Three-dimensional medical image segmentation processing method and system based on deep learning

Abstract

The invention provides a three-dimensional medical image segmentation processing method and system based on deep learning, and the method and system can automatically recognize and construct the spatial relation between complex tissue structures through precise matching and combination of tissue contour data in an initial three-dimensional medical scanning image, and generate a coherent and accurate tissue contour data combination path. Furthermore, the Gaussian filtering technology is adopted to smooth the path organization boundary, so that the influence of noise and artifacts is effectively reduced, and the generated filtering organization boundary is clearer and more continuous. And then, intelligent reconstruction is performed on the filtering organization boundary according to the state node sequence of the organization contour data combination, so that key structure information is reserved, potential boundary deviation is corrected, and the accuracy of the organization contour is greatly improved. And finally, based on the optimized tissue contour, fine adjustment is performed on the tissue segmentation boundary in the initial image, and the generated target three-dimensional medical scanning image shows higher detail retention degree and boundary definition in the aspect of tissue segmentation.

Description

Technical Field

[0001] The present application relates to the field of smart medical technology, and more specifically, to a three-dimensional medical image segmentation processing method and system based on deep learning. Background Art

[0002] In the field of medical image processing, 3D medical image segmentation is a crucial technology, which is of great significance for disease diagnosis, formulation of treatment plans and medical research. Traditional 3D medical image segmentation methods mostly rely on manual operations or simple image processing algorithms. These methods often face problems such as low accuracy, long time consumption and susceptibility to subjective factors when dealing with complex tissue structures. Especially when faced with diverse and complex 3D medical scanning images, such as CT (computed tomography) and MRI (magnetic resonance imaging), traditional methods are difficult to accurately and efficiently segment the tissue area of ​​interest. Summary of the invention

[0003] In view of this, the purpose of this application is to provide a three-dimensional medical image segmentation processing method and system based on deep learning.

[0004] According to the first aspect of the present application, a three-dimensional medical image segmentation processing method based on deep learning is provided, the method comprising:

[0005] Acquire tissue contour data in an initial three-dimensional medical scan image, and match the tissue contour data based on a deep learning model according to tissue structure characteristics to generate a plurality of tissue contour data combinations, each of which includes a first tissue contour and a second tissue contour;

[0006] Based on the three-dimensional coordinate parameters of the tissue boundary starting position of each tissue contour data combination and the three-dimensional coordinate parameters of the tissue boundary ending position of each tissue contour data combination, a plurality of tissue contour data combinations are aggregated to generate a tissue contour data combination path;

[0007] Performing Gaussian filtering on the path tissue boundary corresponding to the tissue contour data combination path to generate a filtered tissue boundary, wherein the path tissue boundary is composed of tissue boundaries of a plurality of tissue contour data combinations;

[0008] Reconstructing the filtered tissue boundary according to the state node sequence of the tissue contour data combination, and updating the tissue boundary covering the tissue contour data combination according to the reconstructed tissue boundary;

[0009] Optimizing the first tissue contour and the second tissue contour respectively according to the reconstructed tissue boundary to generate a first optimized tissue contour and a second optimized tissue contour;

[0010] According to the first optimized tissue contour and the second optimized tissue contour, the tissue segmentation boundary in the initial three-dimensional medical scan image is optimized to generate a target three-dimensional medical scan image.

[0011] According to the second aspect of the present application, a three-dimensional medical image segmentation processing system is provided, wherein the three-dimensional medical image segmentation processing system comprises a machine-readable storage medium and a processor, wherein the machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the three-dimensional medical image segmentation processing system implements the aforementioned three-dimensional medical image segmentation processing method based on deep learning.

[0012] According to the third aspect of the present application, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When the computer-executable instructions are executed, the aforementioned deep learning-based three-dimensional medical image segmentation processing method is implemented.

[0013] According to any one of the above aspects, the technical effect of the present application is:

[0014] The embodiments of the present application significantly improve the accuracy and efficiency of medical image processing. By accurately matching and combining the tissue contour data in the initial three-dimensional medical scan image, the spatial relationship between complex tissue structures can be automatically identified and constructed, and a coherent and accurate tissue contour data combination path can be generated. Furthermore, the Gaussian filtering technology is used to smooth the path tissue boundaries, which effectively reduces the influence of noise and artifacts, making the generated filtered tissue boundaries clearer and continuous. Subsequently, the filtered tissue boundaries are intelligently reconstructed based on the state node sequence of the tissue contour data combination, which not only retains key structural information, but also corrects potential boundary deviations, greatly improving the accuracy of the tissue contour. Finally, the tissue segmentation boundaries in the initial image are finely adjusted based on the optimized tissue contour, and the generated target three-dimensional medical scan image shows higher detail retention and boundary clarity in terms of tissue segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A schematic diagram of the process of a three-dimensional medical image segmentation processing method based on deep learning provided in an embodiment of the present application is shown;

[0017] Figure 2 A schematic diagram of the component structure of a three-dimensional medical image segmentation processing system for implementing the above-mentioned deep learning-based three-dimensional medical image segmentation processing method provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] The embodiments of the present application are described below in conjunction with the drawings in the present application. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0019] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when an element is "connected" or "coupled" to another element, the element may be directly connected or coupled to the other element, or may refer to the element and the other element establishing a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling, and the term "and / or" used herein indicates at least one of the items defined by the term, such as "A and / or B" may be implemented as "A", or as "B", or as "A and B".

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the implementation mode of the present application will be further described in detail with reference to the accompanying drawings. The technical solution of the embodiment of the present application and the technical effect produced by the technical solution of the present application are explained below by describing several exemplary implementation modes. It should be pointed out that the following implementation modes can refer to, draw lessons from or combine with each other, and the same terms, similar features and similar implementation steps in different implementation modes are not described repeatedly.

[0021] Figure 1 The flowchart of the 3D medical image segmentation processing method and system based on deep learning provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps in the 3D medical image segmentation processing method based on deep learning of this embodiment can be shared with each other according to actual needs, or some steps can be omitted or maintained. The detailed steps of the 3D medical image segmentation processing method based on deep learning include:

[0022] Step S110, obtaining tissue contour data in the initial three-dimensional medical scan image, and matching the tissue contour data based on a deep learning model according to tissue structure characteristics to generate multiple tissue contour data combinations, each of which includes a first tissue contour and a second tissue contour.

[0023] In detail, the initial three-dimensional medical scan image refers to the three-dimensional image data of the patient's body tissue or organ obtained by medical imaging equipment (such as CT, MRI, etc.), which is usually stored in digital form and contains rich anatomical structure and pathological information. For example, if a doctor needs to conduct a detailed examination of the heart of a heart disease patient, they may use a CT scanning device to obtain three-dimensional image data of the patient's heart. These three-dimensional image data are the initial three-dimensional medical scan images, which show the position and shape of various parts of the heart (such as atria, ventricles, blood vessels, etc.) in three-dimensional space.

[0024] The tissue contour data refers to information extracted from the initial three-dimensional medical scan image that describes the shape and position of the edge of a specific tissue, usually in the form of a point set, a polygonal mesh, or a voxel mesh, and can accurately represent the boundary of the tissue. For example, in a cardiac CT scan image, the server may use an edge detection algorithm to extract the contour data of the left atrium and left ventricle of the heart. These contour data are composed of a series of points or line segments, which accurately outline the edge shape of the left atrium and left ventricle.

[0025] The tissue structure feature refers to the feature information used to describe and distinguish different tissue types, including shape, size, positional relationship, texture, grayscale value, etc., which helps the server make accurate judgments when extracting and matching tissue contour data. For example, when matching the contour data of the left atrium and left ventricle of the heart, the server may analyze their shape features (such as whether they are elliptical), size features (such as the size of the major and minor axes), positional relationship features (such as whether they are adjacent and in the same direction), etc. This information helps the server determine which contour data should be combined together.

[0026] The tissue contour data combination refers to a unit formed by combining multiple related tissue contour data according to tissue structure characteristics. Each tissue contour data combination usually contains at least two related tissue contours, such as the left atrium and left ventricle of the heart. For example, in the process of processing cardiac CT scan images, the server may combine the contour data of the left atrium and the contour data of the left ventricle into a tissue contour data combination, which represents these two closely connected parts of the heart and will be regarded as a whole in the subsequent analysis and visualization process.

[0027] Based on the above description, in this embodiment, the server first obtains initial three-dimensional medical scan images from a medical imaging device or storage system. These initial three-dimensional medical scan images can be obtained through CT, MRI or other imaging technologies. The server can perform pre-processing on these initial three-dimensional medical scan images, such as denoising, enhancement and other operations, to improve image quality.

[0028] Next, the server will use image processing algorithms, such as edge detection and threshold segmentation, to identify and extract tissue contour data from the initial three-dimensional medical scan image. This tissue contour data may include edge information of different tissues such as the heart, lungs, liver, and blood vessels.

[0029] After extracting the tissue profile data, the server will match them according to the tissue structure characteristics. For example, the server can analyze the shape, size, position relationship and relative position of the profile with other tissues to determine which tissue profile data should be grouped together. This process can be improved with the help of machine learning or deep learning models to improve the accuracy and efficiency of matching.

[0030] Finally, the server generates multiple tissue contour data combinations, each of which contains at least two related tissue contours, namely a first tissue contour and a second tissue contour. For example, in one combination, the first tissue contour may be the left atrium of the heart, and the second tissue contour may be the adjacent left ventricle.

[0031] Step S120, based on the three-dimensional coordinate parameters of the tissue boundary starting position of each tissue contour data combination and the three-dimensional coordinate parameters of the tissue boundary ending position of each tissue contour data combination, multiple tissue contour data combinations are aggregated to generate a tissue contour data combination path.

[0032] In detail, the three-dimensional coordinate parameters of the starting and ending positions of the tissue boundaries refer to the precise position information describing the starting and ending points of the tissue contour in three-dimensional space. These coordinate parameters are usually composed of three components, X, Y, and Z, representing the horizontal, longitudinal, and depth directions in space, respectively. For example, in a cardiac CT scan image, the contour data of the left atrium may have a clear starting point and end point. The server will measure the coordinate parameters of these two points in three-dimensional space (such as X1, Y1, Z1 and X2, Y2, Z2) for subsequent analysis of the spatial relationship between the left atrium and other tissues (such as the left ventricle).

[0033] The tissue contour data combination path refers to a continuous path formed by combining multiple tissue contour data according to their actual positional relationship in three-dimensional space. This path reflects the spatial connection relationship between different tissues and provides a basis for subsequent tissue boundary analysis and optimization. For example, in the process of processing cardiac CT scan images, the server may combine tissue contour data such as the left atrium, left ventricle, right atrium and right ventricle into a continuous tissue contour data combination path. This path starts from a certain starting point of the heart (such as the left atrium), passes through the left ventricle, right atrium and right ventricle in sequence, and finally reaches a certain end point of the heart.

[0034] That is, in this embodiment, after generating multiple tissue contour data combinations, the server will further analyze the three-dimensional coordinate parameters of the starting position and the ending position of the tissue contour boundary in each tissue contour data combination. These three-dimensional coordinate parameters are accurately measured through image processing technology and describe the exact position of the tissue contour in three-dimensional space.

[0035] Next, the server will find the spatial association relationship between each tissue profile data combination based on these three-dimensional coordinate parameters. For example, the server may find that the end position of a tissue profile data combination is very close to the start position of another tissue profile data combination, which indicates that the two tissue profile data combinations are connected in space.

[0036] In this way, the server can aggregate multiple tissue profile data into a continuous tissue profile data combination path, which not only reflects the spatial relationship between different organizations, but also provides a basis for subsequent tissue boundary analysis and optimization.

[0037] Step S130, performing Gaussian filtering on the path tissue boundary corresponding to the tissue contour data combination path to generate a filtered tissue boundary, wherein the path tissue boundary is composed of tissue boundaries of multiple tissue contour data combinations.

[0038] In detail, Gaussian filtering is a commonly used image processing technique that smoothes an image by applying a Gaussian function to remove noise and details in the image while retaining the main structural information. During Gaussian filtering, the new value of each pixel is the weighted average of the pixel values ​​in its neighborhood, with the weight given by the Gaussian function. For example, during the processing of cardiac CT scan images, the server may perform Gaussian filtering on the path tissue boundary corresponding to the combined path of tissue contour data. By adjusting the parameters of the filter (such as kernel size and standard deviation), the server can remove some small fluctuations and burrs on the path tissue boundary to make it smoother and clearer.

[0039] The filtered tissue boundary refers to the tissue boundary after Gaussian filtering. Compared with the original tissue boundary, the filtered tissue boundary is smoother and has less noise, but retains the main structural information. For example, after Gaussian filtering, the tissue boundary in the cardiac CT scan image (such as the boundary between the left atrium and the left ventricle) becomes smoother and clearer, and this filtered tissue boundary will serve as the basic data for subsequent reconstruction and optimization.

[0040] That is, after generating the tissue contour data combination path, the server will perform Gaussian filtering on the path tissue boundary corresponding to the tissue contour data combination path. Specifically, the path tissue boundary is regarded as a special image signal, and a Gaussian filter is applied to smooth its surface. By adjusting the parameters of the filter, such as the kernel size and standard deviation, the server can remove some small fluctuations and burrs while retaining the overall shape of the tissue boundary.

[0041] After Gaussian filtering, the server will obtain a smoother and clearer filtering organization boundary, which will serve as the basis for subsequent organization boundary reconstruction and optimization.

[0042] Step S140, reconstructing the filtered tissue boundary according to the state node sequence of the tissue contour data combination, and updating the tissue boundary covering the tissue contour data combination according to the reconstructed tissue boundary.

[0043] In detail, the state node sequence refers to a sequence that describes the order and connection relationship of tissue contour data combinations in the path. Each state node represents a tissue contour data combination, and the connection relationship between state nodes reflects their actual positional relationship in three-dimensional space. For example, in the process of processing cardiac CT scan images, the server may construct a state node graph to represent the order and connection relationship between tissue contour data combinations such as the left atrium, left ventricle, right atrium and right ventricle, and the state node graph will serve as a reference for subsequent reconstruction of filtered tissue boundaries.

[0044] The reconstruction of the filtered tissue boundary refers to the process of segmenting, adjusting and recombining the filtered tissue boundary according to the state node sequence. Through reconstruction, the server can generate a tissue boundary model that is more in line with the actual anatomical structure. For example, in the process of processing cardiac CT scan images, the server may segment the filtered tissue boundary according to the state node sequence, and recombin and adjust the segments according to the connection relationship between adjacent tissue contour data combinations. The reconstructed filtered tissue boundary finally generated will more accurately reflect the actual boundary relationship between the various parts of the heart.

[0045] That is, after obtaining the filtered organization boundary, the server will reconstruct it according to the state node sequence of the organization profile data combination. The state node sequence describes the order and connection relationship of the organization profile data combination in the path.

[0046] First, the server analyzes the starting and ending positions of each tissue profile data combination to determine their relative relationship and spatial position. Then, based on this information, the server constructs a state node graph, where each node represents a tissue profile data combination and the edge represents the connection relationship between them.

[0047] Next, the server will traverse this state node graph and segment and reconstruct the filtered tissue boundary according to the preset rules or algorithms. For example, the server may cut the filtered tissue boundary at the connection point of two adjacent tissue contour data combinations, and reassemble and connect these segments according to the position information of the cutting point to form a new reconstructed tissue boundary that is more in line with the actual tissue structure.

[0048] Finally, the server will use these reconstructed organizational boundaries to update and overwrite the organizational boundaries of the original organizational contour data combination, so that the server obtains a more accurate and detailed organizational boundary model.

[0049] Step S150, optimizing the first tissue contour and the second tissue contour respectively according to the reconstructed tissue boundary to generate a first optimized tissue contour and a second optimized tissue contour, and optimizing the tissue segmentation boundary in the initial three-dimensional medical scan image according to the first optimized tissue contour and the second optimized tissue contour to generate a target three-dimensional medical scan image.

[0050] After completing the reconstruction of the tissue boundaries, the server will further use these reconstructed tissue boundaries to optimize the tissue segmentation boundaries in the initial three-dimensional medical scan image.

[0051] First, the server reclassifies and labels each pixel or voxel in the original image based on the shape and position information of the reconstructed tissue boundary. For example, the server labels pixels or voxels inside the reconstructed tissue boundary as belonging to a specific tissue, while pixels or voxels outside the boundary are labeled as other tissues or background.

[0052] In this way, the server can generate a new, optimized three-dimensional medical scan image. In this new three-dimensional medical scan image, the segmentation boundaries between different tissues are clearer, more accurate, and more in line with the actual tissue structure, which can not only improve the accuracy and efficiency of medical diagnosis, but also provide more powerful support for subsequent medical analysis and treatment. Therefore, in the subsequent three-dimensional medical image segmentation process, the target tissue image can be segmented from the target three-dimensional medical scan image based on the optimized tissue segmentation boundaries of the target three-dimensional medical scan image.

[0053] In addition, during the optimization process, the server can also pay special attention to protecting important tissue structures and detailed information. For example, for fine structures such as blood vessels and nerves, the server will use more sophisticated segmentation algorithms and parameter settings to ensure the integrity and accuracy of these structures. At the same time, the server will also conduct quality assessment and verification of the optimization results to ensure that they meet the requirements of medical diagnosis and application.

[0054] Based on the above steps, the embodiment of the present application significantly improves the accuracy and efficiency of medical image processing. By accurately matching and combining the tissue contour data in the initial three-dimensional medical scan image, the spatial relationship between complex tissue structures can be automatically identified and constructed, and a coherent and accurate tissue contour data combination path can be generated. Further, the Gaussian filtering technology is used to smooth the path tissue boundaries, which effectively reduces the influence of noise and artifacts, making the generated filtered tissue boundaries clearer and continuous. Subsequently, the filtered tissue boundaries are intelligently reconstructed based on the state node sequence of the tissue contour data combination, which not only retains key structural information, but also corrects potential boundary deviations, greatly improving the accuracy of the tissue contour. Finally, the tissue segmentation boundaries in the initial image are finely adjusted based on the optimized tissue contour, and the generated target three-dimensional medical scan image shows higher detail retention and boundary clarity in terms of tissue segmentation.

[0055] In a possible implementation, step S120 includes:

[0056] Step S121, based on the three-dimensional coordinate parameters of the tissue boundary starting position of each tissue contour data combination and the three-dimensional coordinate parameters of the tissue boundary ending position of each tissue contour data combination, find the associated tissue contour data combination corresponding to each tissue contour data combination.

[0057] Step S122: according to the spatial association relationship between the tissue contour data combinations, a plurality of the tissue contour data combinations are aggregated to generate a tissue contour data combination path.

[0058] In this embodiment, the server uses advanced image processing algorithms (such as edge detection, threshold segmentation, etc.) to identify and extract tissue contour data in the image. These algorithms can accurately identify the edge information of different tissues, such as the heart, lungs, liver, blood vessels, etc. During the extraction process, the server will carefully analyze the gray value, texture features, etc. of each pixel or voxel to ensure the accuracy and completeness of the contour.

[0059] After extracting the tissue contour data, the server matches the data according to the tissue structure characteristics. This step is the key to constructing the tissue contour data combination. The server analyzes the shape, size, position relationship and relative position of the contour with other tissues, and uses this information to determine which contour data should be combined together. For example, the left atrium and left ventricle of the heart can be combined into a tissue contour data combination because of their close proximity in space and their functional interrelationship.

[0060] During the matching process, the server can use machine learning or deep learning models to improve the accuracy and efficiency of matching. These deep learning models are optimized with a large amount of training data and can automatically identify tissue profiles with similar characteristics and accurately combine them together.

[0061] Once multiple tissue contour data combinations are generated, the server will then look for the association between them based on the three-dimensional coordinate parameters of the starting and ending positions of the tissue boundaries of each tissue contour data combination. These three-dimensional coordinate parameters are accurately measured through image processing technology and accurately describe the position of the tissue contour in three-dimensional space.

[0062] The server will traverse all the tissue profile data combinations, and for each tissue profile data combination, it can check the coordinates of the starting and ending positions of its tissue boundaries. Then, the server will find out whether there are other combinations whose tissue boundary coordinates are close or connected to it. If such a combination is found, the server will consider that there is a spatial association relationship between them and mark them as associated tissue profile data combinations.

[0063] For example, suppose that when the server checks the contour data combination of the left ventricle of the heart, it finds that its terminal position coordinates are very close to the starting position coordinates of another contour data combination marked as the left atrium of the heart. At this time, the server will believe that there is an upstream and downstream spatial correlation relationship between the two combinations and mark them as mutually related combinations.

[0064] After determining all associated tissue profile data combinations, the server begins to aggregate them into a continuous tissue profile data combination path, a process similar to drawing a line connecting all related tissue profiles in three-dimensional space.

[0065] The server will first select a starting organizational profile data combination as the starting point of the path. Then, based on the spatial association relationship, it can find the next combination with downstream spatial association in the starting combination and add it to the path. Next, the server will continue to find downstream associated combinations in the added combinations and add them to the path in turn. This process will be repeated until all relevant organizational profile data combinations are added to the path.

[0066] During the convergence process, the server pays special attention to maintaining the continuity and smoothness of the path. It analyzes the coordinate differences and angle changes between adjacent combinations to ensure that the path presents a natural curve shape in three-dimensional space. At the same time, the server also performs necessary optimization and adjustments to the path to eliminate any possible deviations or errors.

[0067] In one example, suppose the server is processing a 3D medical scan image containing the heart, lungs, and blood vessels. First, the server extracts the tissue contour data of the left atrium, left ventricle, right atrium, right ventricle, lungs, and blood vessels through an image processing algorithm. Then, the server matches and combines these contour data according to their shapes, sizes, and positional relationships, generating multiple tissue contour data combinations.

[0068] Next, the server starts to look for the associations between these tissue contour data combinations. When checking the contour data combination of the left ventricle of the heart, the server finds that its end position coordinates are very close to the start position coordinates of the left atrium of the heart. At the same time, it also finds that the downstream of the left ventricle (i.e., downstream of the end position) is connected to the start position of the right ventricle. Therefore, the server marks the left atrium, left ventricle, and right ventricle as mutually associated tissue contour data combinations.

[0069] Similarly, the server also checked the contour data combinations of the lungs and blood vessels and found the correlation between them. Finally, the server aggregated these interrelated combinations into a continuous tissue contour data combination path, which starts from the left atrium of the heart, passes through the left ventricle and right ventricle, and finally extends to the lungs and blood vessels. It not only accurately reflects the positional relationship of these tissues in three-dimensional space, but also provides an important reference for subsequent tissue boundary analysis and optimization.

[0070] In a possible implementation, step S122 includes:

[0071] Step S1221, determining a target tissue profile data combination from the plurality of tissue profile data combinations.

[0072] Step S1221, searching for an upstream associated tissue profile data combination having an upstream spatial association with the target tissue profile data combination from the associated tissue profile data combinations corresponding to the target tissue profile data combination, and arranging the upstream associated tissue profile data combination before the target tissue profile data combination in the tissue profile data combination path.

[0073] Step S1221, searching for a downstream associated tissue profile data combination having a downstream spatial association with the target tissue profile data combination from the associated tissue profile data combination corresponding to the target tissue profile data combination, and arranging the downstream associated tissue profile data combination after the target tissue profile data combination in the tissue profile data combination path.

[0074] Return to step S1221 until each tissue profile data combination is polled to generate a tissue profile data combination path.

[0075] In this embodiment, the server first loads all matched and generated tissue contour data combinations into an internal data structure, such as a list or database table, which contains information such as the unique identifier of each tissue contour data combination, the three-dimensional coordinate parameters of the starting and ending positions of the tissue boundaries, etc.

[0076] Then, the server will select a starting point to determine the first target tissue contour data combination. The starting point can be arbitrary, but usually for efficiency considerations, the server will select the tissue contour data combination located at the center of the image or a specific anatomical position (such as the bottom of the heart) as the starting point. The server traverses this data structure to find the tissue contour data combination corresponding to this starting point and determines it as the current target tissue contour data combination.

[0077] Once the target tissue contour data combination is determined, the server will then search for an upstream associated tissue contour data combination with an upstream spatial association with it. The server will again traverse all tissue contour data combinations and compare the three-dimensional coordinate parameters of the tissue boundary end position of each combination with the tissue boundary start position of the target tissue contour data combination.

[0078] Specifically, the server calculates the Euclidean distance or other spatial distance measure between the two locations to determine whether they are close enough to indicate upstream spatial association. At the same time, the server may also consider directionality factors, such as ensuring that the end location of the upstream tissue profile data combination points to the start location of the target tissue profile data combination.

[0079] If an upstream associated organization profile data combination that meets the conditions is found, the server will arrange its position in the organization profile data combination path before the current target organization profile data combination, which usually means that the server will update the representation of the path in the internal data structure, such as by adding a link or pointer to the upstream associated organization profile data combination.

[0080] After determining the upstream association (if any), the server proceeds to look for downstream associated tissue profile data combinations that have downstream spatial associations with the target tissue profile data combination. The process is similar to looking for upstream associations, but the server now compares the end position of the tissue boundary of the target tissue profile data combination with the start position of the tissue boundary of each other combination.

[0081] The server also calculates the spatial distance between the two locations and considers the directional factors to determine the downstream spatial association. Once a downstream associated tissue profile data combination that meets the conditions is found, the server will arrange its position in the tissue profile data combination path after the current target tissue profile data combination and update the internal data structure to reflect this change.

[0082] After completing the upstream and downstream association searches for the current target tissue profile data combination, the server returns to the step of determining the target tissue profile data combination from multiple tissue profile data combinations, but this time selects the next tissue profile data combination that has not been processed in the path as the new target tissue profile data combination.

[0083] The server will repeat the above upstream and downstream association search process until each tissue profile data combination is polled to ensure that all combinations are correctly arranged in the tissue profile data combination path according to their actual positional relationship in the three-dimensional space.

[0084] In a specific example, assume that the server is processing a 3D medical scan image containing different parts of the heart (such as the left atrium, left ventricle, right atrium, and right ventricle). The server has extracted and matched the tissue contour data of these parts through the previous steps to generate a corresponding tissue contour data combination.

[0085] The server first selects the left atrium as the starting point and determines it as the first target tissue contour data combination. Then, the server searches for an upstream associated tissue contour data combination that has an upstream spatial association with the left atrium, but in this example, the left atrium is the most upstream part of the heart, so there is no upstream association.

[0086] Next, the server searches for a downstream associated tissue contour data combination that has a downstream spatial association with the left atrium. The server finds that the starting position of the left ventricle is adjacent to the ending position of the left atrium, so the left ventricle is determined as the downstream associated tissue contour data combination and arranged after the left atrium.

[0087] The server continues this process, adding the right atrium and right ventricle to the tissue contour data combination path according to their actual positional relationship in the heart. Finally, the server generates a complete tissue contour data combination path that accurately reflects the positional relationship of each part of the heart in three-dimensional space.

[0088] Through this process, not only the tissue contour data combination path is constructed, but also important spatial structure information is provided for subsequent tissue boundary analysis and optimization.

[0089] In a possible implementation, step S1221 includes:

[0090] Step S1221-1: acquiring, from the associated tissue profile data combinations corresponding to the target tissue profile data combination, an upstream tissue profile data combination having an upstream spatial association with the target tissue profile data combination.

[0091] Step S1221-1, obtaining the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination.

[0092] Step S1221-1, based on the starting three-dimensional coordinate parameters of the tissue boundaries of the upstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundaries of the upstream tissue contour data combination, and the starting three-dimensional coordinate parameters of the tissue boundaries of the target tissue contour data combination, analyze whether the upstream tissue contour data combination is the upstream associated tissue contour data combination of the target tissue contour data combination.

[0093] Return to step S1221 - 1 until all upstream tissue profile data combinations that have upstream spatial associations with the target tissue profile data combination are polled.

[0094] In this embodiment, when the server performs this step, it has determined the current target organization profile data combination and has identified other organization profile data combinations associated with the target organization profile data combination. The server maintains a data structure (such as a list, database table or graph structure) for storing these associations.

[0095] The server will first retrieve all upstream candidate tissue profile data combinations associated with the target tissue profile data combination from this data structure. "Upstream" here refers to the tissue profile data combination that is spatially located before the target tissue profile data combination, which usually means that their tissue boundary ending positions are close to or connected to the tissue boundary starting positions of the target tissue profile data combination.

[0096] For each upstream candidate tissue contour data combination, the server needs to obtain the starting three-dimensional coordinate parameters and ending three-dimensional coordinate parameters of its tissue boundary. At the same time, the server also obtains the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination for subsequent spatial relationship analysis.

[0097] These three-dimensional coordinate parameters are accurately measured by the previous image processing steps, and they describe the exact position of the tissue contour in three-dimensional space. The server retrieves the required information from the data structure storing these parameters.

[0098] The server will then analyze whether the upstream candidate tissue profile data combination is the upstream associated tissue profile data combination of the target tissue profile data combination based on the obtained three-dimensional coordinate parameters. In detail, the server will first calculate the spatial distance between the end position of the upstream candidate tissue profile data combination and the start position of the target tissue profile data combination, which can be the Euclidean distance or other suitable spatial distance measurement. In addition to the distance, the server will also consider the directionality factor. It will check whether the end position of the upstream candidate tissue profile data combination points to the start position of the target tissue profile data combination, or whether there is a natural connection path between them. The server will apply certain tolerances or thresholds to determine whether the distance and directionality meet the conditions for upstream association. These tolerances and thresholds can be set based on experience or dynamically adjusted through machine learning or optimization algorithms. If the upstream candidate tissue profile data combination meets all of the above conditions, the server will mark it as the upstream associated tissue profile data combination of the target tissue profile data combination.

[0099] The server will repeatedly execute the steps of obtaining the upstream tissue profile data combination from the associated tissue profile data combination, obtaining the three-dimensional coordinate parameters, and analyzing the upstream association relationship until all upstream candidate tissue profile data combinations associated with the target tissue profile data combination are polled.

[0100] During this process, the server may discover multiple upstream associated organization profile data combinations, especially when the target organization profile data combination is located at the intersection of multiple organizations. The server will record all these upstream associations and use them in subsequent steps to build a complete organization profile data combination path.

[0101] In a specific example, assume that the server is processing a three-dimensional medical scan image containing different parts of the heart, and has determined that the current target tissue contour data combination is the left ventricle. The server knows that the left ventricle is usually closely connected to the left atrium in space, so it first searches for upstream candidate tissue contour data combinations associated with the left ventricle.

[0102] In the associated data structure, the server finds the left atrium as one of the upstream candidate tissue contour data combinations. The server then obtains the tissue boundary start and end three-dimensional coordinate parameters of the left atrium, and the start three-dimensional coordinate parameters of the left ventricle.

[0103] By analyzing these coordinate parameters, the server finds that the end position of the left atrium is very close to the start position of the left ventricle, and the directionality also meets the upstream association condition (i.e., the end position of the left atrium points to the start position of the left ventricle). Therefore, the server marks the left atrium as the upstream association tissue contour data combination of the left ventricle.

[0104] If there are other upstream candidate tissue contour data combinations (although unlikely in this particular scenario), the server will continue to repeat the above steps until all candidate combinations are polled. However, in this example, the server only finds the left atrium as an upstream associated tissue contour data combination and adds it to the position before the left ventricle in the tissue contour data combination path.

[0105] In a possible implementation, step S1221-1 includes:

[0106] Step S1221-11, if, based on the ending three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination, it is analyzed that there is no deviation between the ending position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, the upstream tissue contour data is output as the upstream associated tissue contour data combination of the target tissue contour data combination.

[0107] Step S1221-12, if based on the ending three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination, it is analyzed whether there is a deviation between the ending position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, based on the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination, it is analyzed whether there is a deviation between the starting position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination.

[0108] Step S1221-13, if analysis shows that there is no deviation between the starting position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, the boundary feature points on the tissue boundary of the upstream tissue contour data combination are rearranged in reverse order to optimize the upstream tissue contour data combination, and the optimized upstream tissue contour data is output as the upstream associated tissue contour data combination of the target tissue contour data combination.

[0109] Step S1221-14, if the analysis shows that there is a deviation between the starting position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, it is determined that the upstream tissue contour data is not the upstream associated tissue contour data combination of the target tissue contour data combination.

[0110] In this embodiment, the server first obtains the tissue boundary ending three-dimensional coordinate parameters (recorded as P1) of the upstream tissue contour data combination and the tissue boundary starting three-dimensional coordinate parameters (recorded as P2) of the target tissue contour data combination. Then, the server calculates the distance between the two coordinate points and determines whether the distance is less than a preset threshold (the threshold is used to define the standard of "no deviation").

[0111] For example, suppose the server is processing a three-dimensional medical scan image containing the left ventricle and left atrium of the heart. The starting position coordinate P2 of the left ventricle is known, and the server now checks the relationship between the ending position coordinates P1 and P2 of the left atrium. If the calculated distance is less than a preset threshold (e.g., 5 mm), the server believes that there is no significant deviation between the ending position of the left atrium and the starting position of the left ventricle, and therefore preliminarily determines that the left atrium can be the upstream associated tissue contour data combination of the left ventricle.

[0112] If the distance between P1 and P2 is greater than the preset threshold, indicating that there is a deviation between the end position and the starting position, the server will then analyze the relationship between the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination (denoted as P0) and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination (P2).

[0113] For example, if the distance between the end position P1 of the left atrium and the start position P2 of the left ventricle is too large, the server further checks the relationship between the start positions P0 and P2 of the left atrium. The server calculates the distance between P0 and P2 and applies the preset threshold again for judgment.

[0114] If the distance between P0 and P2 is less than a preset threshold, it means that the starting position of the upstream tissue contour data combination is spatially close to the starting position of the target tissue contour data combination, although there is a deviation between their ending position and starting position.

[0115] For example, if the starting position P0 of the left atrium is close enough to the starting position P2 of the left ventricle (the distance is less than a preset threshold), the server will consider that this may be caused by a slight error in the image acquisition process or a slight variation in the tissue structure itself.

[0116] In this case, the server will attempt to reversely rearrange the boundary feature points on the tissue boundary of the upstream tissue contour data combination to adjust its shape and direction so that it is better aligned with the target tissue contour data combination. Reverse rearrangement may involve transforming or reordering the coordinates of the boundary feature points to ensure that the end position of the upstream tissue contour data combination can more accurately point to the start position of the target tissue contour data combination.

[0117] For example, the server will reverse the order of the tissue boundary feature points of the left atrium to fine-tune its shape and orientation so that the end position of the left atrium can be more closely connected to the starting position of the left ventricle. This optimization may involve some subtle adjustments to the edge of the left atrium contour to ensure its continuity with the left ventricle.

[0118] After completing the reverse rearrangement and optimization, the server will output the optimized upstream tissue profile data combination as the upstream associated tissue profile data combination of the target tissue profile data combination, which means that the server has confirmed that this upstream tissue profile data combination is spatially associated with the target tissue profile data combination, and has made necessary adjustments to it to optimize this association relationship.

[0119] For example, after reverse rearrangement and optimization, the server confirms that the left atrium is the upstream associated tissue contour data combination of the left ventricle, and adds it to the corresponding position in the tissue contour data combination path. In this way, the server constructs a more accurate and continuous tissue contour data combination path that reflects the actual spatial relationship between different parts of the heart.

[0120] If the distance between P0 and P2 is greater than the preset threshold, it means that there is a significant spatial deviation between the starting position of the upstream tissue profile data combination and the starting position of the target tissue profile data combination. The server will determine that this upstream tissue profile data combination is not the upstream associated tissue profile data combination of the target tissue profile data combination.

[0121] For example, if the distance between the starting position P0 of the left atrium and the starting position P2 of the left ventricle is too large (exceeding a preset threshold), the server will consider that the left atrium is not the direct upstream associated tissue contour data combination of the left ventricle. In this case, the server may continue to search for other possible upstream associated tissue contour data combinations, or mark the current target tissue contour data combination as an isolated node in the path (if there is no other associated tissue contour data combination).

[0122] In a possible implementation, step S1221 may further include:

[0123] Step A110: acquiring, from the associated tissue contour data combinations corresponding to the target tissue contour data combination, a downstream tissue contour data combination having a downstream spatial association with the target tissue contour data combination.

[0124] Step A120, obtaining the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, and the ending three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination.

[0125] Step A130, based on the starting three-dimensional coordinate parameters of the tissue boundaries of the downstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundaries of the downstream tissue contour data combination, and the ending three-dimensional coordinate parameters of the tissue boundaries of the target tissue contour data combination, analyze whether the downstream tissue contour data combination is a downstream associated tissue contour data combination of the target tissue contour data combination.

[0126] Return to step A110 until all downstream tissue contour data combinations having downstream spatial association with the target tissue contour data combination are found through polling.

[0127] In this embodiment, when processing a three-dimensional medical scan image, the server has determined the current target tissue contour data combination and has identified other tissue contour data combinations associated with the target tissue contour data combination. The server will traverse these associated tissue contour data combinations to find a downstream tissue contour data combination that has a downstream spatial association with the target tissue contour data combination.

[0128] For example, suppose the server is processing a 3D medical scan image containing different parts of the heart, and the current target tissue contour data set is the left ventricle. The server knows that the left ventricle is usually spatially connected to the right ventricle, so it first looks for downstream candidate tissue contour data sets associated with the left ventricle.

[0129] The server retrieves all possible downstream candidate tissue contour data combinations from the associated data structure. These candidate combinations may be tissue parts such as the right ventricle, aorta, etc., which are located downstream of the left ventricle.

[0130] For each downstream candidate tissue contour data combination, the server needs to obtain the starting three-dimensional coordinate parameters and the ending three-dimensional coordinate parameters of its tissue boundary. At the same time, the server also obtains the ending three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination for subsequent spatial relationship analysis.

[0131] For example, the server will obtain the starting three-dimensional coordinate parameters and the ending three-dimensional coordinate parameters of the tissue boundary of the right ventricle, and will also obtain the ending three-dimensional coordinate parameters of the left ventricle. These coordinate parameters accurately describe the positional relationship between the right ventricle and the left ventricle in three-dimensional space.

[0132] The server will then analyze whether the downstream candidate tissue contour data combination is the downstream associated tissue contour data combination of the target tissue contour data combination based on the obtained three-dimensional coordinate parameters. In detail, the server will first calculate the spatial distance between the starting position of the downstream candidate tissue contour data combination and the ending position of the target tissue contour data combination, which can be the Euclidean distance or other suitable spatial distance measurement. In addition to the distance, the server will also consider the directionality factor. It will check whether the starting position of the downstream candidate tissue contour data combination is naturally extended from the ending position of the target tissue contour data combination, or whether there is a natural connection path between them. Among them, the server will apply a certain tolerance or threshold to determine whether the distance and directionality meet the conditions for downstream association. These tolerances and thresholds can be set based on experience or dynamically adjusted through machine learning or optimization algorithms. For example, the server will calculate the distance between the starting position of the right ventricle and the ending position of the left ventricle, and check whether the directionality is consistent. If the distance is less than the preset threshold and the directionality is correct, the server will preliminarily determine that the right ventricle is the downstream associated tissue contour data combination of the left ventricle.

[0133] The server will repeatedly execute the steps of obtaining downstream tissue profile data combinations from associated tissue profile data combinations, obtaining three-dimensional coordinate parameters, and analyzing downstream association relationships until all downstream candidate tissue profile data combinations associated with the target tissue profile data combination are polled.

[0134] For example, the server will continue to check other possible downstream candidate tissue profile data combinations (such as the aorta, etc.) and repeat the above steps for analysis. If multiple downstream associated tissue profile data combinations are found, the server will record the association relationship between them and the target tissue profile data combination.

[0135] After analyzing the above steps, the server will determine which downstream candidate tissue profile data combinations are downstream associated tissue profile data combinations of the target tissue profile data combination, and these downstream associated tissue profile data combinations will be added to the tissue profile data combination path according to their actual position relationship in three-dimensional space.

[0136] For example, after determining that the right ventricle is the downstream associated tissue contour data combination of the left ventricle, the server will add the right ventricle to the position after the left ventricle in the tissue contour data combination path. In this way, the server constructs a more complete and accurate tissue contour data combination path that reflects the spatial relationship between different parts of the heart.

[0137] In a possible implementation, step A130 includes:

[0138] Step A131, if, based on the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, it is analyzed that there is no deviation between the starting position of the tissue boundary of the downstream tissue contour data combination and the ending position of the tissue boundary of the downstream tissue contour data combination, the downstream tissue contour data is output as a downstream associated tissue contour data combination of the downstream tissue contour data combination.

[0139] Step A132, if based on the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, analyze whether there is a deviation between the starting position of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, then based on the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, analyze whether there is a deviation between the ending position of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination.

[0140] Step A133, if the analysis shows that there is no deviation between the termination position of the tissue boundary of the downstream tissue contour data combination and the termination position of the tissue boundary of the downstream tissue contour data combination, the boundary feature points on the tissue boundary of the downstream tissue contour data combination are rearranged in reverse order to optimize the downstream tissue contour data combination, and the optimized downstream tissue contour data is output as the downstream associated tissue contour data combination of the downstream tissue contour data combination.

[0141] Step A134, if the analysis shows that there is a deviation between the end position of the tissue boundary of the downstream tissue contour data combination and the end position of the tissue boundary of the downstream tissue contour data combination, determine that the downstream tissue contour data is not the downstream associated tissue contour data combination of the downstream tissue contour data combination.

[0142] In this embodiment, the server first obtains the starting three-dimensional coordinate parameters (denoted as S) and the ending three-dimensional coordinate parameters (denoted as E) of the tissue boundary of the downstream tissue contour data combination. Then, the server analyzes the spatial relationship between the two coordinate points to determine the internal consistency of the downstream tissue contour data combination.

[0143] For example, suppose the server is processing a 3D medical scan image containing a vascular branch, and the current target tissue contour data set is part of the main vessel. The server is checking a candidate downstream tissue contour data set that represents a possible vascular branch. The server first obtains the coordinates of the starting point S and the ending point E of the vascular branch.

[0144] Next, the server calculates the vector from S to E and analyzes whether the vector points to a reasonable direction (for example, a branch vessel usually extends naturally from the main vessel). If the line between S and E is almost straight and in a reasonable direction, the server considers that the downstream tissue contour data combination is internally consistent, that is, there is no significant deviation between the starting position and the ending position.

[0145] If the server determines that the downstream tissue profile data combination is internally consistent, that is, there is no significant deviation between the starting position and the ending position, then the server will output this downstream tissue profile data combination as the downstream associated tissue profile data combination of the current target tissue profile data combination.

[0146] For example, in the above-mentioned vascular branch example, if the server determines that the line between the starting point S and the ending point E is a straight line and the direction is reasonable, then the server will consider this vascular branch to be the downstream associated tissue contour data combination of the main blood vessel and add it to the tissue contour data combination path.

[0147] If the server finds internal inconsistency in the downstream tissue contour data combination, that is, there is a significant deviation between the starting position and the ending position, the server will further check whether the ending position of this tissue boundary is repeated or abnormal.

[0148] For example, suppose that when the server processes another candidate downstream tissue contour data combination, it finds that the line between its starting point S and the ending point E is tortuous or has an unreasonable direction. At this point, the server will suspect that there may be errors or anomalies in this tissue contour data combination. For further confirmation, the server will analyze whether the coordinates of the ending point E overlap with other known tissue boundary ending positions.

[0149] If the server determines that the end position of the downstream tissue contour data combination is not repeated, the server will try to optimize the tissue contour data combination. The optimization method may include rearranging the boundary feature points on the tissue boundary in reverse order to adjust its shape and direction.

[0150] For example, in the case of a blood vessel branch, if the server finds that the end point E of the candidate branch is not repeated but the line between the start point S and the end point E is unreasonable, the server may try to reverse the order of the feature points on the branch contour. By adjusting the order of these points, the server hopes to obtain a smoother and more reasonable blood vessel branch contour. After optimization, the server will re-analyze the association between this tissue contour data combination and the target tissue contour data combination.

[0151] After optimizing the downstream tissue profile data combination, the server will re-acquire the starting three-dimensional coordinate parameters and the ending three-dimensional coordinate parameters of its tissue boundary, and analyze again whether it is the downstream associated tissue profile data combination of the current target tissue profile data combination.

[0152] For example, after reverse order rearrangement optimization, the server will recheck the spatial relationship between the starting point S' (optimized starting point) and the end point E' (usually the end point will not change unless there are special circumstances) of the vascular branch. If the optimized contour is internally consistent and has a reasonable direction, the server will determine it as the downstream associated tissue contour data combination.

[0153] If the server finds that the end position of the downstream tissue profile data combination is repeated, or is still inconsistent internally after optimization, the server will determine that this tissue profile data combination is not the downstream associated tissue profile data combination of the current target tissue profile data combination.

[0154] For example, if the server finds that the end point E' of the optimized vascular branch still overlaps with other known tissue boundary end positions when checking the optimized vascular branch contour, or the line between the start point S' and the end point E' is still unreasonable, the server will exclude the possibility of this candidate branch as a downstream associated tissue contour data combination. The server will continue to search for other possible downstream associated tissue contour data combinations.

[0155] In a possible implementation, step S140 includes:

[0156] Step S141, based on the starting position and ending position of the first tissue contour and the starting position and ending position of the second tissue contour in each tissue contour data combination, determine the starting position decomposition plane and the ending position decomposition plane corresponding to the tissue contour data combination.

[0157] Step S142, reconstructing the filtered tissue boundary according to the start position decomposition plane and the end position decomposition plane to generate a plurality of reconstructed tissue boundaries.

[0158] Step S143, updating the tissue boundary covering the tissue contour data combination according to the reconstructed tissue boundary.

[0159] In this embodiment, when processing a three-dimensional medical scan image, the server has acquired multiple tissue contour data combinations, each of which includes a first tissue contour and a second tissue contour, and the starting positions and ending positions of these contours have been accurately located in the three-dimensional space. The server now needs to determine the starting position decomposition plane and the ending position decomposition plane corresponding to each tissue contour data combination based on these position information.

[0160] For example, suppose the server is processing a three-dimensional medical scan image containing the left ventricle and the right ventricle of the heart. The left ventricle and the right ventricle are respectively used as the first tissue contour and the second tissue contour in the two tissue contour data combinations (or vice versa, depending on the processing order). The server first determines the starting position and the ending position of the left ventricle, and then constructs a plane based on these two positions as the starting position decomposition surface. Similarly, the server also determines the starting position and the ending position of the right ventricle, and constructs another plane as the ending position decomposition surface. These two decomposition surfaces will be used for the subsequent reconstruction of the filtered tissue boundary.

[0161] It should be noted that the starting position decomposition plane and the ending position decomposition plane are not necessarily strict physical planes. They are more used to describe the starting and ending areas of the tissue contour in three-dimensional space. In practical applications, these decomposition planes may be approximated by fitting algorithms (such as least squares fitting planes).

[0162] After determining the starting position decomposition plane and the ending position decomposition plane, the server will use these decomposition planes to reconstruct the filtered tissue boundary. The filtered tissue boundary is obtained by smoothing such as Gaussian filtering, which may blur the details of the original tissue contour to a certain extent. Through reconstruction, the server aims to restore these details and make the tissue boundary more consistent with the actual anatomical structure.

[0163] For example, the server first loads the filtered tissue boundary data, which may exist in the form of point cloud, polygon mesh or voxel mesh. Then, the server uses ray tracing or plane and curve intersection algorithm to calculate the intersection points or intersection lines of the start position decomposition surface and the end position decomposition surface with the filtered tissue boundary, and these intersection points or intersection lines divide the filtered tissue boundary into multiple segments, each of which is located between two adjacent decomposition surfaces.

[0164] Next, the server further analyzes and processes each segment. It may adjust its position, orientation, or shape based on the segment's shape, size, curvature, and other characteristic parameters to make it more consistent with anatomical constraints. For example, the server may use spline curves or B-spline surfaces to fit each segment to generate a smooth reconstructed tissue boundary that conforms to the actual anatomical structure.

[0165] After completing the reconstruction of the filtered tissue boundary, the server will use the reconstructed tissue boundary to update the tissue boundary covering the original tissue contour data combination, which means that the server will replace the original tissue contour data with a more accurate and more anatomically accurate reconstructed tissue boundary.

[0166] For example, returning to the example of the left and right ventricles of the heart, after the server completes the reconstruction of the filtered tissue boundary, it will obtain a series of more accurate and smooth tissue boundary segments, which describe the connection area between the left and right ventricles more clearly and accurately. The server will now integrate these reconstructed tissue boundary segments to form a complete and updated tissue contour data combination, which will be used for subsequent operations such as tissue segmentation, visualization or quantitative analysis.

[0167] Through this process, the server not only improves the accuracy and reliability of tissue profile data, but also provides more powerful support for medical diagnosis, treatment and research.

[0168] In a possible implementation, step S142 includes:

[0169] Step S1421, obtaining three-dimensional coordinate parameters of the starting position decomposition surface and the ending position decomposition surface, wherein the three-dimensional coordinate parameters include center point coordinates, normal vectors, and existing boundary point coordinates.

[0170] Step S1422, using ray tracing or plane and curve intersection algorithm, respectively calculate the intersection points or intersection lines of the starting position decomposition surface and the ending position decomposition surface with the filtered tissue boundary, and based on the intersection points or intersection lines, segment the filtered tissue boundary to generate multiple tissue boundary segments, each tissue boundary segment corresponds to the part between a starting position decomposition surface and a ending position decomposition surface.

[0171] Step S1423, marking each tissue boundary segment, and recording the starting decomposition surface information and the ending decomposition surface information corresponding to the tissue boundary segment.

[0172] Step S1424, analyzing characteristic parameters of each tissue boundary segment, wherein the characteristic parameters include curvature, length, and connection relationship with other tissue boundary segments.

[0173] Step S1425 , for each tissue boundary segment, a preset reconstruction algorithm is executed according to the starting decomposition plane information and the ending decomposition plane information marked by the tissue boundary segment, and the characteristic parameters of the tissue boundary segment, to generate a plurality of reconstructed tissue boundaries.

[0174] In this embodiment, when processing the three-dimensional medical scan image, the server has determined the corresponding starting position decomposition surface and ending position decomposition surface according to the starting position and ending position of the tissue contour data combination. Now, the server needs to obtain the three-dimensional coordinate parameters of these decomposition surfaces for subsequent use.

[0175] For example, suppose the server is processing a 3D medical scan image containing the bronchus of the lungs. The server has determined the start and end position decomposition surfaces of a bronchus segment. To obtain the 3D coordinate parameters of these decomposition surfaces, the server performs the following operations:

[0176] Calculate the coordinates of the geometric center point of the decomposition surface, which is usually obtained by averaging the positions of all points on the decomposition surface. For planar decomposition surfaces, the center point coordinates may be located at the intersection of the normal lines of the plane.

[0177] Determine the normal vector of the decomposition surface, that is, the unit vector perpendicular to the decomposition surface and pointing outward. This normal vector is crucial for subsequent ray tracing and intersection calculations.

[0178] Record the coordinates of some boundary points of the decomposition surface, which define the edge range of the decomposition surface. Although these boundary points are not required, they can help the server describe the shape and position of the decomposition surface more accurately.

[0179] After obtaining the three-dimensional coordinate parameters of the decomposition surface, the server will then use ray tracing or a plane-curve intersection algorithm to calculate the intersection point or intersection line between the decomposition surface and the filter tissue boundary.

[0180] For example, continuing with the example of the lung bronchi, the server now has the three-dimensional coordinate parameters of the starting position decomposition surface and the ending position decomposition surface. In order to find the intersection points or intersection lines of these decomposition surfaces with the filtered bronchial tissue boundaries, the server will perform the following steps:

[0181] The server may emit rays from the boundary points or center points of the decomposition surface, and these rays extend along the normal vector direction of the decomposition surface (or its opposite direction) until they intersect with the filtering tissue boundary. The intersection point is the point where these rays touch the tissue boundary.

[0182] Another method is to directly calculate the intersection points or intersection lines of the decomposition surface (as a plane) and the filter tissue boundary (as a curve or surface), which usually involves the problem of solving the plane and curve equations simultaneously in analytic geometry.

[0183] Through these methods, the server can divide the filtering organization boundary into multiple sections located between the starting position decomposition plane and the ending position decomposition plane, namely, organization boundary segments.

[0184] After the organization boundary segments are generated, the server will mark each organization boundary segment and record their corresponding starting decomposition surface information and ending decomposition surface information.

[0185] For example, for each bronchial tissue boundary segment, the server assigns a unique identifier and records which start and end position decomposition surfaces define the segment, which is crucial for subsequent reconstruction and analysis.

[0186] The server then analyzes the characteristic parameters of each organizational boundary segment, which help understand the segment's shape, size, and connectivity with other segments.

[0187] For example, the server might calculate the curvature of each tissue boundary segment to understand how curved it is; calculate the length of the segment to understand its size; and analyze the points at which the segment connects to other segments to build a structural model of the entire bronchial tree.

[0188] Finally, the server executes a preset reconstruction algorithm to generate multiple reconstructed tissue boundaries according to the starting decomposition surface information, the ending decomposition surface information and the characteristic parameters of each tissue boundary segment.

[0189] For example, the server may use spline interpolation, B-spline surface fitting, or other advanced geometric modeling techniques to optimize the shape and position of each tissue boundary segment, which can generate a smooth and anatomically correct reconstructed tissue boundary. During the reconstruction process, the server also considers the connectivity between segments to ensure the continuity and consistency of the entire tissue boundary.

[0190] After completing these steps, the server will have more accurate and reliable reconstructed tissue boundary data, which can be used for subsequent medical diagnosis, treatment plan formulation, research analysis, etc.

[0191] Figure 2A three-dimensional medical image segmentation processing system 100 provided in an embodiment of the present application is shown, including a processor 1001 and a memory 1003 and a program code stored in the memory 1003. The processor 1001 executes the above program code to implement the steps of the three-dimensional medical image segmentation processing method based on deep learning.

[0192] Figure 2 The three-dimensional medical image segmentation processing system 100 shown includes: a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the three-dimensional medical image segmentation processing system 100 may also include a transceiver 1004, which can be used for data interaction between the three-dimensional medical image segmentation processing system and other three-dimensional medical image segmentation processing systems, such as data transmission and / or data reception. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of the three-dimensional medical image segmentation processing system 100 does not constitute a limitation on the embodiments of the present application.

[0193] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0194] The bus 1002 may include a path to transmit information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc.

[0195] The memory 1003 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium, other magnetic storage devices, or any other medium that can be used to have or store program code and can be read by a computer, without limitation herein.

[0196] The memory 1003 is used to store program codes for executing the embodiments of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the program codes stored in the memory 1003 to implement the steps shown in the above method embodiments.

[0197] An embodiment of the present application provides a computer-readable storage medium having program code stored thereon. When the program code is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0198] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the implementation order of these steps is not limited to the order represented by the arrows. Unless there is a clear explanation in this article, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders based on demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages according to the actual implementation scenario, and some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage in these sub-steps or stages may also be executed at different times respectively. In different scenarios at the execution time, the execution order of these sub-steps or stages can be flexibly configured based on demand, and the embodiment of the present application does not limit this.

[0199] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the scheme of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. A three-dimensional medical image segmentation processing method based on deep learning, characterized in that: The method comprises: Acquire tissue contour data in an initial three-dimensional medical scan image, and match the tissue contour data based on a deep learning model according to tissue structure characteristics to generate a plurality of tissue contour data combinations, each of which includes a first tissue contour and a second tissue contour; Based on the three-dimensional coordinate parameters of the tissue boundary starting position of each tissue contour data combination and the three-dimensional coordinate parameters of the tissue boundary ending position of each tissue contour data combination, a plurality of tissue contour data combinations are aggregated to generate a tissue contour data combination path; Performing Gaussian filtering on the path tissue boundary corresponding to the tissue contour data combination path to generate a filtered tissue boundary, wherein the path tissue boundary is composed of tissue boundaries of a plurality of tissue contour data combinations; Reconstructing the filtered tissue boundary according to the state node sequence of the tissue contour data combination, and updating the tissue boundary covering the tissue contour data combination according to the reconstructed tissue boundary; Optimizing the first tissue contour and the second tissue contour respectively according to the reconstructed tissue boundary to generate a first optimized tissue contour and a second optimized tissue contour; According to the first optimized tissue contour and the second optimized tissue contour, the tissue segmentation boundary in the initial three-dimensional medical scan image is optimized to generate a target three-dimensional medical scan image.

2. The three-dimensional medical image segmentation processing method based on deep learning according to claim 1, characterized in that: The step of aggregating multiple tissue contour data combinations to generate a tissue contour data combination path based on the three-dimensional coordinate parameters of the tissue boundary starting position of each tissue contour data combination and the three-dimensional coordinate parameters of the tissue boundary ending position of each tissue contour data combination comprises: Based on the three-dimensional coordinate parameters of the tissue boundary starting position of each tissue contour data combination and the three-dimensional coordinate parameters of the tissue boundary ending position of each tissue contour data combination, searching for the associated tissue contour data combination corresponding to each tissue contour data combination; According to the spatial association relationship between the tissue contour data combinations, a plurality of the tissue contour data combinations are aggregated to generate a tissue contour data combination path.

3. The deep learning-based three-dimensional medical image segmentation processing method according to claim 2, characterized in that: The step of aggregating a plurality of tissue contour data combinations to generate a tissue contour data combination path according to the spatial association relationship between the tissue contour data combinations includes: Determine a target tissue contour data combination from a plurality of tissue contour data combinations; Searching for an upstream associated tissue profile data combination having an upstream spatial association with the target tissue profile data combination from the associated tissue profile data combination corresponding to the target tissue profile data combination, and arranging the upstream associated tissue profile data combination before the target tissue profile data combination in the tissue profile data combination path; Searching for a downstream associated tissue profile data combination having a downstream spatial association with the target tissue profile data combination from the associated tissue profile data combination corresponding to the target tissue profile data combination, and arranging the downstream associated tissue profile data combination after the target tissue profile data combination in the tissue profile data combination path; The step of determining a target tissue contour data combination from the plurality of tissue contour data combinations is returned to be executed until each tissue contour data combination is polled to generate a tissue contour data combination path.

4. The three-dimensional medical image segmentation processing method based on deep learning according to claim 3, characterized in that: The step of searching, from the associated tissue profile data combination corresponding to the target tissue profile data combination, for an upstream associated tissue profile data combination having an upstream spatial association with the target tissue profile data combination comprises: Acquire, from the associated tissue contour data combination corresponding to the target tissue contour data combination, an upstream tissue contour data combination having an upstream spatial association with the target tissue contour data combination; Acquire the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination; Analyzing whether the upstream tissue contour data combination is an upstream associated tissue contour data combination of the target tissue contour data combination based on the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination; Return to the step of acquiring an upstream tissue profile data combination having an upstream spatial association with the target tissue profile data combination from the associated tissue profile data combination corresponding to the target tissue profile data combination, until all upstream tissue profile data combinations having an upstream spatial association with the target tissue profile data combination are polled.

5. The deep learning-based 3D medical image segmentation processing method according to claim 4, characterized in that: The step of analyzing whether the upstream tissue contour data combination is an upstream associated tissue contour data combination of the target tissue contour data combination based on the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination, and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination comprises: If, based on the terminating three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination, it is analyzed that there is no deviation between the terminating position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, the upstream tissue contour data is output as an upstream associated tissue contour data combination of the target tissue contour data combination; If, based on the terminating three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination, it is analyzed whether there is a deviation between the terminating position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, based on the starting three-dimensional coordinate parameters of the tissue boundary of the upstream tissue contour data combination and the starting three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination, it is analyzed whether there is a deviation between the starting position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination; If it is analyzed that there is no deviation between the starting position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, the boundary feature points on the tissue boundary of the upstream tissue contour data combination are rearranged in reverse order to optimize the upstream tissue contour data combination, and the optimized upstream tissue contour data is output as the upstream associated tissue contour data combination of the target tissue contour data combination; If a deviation is found between the starting position of the tissue boundary of the upstream tissue contour data combination and the starting position of the tissue boundary of the target tissue contour data combination, it is determined that the upstream tissue contour data is not an upstream associated tissue contour data combination of the target tissue contour data combination.

6. The deep learning-based three-dimensional medical image segmentation processing method according to claim 3, characterized in that: The step of searching for a downstream associated tissue profile data combination having a downstream spatial association with the target tissue profile data combination from the associated tissue profile data combination corresponding to the target tissue profile data combination comprises: Acquire a downstream tissue contour data combination having a downstream spatial association with the target tissue contour data combination from the associated tissue contour data combination corresponding to the target tissue contour data combination; Acquire the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, and the ending three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination; Analyzing whether the downstream tissue contour data combination is a downstream associated tissue contour data combination of the target tissue contour data combination based on the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, and the ending three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination; Return to the step of acquiring a downstream tissue profile data combination having a downstream spatial association with the target tissue profile data combination from the associated tissue profile data combination corresponding to the target tissue profile data combination, until all downstream tissue profile data combinations having a downstream spatial association with the target tissue profile data combination are found by polling.

7. The deep learning-based three-dimensional medical image segmentation processing method according to claim 6, characterized in that: The step of analyzing whether the downstream tissue contour data combination is a downstream associated tissue contour data combination of the target tissue contour data combination based on the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, and the ending three-dimensional coordinate parameters of the tissue boundary of the target tissue contour data combination comprises: If, based on the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, it is analyzed that there is no deviation between the starting position of the tissue boundary of the downstream tissue contour data combination and the ending position of the tissue boundary of the downstream tissue contour data combination, the downstream tissue contour data is output as a downstream associated tissue contour data combination of the downstream tissue contour data combination; If, based on the starting three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, it is analyzed whether there is a deviation between the starting position of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, and based on the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination, it is analyzed whether there is a deviation between the ending position of the tissue boundary of the downstream tissue contour data combination and the ending three-dimensional coordinate parameters of the tissue boundary of the downstream tissue contour data combination; If it is analyzed that there is no deviation between the end position of the tissue boundary of the downstream tissue contour data combination and the end position of the tissue boundary of the downstream tissue contour data combination, the boundary feature points on the tissue boundary of the downstream tissue contour data combination are rearranged in reverse order to optimize the downstream tissue contour data combination, and the optimized downstream tissue contour data is output as a downstream associated tissue contour data combination of the downstream tissue contour data combination; If there is a deviation between the analysis of the end position of the tissue boundary of the downstream tissue contour data combination and the end position of the tissue boundary of the downstream tissue contour data combination, it is determined that the downstream tissue contour data is not the downstream associated tissue contour data combination of the downstream tissue contour data combination.

8. The deep learning-based three-dimensional medical image segmentation processing method according to claim 1, characterized in that: The step of reconstructing the filtered tissue boundary according to the state node sequence of the tissue contour data combination, and updating the tissue boundary covering the tissue contour data combination according to the reconstructed tissue boundary, comprises: Determine a starting position decomposition plane and an ending position decomposition plane corresponding to the tissue contour data combination based on the starting position and the ending position of the first tissue contour and the starting position and the ending position of the second tissue contour in each tissue contour data combination; Reconstructing the filtered tissue boundary according to the starting position decomposition surface and the ending position decomposition surface to generate a plurality of reconstructed tissue boundaries; The tissue boundary covering the tissue contour data combination is updated according to the reconstructed tissue boundary.

9. The deep learning-based three-dimensional medical image segmentation processing method according to claim 8, characterized in that: The step of reconstructing the filtered tissue boundary according to the starting position decomposition surface and the ending position decomposition surface to generate a plurality of reconstructed tissue boundaries includes: Acquire three-dimensional coordinate parameters of the starting position decomposition surface and the ending position decomposition surface, wherein the three-dimensional coordinate parameters include center point coordinates, normal vectors, and existing boundary point coordinates; Using ray tracing or a plane-curve intersection algorithm, respectively calculating the intersection points or intersection lines of the starting position decomposition surface and the ending position decomposition surface with the filtering tissue boundary, and segmenting the filtering tissue boundary based on the intersection points or intersection lines to generate a plurality of tissue boundary segments, each of which corresponds to a portion between a starting position decomposition surface and a ending position decomposition surface; Marking each tissue boundary segment, and recording the starting decomposition surface information and the ending decomposition surface information corresponding to the tissue boundary segment; Analyzing characteristic parameters of each tissue boundary segment, wherein the characteristic parameters include curvature, length, and connection relationship with other tissue boundary segments; For each tissue boundary segment, a preset reconstruction algorithm is executed according to the starting decomposition surface information and the ending decomposition surface information marked by the tissue boundary segment and the characteristic parameters of the tissue boundary segment to generate a plurality of reconstructed tissue boundaries.

10. A three-dimensional medical image segmentation processing system, characterized in that: It includes a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by the processor, the three-dimensional medical image segmentation processing method based on deep learning according to any one of claims 1 to 9 is implemented.

Images