (x, y) represents the segmented organ. For example, if organ R(c1) represents the set of pixels in the region where the left pulmonary artery is located, then the pixel values ​​in the region and edge of the left pulmonary artery are all set to 1.

[0071] S2. Obtaining the lymph node region: The mediastinal region R... core Pixels are represented as the complete set of pixels, and each type of mediastinal organ R(c i If each pixel is a subset of pixels, then the lymph node region D can be derived using the set relationships between pixels, as expressed by:

[0072]

[0073] Equation (3) represents a difference operation method. It is the union of the sets of all mediastinal organ pixels. R represents core The unsegmented pixel region, i.e., the lymph node region;

[0074] S3. Divide the lymph node regions: such as Figure 2 As shown, obtain the shortest path d between mediastinal organs. ij And the shortest path d from the mediastinal organs to the edge of the mediastinal region i0 The lymph node region is divided using the shortest path method, generating multiple connected lymph node sub-regions. The specific calculation method is as follows:

[0075] (1) Calculate the mediastinal organ R(c) i The centroid coordinates P(c) i The calculation formula is:

[0076]

[0077] Where, |R(c i )| represents the total number of pixels for each type of mediastinal organ.

[0078] (2) Construction of an undirected graph of organs: Construct an undirected weighted graph G = (V, E), where V represents the graph nodes, i.e., mediastinal organs, and E is the edge connecting two mediastinal organs, with the weight of the edge being the distance d(c) between the centroids of the two mediastinal organs. i c j The calculation formula is:

[0079]

[0080] Each line represents d(c) i c j ) path d ij The path is plotted in the longitudinal diaphragm region image;

[0081] (3) Calculate the shortest distance d(c) from each mediastinal organ to the edge of the mediastinal region. i ,c0), and will represent d(c i The path d of c0) i0 Plotted in the diaphragmatic region image:

[0082] d(c i c0)=min||P(c i )-m|| (6)

[0083] Wherein, P(c i Let P(c) represent the centroid coordinates of the i-th organ, m represent any point on the edge of the mediastinal region, and ||P(c) represent the centroid coordinates of the i-th organ. i )-m|| represents the Euclidean distance from the centroid of the mediastinal organ to point m;

[0084] (4) Delete the shortest path through the mediastinal organ: For any shortest path, if the coordinate point on it is contained inside the segmented mediastinal organ, then delete the path.

[0085] S4. Divide lymph node sub-regions: such as Figure 3 As shown, with d ij d i0 The edges of the mediastinal organs and the mediastinal region serve as the boundary lines of the lymph node region D. Within D, a region generation method is used to generate sub-regions of lymph nodes D. i The specific calculation method is as follows:

[0086] (1) From d ij d i0 The boundaries of mediastinal organs and mediastinal regions constitute the boundary set B. all The partition of D is determined by the total boundary set B. all Sure:

[0087]

[0088] D idenoted by B all the divided lymph node sub-regions, D i are connected and the region boundaries belong to B all ;

[0089] (2) Select a seed point p inside each connected lymph node sub-region i , and the selection rule is:

[0090]

[0091] The seed point is located inside D i and not on B all ;

[0092] (3) Taking the seed point p i as the starting point, generate the connected region D by using the region growing method i until the boundary B all is reached;

[0093] (4) Set the internal pixel value of D i as c p (x,y):

[0094] c p (x,y)≠0, c i (x,y), c j (x,y) (9)

[0095] c i (x,y) represents the pixel values inside and at the edge of R(c i ), c j (x,y) represents the pixel point values inside the already segmented D j (j<i);

[0096] S5. Identify lymph node stations: Analyze the edge components of each D i , classify D using the sub-region edge classification model ResNet i , and the recognition result of each type of D i is the lymph node station division result of the 2D mediastinal image, with a specific example as follows:

[0097] For the lymph node sub-region D i , its edges are composed of four categories: superior vena cava, aortic arch, trachea and mediastinal region edge, and the classification model ResNet identifies it as the mediastinal lymph node 4R station.

Claims

1. A method for segmenting lymph nodes in a 2D mediastinal image based on difference set derivation, characterized in that, The mediastinal region in lung CT images includes mediastinal organs and lymph node regions. The method described first uses a segmentation model to segment the mediastinal region and mediastinal organs, then uses difference operations to obtain the lymph node regions. Next, a mediastinal organ shortest path planning method is used to divide the region into sub-regions. Finally, a sub-region edge classification model is used to identify the sub-regions and obtain the lymph node partitioning results. The method includes the following steps: S1. Segmenting the mediastinal region and mediastinal organs: In any 2D mediastinal image, the mediastinal region R is segmented using different segmentation models. core and each type of mediastinal organ R(c i ), and obtain R core and R(c i Edge information E core and E info i ; S11. Construct a longitudinal septum region segmentation model and train the model parameters using the longitudinal septum region segmentation dataset; S12. Using a mediastinal region segmentation model, extract the mediastinal region R from the lung mediastinal image. core ; S13. Using an edge detection algorithm, calculate R. core Edge information E core We obtain the coordinate and category information of the edge points, which are represented as follows: AND core ={((x,y),c(x,y))|(x,y)∈R core } (1) Where (x, y) represents the coordinate information of the edge point of the longitudinal diaphragm region, and c(x, y) = 0 represents the category information of the edge point of the longitudinal diaphragm region; S14. Construct a mediastinal organ segmentation model and train it using a mediastinal organ segmentation dataset. S15. Use the mediastinal organ segmentation model to segment each mediastinal organ R(c) i ), and use the edge detection algorithm in S13 to calculate R(c i The edge point information constitutes set E. info i : E info i ={((x,y),c i (x,y))|(x,y)∈R(c i )} (2) Where (x, y) represents the coordinate information of the edge point of the mediastinal organ, c i (x, y)∈{1, 2, ..., n} represents the number of the mediastinal organ category; S16, R(c) i The pixel value inside is set to c. i The value of (x, y) represents the organ that has been separated; S2. Obtaining the lymph node region: The mediastinal region R... core Pixels are represented as the complete set of pixels, and each type of mediastinal organ R(c i If each pixel is a subset of pixels, then the lymph node region D can be derived using the set relationships between pixels, as expressed by: Equation (3) represents a difference operation method. It is the union of the sets of all mediastinal organ pixels. R represents core The unsegmented pixel region, i.e., the lymph node region; S3. Divide the lymph node regions: Obtain the shortest path d between mediastinal organs. ij And the shortest path d from the mediastinal organs to the edge of the mediastinal region i0 The lymph node region is divided using the shortest path method, generating multiple connected lymph node sub-regions. The specific calculation method is as follows: S31. Calculate the mediastinal organ R(c) i The centroid coordinates P(c) i The calculation formula is: Where, |R(c i )| represents the total number of pixels for each type of mediastinal organ. S32. Construction of an Undirected Graph of Organs: Construct an undirected weighted graph G = (V, E), where V represents a graph node, i.e., a mediastinal organ, and E is an edge connecting two mediastinal organs, with the weight of the edge being the distance d(c) between the centroids of the two mediastinal organs. i c j The calculation formula is: Each line represents d(c) i c j ) path d ij The path is plotted in the longitudinal diaphragm region image; S33. Calculate the shortest distance d(c) from each mediastinal organ to the edge of the mediastinal region. i ,c0), and will represent d(c i The path d of c0) i0 Plotted in the diaphragmatic region image: d(c i ,c0)=min||P(c i )-m|| (6) Wherein, P(c i Let P(c) represent the centroid coordinates of the i-th organ, m represent any point on the edge of the mediastinal region, and ||P(c) represent the centroid coordinates of the i-th organ. i )-m|| represents the Euclidean distance from the centroid of the mediastinal organ to point m; S34. Delete the shortest path through the mediastinal organ: For any shortest path, if the coordinates of its points are contained inside the segmented mediastinal organ, then delete the path. S4. Divide lymph node sub-regions: using d ij d i0 The edges of the mediastinal organs and the mediastinal region serve as the boundary lines of the lymph node region D. Within D, a region generation method is used to generate sub-regions of lymph nodes D. i The specific calculation method is as follows: S41, by d ij d i0 The boundaries of mediastinal organs and mediastinal regions constitute the boundary set B. all The partition of D is determined by the total boundary set B. all Sure: D i Indicates that by B all The divided lymph node subregions, D i The regions are connected and have boundaries. Belongs to B all ; S42. Select a seed point p from within each connected lymph node subregion. i The selection rules are as follows: Seed point is located at D i Inside and not in B all superior; S43, using seed point p i Starting from the region, the connected region D is generated using the region growing method. i until boundary B all until; S44, D i The internal pixel value is set to c p (x, y): c p (x,y)≠0,c i (x,y),c j (x,y) (9) c i (x, y) represents R(c i ) Internal and edge pixel values, c j (x, y) represents the already segmented D. j (j < i) are the pixel values ​​inside; S5. Identify lymph node regions: Analyze each D i Edge components, using sub-region edge classification models for D i Classify them, each category D i The identification result is the lymph node partitioning result of the 2D mediastinal image.

Images