Breast endoscope preoperative anatomical variation identification method and system

Abstract

The invention discloses a mammary gland endoscope preoperative anatomical variation identification method and system, and relates to the technical field of computer data processing. The method comprises the following steps: S1, acquiring heterogeneous source data, and performing spatial registration and data standardization fusion to generate a multi-dimensional fusion data set; s2, performing basic topological feature extraction based on the multi-dimensional fusion data set, and calculating edge weight information to construct a target topological graph; s3, matching and comparing the target topological graph with a preset standard topological database, and calculating topological compliance parameters based on comparison differences; s4, identifying a parameter abnormal region according to the spatial distribution characteristics of the topological compliance parameters, and generating a spatial security constraint model in combination with local spatial medium attributes; and S5, based on the spatial security constraint model, outputting structure abnormal information and corresponding optimized path planning parameters. And through multi-channel tensor quantization variation characteristics, safety constraints are constructed by using an anisotropic bounding box, and a path is optimized in combination with a potential field algorithm, so that high-precision automatic obstacle avoidance is realized.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, specifically to a method and system for identifying anatomical variations before breast endoscopic surgery. Background Technology

[0002] In many technical fields requiring precise manipulation of enclosed or complex structures, the safety and success rate of the procedure highly depend on a precise understanding of the three-dimensional topological features within the target structure. For example, in laparoscopic breast surgery, accurately identifying the anatomical structures and potential anatomical variations within the target area is crucial for reducing surgical risks and improving the safety of pathway planning. Furthermore, in practice, the internal tissue structure of the breast region is complex, and the distribution of blood vessels and ductal systems varies significantly among individuals.

[0003] Currently, the analysis of such complex structures largely relies on the correlation analysis of multi-source heterogeneous detection data. However, relying solely on discrete images of a single modality makes it difficult to achieve digital modeling of complex three-dimensional connected structures. Furthermore, due to the lack of an effective spatial quantitative evaluation mechanism, it is difficult to achieve automated identification and risk quantification of anatomical variation regions. When constructing auxiliary path planning models, if static anatomical data cannot be combined with dynamic spatial constraint logic, the generated planned paths often fail to avoid high-risk variation regions. This results in the operational planning process still relying heavily on human experience, making it difficult to systematically anticipate and avoid potential operational risks caused by structural irregularities.

[0004] Therefore, how to effectively integrate multi-source data in the digital space, automatically identify topological variations based on structural features, and then construct a planning model with spatial constraints has become an urgent problem to be solved in the field of data processing and assisted navigation. Summary of the Invention

[0005] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide a method and system for identifying anatomical variations before breast endoscopic surgery, so as to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying anatomical variations before breast endoscopic surgery, comprising: S1: Acquire at least two types of heterogeneous source data for the target region, perform spatial registration and data standardization fusion based on a preset global reference coordinate system, and generate a multidimensional fused dataset; S2: Based on the multi-dimensional fusion dataset, basic topological features are extracted, and edge weight information is calculated to construct a target topological graph representing the structural connectivity within the target region; S3: Match and compare the target topology map with the preset standard topology database, and calculate the topology compliance parameter that reflects the degree of structural deviation of the target area based on the comparison difference; S4: Identify abnormal regions of parameters based on the spatial distribution characteristics of topology compliance parameters, and generate a spatial safety constraint model with directional differences by combining local spatial medium properties; S5: Based on the spatial safety constraint model, output structural anomaly information and corresponding optimized path planning parameters.

[0007] The present invention is further configured such that the spatial registration in S1 specifically includes: The first static discrete data matrix of the same target object is obtained as the first mode and the second dynamic time-series data stream is obtained as the second mode through the data transmission interface. Parse the file header information of the first and second modes to extract the original resolution parameters and physical space mapping parameters; The weight allocation logic based on resolution numerical comparison is executed to generate corresponding data source confidence weight values ​​for the first and second modes respectively. Data with higher original resolution parameters are assigned higher source confidence weight values. Using the neighborhood numerical difference calculation rule, the numerical distribution of the data change rate of the first mode and the second mode in the three-dimensional logical space is calculated by traversing them respectively, and discrete coordinate points with change rate values ​​exceeding a preset threshold are selected to construct the first feature coordinate set and the second feature coordinate set. Based on the minimum distance iterative optimization strategy, the Euclidean distance between the first feature coordinate set and the second feature coordinate set is calculated, and the spatial transformation matrix that minimizes the Euclidean distance is used as the transformation operator for spatial dimension alignment.

[0008] The present invention is further configured such that the data standardization and fusion in S1 specifically includes: The second mode is mapped to the coordinate system of the first mode using a transformation operator, and numerical interpolation based on spatial neighborhood is performed on the mapped discrete data points to generate a normalized data volume. Construct a multidimensional composite data tensor, map the basic numerical attributes of the standardized data volume to the first dimension, map the numerical distribution of the data change rate to the second dimension, and map the corresponding source confidence weights to the third dimension to generate a multidimensional fused dataset.

[0009] The present invention is further configured such that the basic topological feature extraction in S2 specifically includes: Based on a multidimensional fusion dataset, continuous high-dimensional data that meets preset connectivity conditions is extracted by utilizing the numerical distribution features mapped to the first and second dimensions. Three-dimensional morphological thinning operations are performed on continuous high-dimensional data. While keeping the topological Euler number unchanged, boundary voxel points are iteratively removed to generate a set of central geometric skeletons of voxel width as the basic topological features of the target region.

[0010] The present invention is further configured such that the construction of the target topology graph in S2 specifically includes: Traverse each voxel in the basic topological features and calculate the number of connected components in the three-dimensional spatial neighborhood of each voxel. Mark voxels with a number of connected components greater than a preset threshold as branch nodes and voxels with a number of connected components equal to 1 as endpoint nodes. Tracing the skeleton path connecting each branch node and endpoint node generates a set of topological edges, and calculating the spatial geometric properties and medium density properties corresponding to each topological edge based on the multidimensional fusion dataset; Using branch nodes and endpoint nodes as graph nodes, and topological edge sets as graph connecting edges, the target topological graph is constructed by using spatial geometric properties and medium density properties as edge weight information.

[0011] The present invention is further configured such that the matching comparison in S3 specifically includes: A preset standard topology database is invoked, which contains a reference topology model generated based on statistics of a large number of historical samples; Using edge weight information as the similarity metric, subgraph isomorphic matching operations are performed on the target topology graph and the reference topology model to establish node mapping relationships; Based on node mapping relationships, redundant nodes that cannot be mapped in the target topology graph are identified and marked as topology variation components.

[0012] The present invention is further configured such that the computational topology compliance parameters in S3 specifically include: Based on the graph nodes with established mapping relationships in the target topology graph, calculate the spatial displacement vectors of the corresponding reference nodes in the reference topology model; Extract the local gradient magnitude of the corresponding graph node position from the second dimension of the multidimensional fusion dataset, and use it as the physical rigidity weight of the graph node; Calculate the product of the magnitude of the spatial displacement vector and the physical rigidity weight to generate the geometric strain energy component; Preset structural penalty weights are assigned to the topological variation components, and the structural penalty weights are weighted and fused with the geometric strain energy components to obtain the topological compliance parameters.

[0013] The present invention is further configured such that S4 specifically includes: The topological compliance parameters are mapped to the three-dimensional logical space corresponding to the multidimensional fusion dataset to construct the potential energy density distribution field. Traverse the potential energy density distribution field and define the spatial region where the value exceeds the preset deviation threshold as the parameter abnormal region; Based on the parameter anomaly region, the corresponding local gradient principal direction is extracted from the first and second dimensions of the multidimensional fusion dataset and used as the directional difference guiding vector. Centered on the region of parameter anomalies, spatial asymmetric expansion is performed along the direction of the directional difference guiding vector to generate an anisotropic bounding box set; By utilizing spatial set union operations and collision detection algorithms, the mutual exclusion influence range of each anisotropic bounding box in the three-dimensional logical space is calculated, and a spatial safety constraint model is constructed.

[0014] The present invention is further configured such that S5 specifically includes: Using the spatial security constraint model as an impassable constraint, a global energy potential field is constructed in the three-dimensional logical space with the cumulative integral of the topological compliance parameter as the cost function; Based on the preset starting coordinates and target coordinates, a heuristic search algorithm is used to retrieve the discrete point sequence with the minimum cost value in the global energy potential field, which is used as the original planned path. Curve fitting is performed on the original planned path to extract the coordinates of key control points and the corresponding directional difference guidance vectors, and to generate optimized path planning parameters. Extract the topological features and corresponding topological compliance parameters of the parameter anomaly regions and encapsulate them as structural anomaly information; Establish the correlation between structural anomaly information and optimized path planning parameters, and execute asynchronous output in the form of structured data reports.

[0015] The present invention also provides a preoperative anatomical variation identification system for breast endoscopic surgery, the system comprising: Multi-source data fusion module: acquires at least two types of heterogeneous source data for the target area, performs spatial registration and data standardization fusion based on a preset global reference coordinate system, and generates a multi-dimensional fused dataset; Topology modeling module: Based on a multi-dimensional fusion dataset, basic topological features are extracted, and edge weight information is calculated to construct a target topology graph representing the structural connectivity relationships within the target region; The variation quantification and comparison module matches and compares the target topology map with a preset standard topology database. Based on the comparison differences, it calculates the topology compliance parameter that reflects the degree of structural deviation of the target region. Risk Space Constraint Module: Identifies abnormal parameter regions based on the spatial distribution characteristics of topology compliance parameters, and generates a spatial safety constraint model with directional differences by combining local spatial medium properties; Intelligent planning output module: Based on the spatial safety constraint model, it outputs structural anomaly information and corresponding optimized path planning parameters.

[0016] This invention provides a method and system for identifying anatomical variations before breast laparoscopy. The method comprises: S1: acquiring at least two types of heterogeneous source data for a target region, performing spatial registration and data standardization fusion based on a preset global reference coordinate system to generate a multidimensional fusion dataset; S2: extracting basic topological features based on the multidimensional fusion dataset, calculating edge weight information to construct a target topological map representing the structural connectivity within the target region; S3: matching and comparing the target topological map with a preset standard topological database, calculating topological compliance parameters reflecting the degree of structural deviation in the target region based on the comparison differences; S4: identifying abnormal regions based on the spatial distribution characteristics of the topological compliance parameters, and generating a spatial safety constraint model with directional differences by combining local spatial medium properties; S5: outputting structural anomaly information and corresponding optimized path planning parameters based on the spatial safety constraint model. The beneficial effects include: Achieving high-precision semantic fusion and quantitative evaluation of complex heterogeneous data: By constructing a multi-channel composite tensor containing density, gradient, and confidence weights, discrete physical values ​​are transformed into topological features with spatial correlation. Furthermore, a topological compliance parameter is introduced to solve the problem that traditional methods cannot quantitatively describe the degree of variation in anatomical structures, thereby improving the objectivity and accuracy of variation identification.

[0017] Establish a dynamic spatial safety constraint model with biomechanical characteristics: construct an anisotropic bounding box using the principal direction of local gradients to simulate the spatial directional differences of real tissues, overcoming the shortcomings of traditional obstacle avoidance models that are singular and rigid; combine with spatial asymmetric expansion logic to enable the safety boundary to dynamically fit the actual shape of the variable structure, enhancing the safety redundancy of preoperative planning.

[0018] Provides data-driven global optimal path planning decision support: Transforms the results of anatomical mutation identification into a cost function in the global energy potential field, and automatically retrieves the optimal path that balances safety and efficiency through heuristic algorithms, fundamentally reducing the reliance on human experience in the operation process, and providing systematic risk avoidance solutions and standardized path parameters for refined operations in complex environments.

[0019] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a method for identifying anatomical variations before breast laparoscopy, as shown in an exemplary embodiment of the present invention; Figure 2 This is a schematic diagram of a preoperative anatomical variation identification system for breast endoscopic surgery, as shown in an exemplary embodiment of the present invention. Detailed Implementation

[0021] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0023] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0024] Example 1: A method for identifying anatomical variations before breast endoscopic surgery, such as... Figure 1 As shown, it includes: S1: Acquire at least two types of heterogeneous source data for the target region, perform spatial registration and data standardization fusion based on a preset global reference coordinate system, and generate a multidimensional fused dataset; S2: Based on the multi-dimensional fusion dataset, basic topological features are extracted, and edge weight information is calculated to construct a target topological graph representing the structural connectivity within the target region; S3: Match and compare the target topology map with the preset standard topology database, and calculate the topology compliance parameter that reflects the degree of structural deviation of the target area based on the comparison difference; S4: Identify abnormal regions of parameters based on the spatial distribution characteristics of topology compliance parameters, and generate a spatial safety constraint model with directional differences by combining local spatial medium properties; S5: Based on the spatial safety constraint model, output structural anomaly information and corresponding optimized path planning parameters.

[0025] The present invention is further configured such that the spatial registration in S1 specifically includes: The first static discrete data matrix of the same target object is obtained as the first mode and the second dynamic time-series data stream is obtained as the second mode through the data transmission interface. Parse the file header information of the first and second modes to extract the original resolution parameters and physical space mapping parameters; The weight allocation logic based on resolution numerical comparison is executed to generate corresponding data source confidence weight values ​​for the first and second modes respectively. Data with higher original resolution parameters are assigned higher source confidence weight values. Using the neighborhood numerical difference calculation rule, the numerical distribution of the data change rate of the first mode and the second mode in the three-dimensional logical space is calculated by traversing them respectively, and discrete coordinate points with change rate values ​​exceeding a preset threshold are selected to construct the first feature coordinate set and the second feature coordinate set. Based on a minimum distance iterative optimization strategy, the Euclidean distance between the first and second feature coordinate sets is calculated, and the spatial transformation matrix that minimizes the Euclidean distance is used as the transformation operator for spatial dimension alignment. Specifically, firstly, the first static discrete data matrix and the second dynamic temporal data stream of the target object are acquired in parallel through a data transmission interface; the first modality is high-resolution three-dimensional volume data, and the second modality is real-time acquired sequence data. For example, in breast laparoscopic surgery, the first modality is preoperative magnetic resonance imaging volume data, and the second modality is the real-time ultrasound scan sequence. Next, a metadata parsing algorithm is called to perform deep parsing of the file header tags of these two modalities, extracting the original resolution parameters and physical space mapping parameters, including: voxel spacing, slice thickness, coordinate origin and orientation matrix. Based on this, weight allocation logic is executed based on the extracted parameters, calculating the reciprocal of the effective resolution for each modality. Effective resolution is defined as the minimum voxel spacing. The reciprocals are then normalized to assign higher confidence weights to data sources with higher resolutions. For example, if the effective resolution of the first modality is 0.5 mm, its reciprocal is 2; if the effective resolution of the second modality is 0.3 mm, its reciprocal is approximately 3.33. After normalization, the weight of the first modality is approximately 0.375, and the weight of the second modality is approximately 0.625. Subsequently, the numerical distribution of the rate of change of data for both modalities in the three-dimensional logical space is calculated using the 3D Sobel operator. Feature points are extracted through adaptive thresholding and spatial clustering. The adaptive threshold is determined using Otsu's method or the 95th quantile method based on gradient histograms. Spatial clustering uses the density-based DBSCAN algorithm, merging feature point groups with a spatial distance of less than 2 mm and using the cluster centroid as the representative point. Finally, a first feature coordinate set and a second feature coordinate set are constructed. Finally, in the spatial registration stage, the Iterative Nearest Point Registration Strategy (ICP Registration Strategy) is adopted. With the centroid alignment of the feature point set as the initial condition, the nearest neighbor search, outlier removal, and rigid body transformation are performed through multiple rounds of iterative loops. The outlier removal adopts the three-standard-deviation principle, and the rigid body transformation is solved using the singular value decomposition method. The iteration termination condition is that the change of the transformation matrix is ​​less than five per thousand or the maximum number of iterations of one hundred is reached. Finally, the spatial transformation matrix that minimizes the Euclidean distance between the two feature coordinate sets is used as the final transformation operator for spatial dimension alignment, thus completing the spatial registration of multi-source heterogeneous data.

[0026] The present invention is further configured such that the data standardization and fusion in S1 specifically includes: The second mode is mapped to the coordinate system of the first mode using a transformation operator, and numerical interpolation based on spatial neighborhood is performed on the mapped discrete data points to generate a normalized data volume. A multidimensional composite data tensor is constructed, mapping the basic numerical attributes of the standardized data volume to the first dimension, the numerical distribution of the data change rate to the second dimension, and the corresponding source confidence weights to the third dimension, generating a multidimensional fused dataset. Specifically, firstly, the final transformation operator obtained in the spatial registration stage is applied to map each discrete data point of the second mode from its original coordinate system to the coordinate system of the first mode; the mapping process is completed through matrix multiplication, that is, multiplying a 4×4 homogeneous transformation matrix with the homogeneous coordinate vector of each data point to obtain its new spatial coordinates in the target coordinate system. Subsequently, numerical interpolation based on spatial neighborhood is performed on the second modality data points whose positions may no longer be regularly arranged after mapping to generate a standardized data volume with the same regular grid as the first modality. Specifically, a trilinear interpolation method is used. For each voxel position in the target regular grid, the eight nearest second modality mapped data points are found in its cubic neighborhood. The weighted average of these points' original values ​​and their distances relative to the target voxel center is calculated to assign an interpolated value to the voxel. The voxel resolution of the regular grid is uniformly set to 0.5 mm cubic, and all interpolated values ​​are normalized to zero mean and unit variance to finally generate a standardized data volume with the same spatial scale as the first modality. After constructing the standardized data volume, the generation of the multidimensional fused dataset begins. First, a four-dimensional composite data tensor is constructed, with the first three dimensions corresponding to the spatial coordinate axes and the fourth dimension representing the feature channels. The normalized grayscale value of each voxel in the standardized data volume is used as the basic numerical attribute and filled into the corresponding spatial position of the tensor's first channel. Next, based on the distribution of the basic numerical attribute, the three-dimensional Sobel operator is used to calculate the gradient components of each voxel position in three directions, and its magnitude is calculated as the data change rate value. This data change rate value is filled into the corresponding position of the tensor's second channel. Finally, the... In the spatial registration stage, the source confidence weight values ​​calculated for each data source are spatially broadcast. For voxel regions originating from the first modality, the third channel of its tensor is uniformly filled with the confidence weight value of the first modality, for example, a confidence weight of 0.375 obtained in the previous embodiment. For voxel regions originating from the interpolation data of the second modality, the confidence weight value of the second modality is uniformly filled, for example, a confidence weight of 0.625 obtained in the previous embodiment. Finally, a multidimensional fusion dataset integrating basic attributes, rate of change attributes, and reliability weight attributes is formed for use by the subsequent topological feature extraction module.

[0027] The present invention is further configured such that the basic topological feature extraction in S2 specifically includes: Based on a multidimensional fusion dataset, continuous high-dimensional data that meets preset connectivity conditions is extracted by utilizing the numerical distribution features mapped to the first and second dimensions. Three-dimensional morphological thinning operations are performed on continuous high-dimensional data. While maintaining the topological Euler number, boundary voxel points are iteratively removed to generate a set of central geometric skeletons of voxel widths as the basic topological features of the target region. Specifically, firstly, based on the multi-dimensional fusion dataset, continuous high-dimensional data volumes satisfying preset connectivity conditions are extracted by comprehensively utilizing the basic numerical attributes of the first channel and the numerical distribution of the data change rate of the second channel. Specifically, a dual threshold screening condition is set: the basic value of the first channel must be greater than a preset density threshold (e.g., the normalized value must be greater than the preset default density threshold of 0.2) to distinguish the target structure from the background region; simultaneously, the data change rate of the voxel position in the second channel must be greater than the average of its 8 neighbors to strengthen the structural edge region. Next, using the three-dimensional twenty-six-neighbor connectivity criterion, region growing clustering is performed on voxels that simultaneously satisfy the above dual conditions, aggregating spatially adjacent voxels with similar attributes into the same connected region; finally, continuous regions with a voxel size greater than one thousand are selected as the continuous high-dimensional data to be processed. These regions typically correspond to tubular structures such as blood vessels and ducts. After obtaining the continuous high-dimensional data volume, a three-dimensional morphological thinning operation is performed to extract the central geometric skeleton. This three-dimensional morphological thinning operation adopts the iterative thinning algorithm proposed by Paladin, which systematically removes boundary voxels while maintaining the 3D topological Euler number. Specifically, in each iteration, boundary voxels that meet the "simple point" condition are detected and removed sequentially from six main directions (up, down, left, right, front, and back). The determination of "simple point" requires that three conditions be met simultaneously: First, the removal of the voxel will not disrupt the connectivity of its 26 neighborhoods, that is, it will not create holes or disconnections in the original connected region; second, the removal of the voxel will not change the local topological type, which is verified by calculating the Euler eigenvalues ​​of the voxel configuration in its 3×3×3 neighborhood; third, the voxel is not an endpoint, that is, there are at least two voxels belonging to the same connected region in its 26 neighborhoods. After each iteration, the Euler number of the remaining voxel set is recalculated to ensure consistency with that before thinning. The above iterative thinning process is repeated until no more boundary voxels meet the removal conditions. At this point, the original continuous tubular data volume is gradually eroded, ultimately retaining a set of spatial curves representing the width of a single voxel, i.e., the central geometric skeleton. This skeleton is strictly located on the centerline of the original structure and completely preserves the topological connectivity and branching structure of the original data volume. Finally, the skeleton is post-processed, using morphological opening operations to remove short, burr branches caused by noise (branches shorter than five voxels are pruned), generating the final basic topological features, i.e., the central geometric skeleton set used to characterize the pipe connectivity within the target region.

[0028] The present invention is further configured such that the construction of the target topology graph in S2 specifically includes: Traverse each voxel in the basic topological features and calculate the number of connected components in the three-dimensional spatial neighborhood of each voxel. Mark voxels with a number of connected components greater than a preset threshold as branch nodes and voxels with a number of connected components equal to 1 as endpoint nodes. Tracing the skeleton path connecting each branch node and endpoint node generates a set of topological edges, and calculating the spatial geometric properties and medium density properties corresponding to each topological edge based on the multidimensional fusion dataset; Using bifurcation nodes and endpoint nodes as graph nodes, and the set of topological edges as graph connections, the system constructs a target topology graph by using spatial geometric properties and medium density properties as edge weights. Specifically, firstly, the system traverses each voxel in the input central geometric skeleton set (i.e., basic topological features). For each skeleton voxel, the system checks its 3×3×3 cubic neighborhood, identifies all voxels within that neighborhood that also belong to the skeleton, and performs connectivity analysis on these voxels based on the 26-neighborhood connectivity rule. The system calculates the number of independent connected components formed within the cube and records this value as the number of connected components for that central voxel. After analyzing all voxels, the system labels nodes based on the number of connected components: all points with a connected component count of 1 are classified as endpoint nodes; all points with a connected component count greater than or equal to 3 are classified as bifurcation nodes. To eliminate redundant bifurcations caused by minor irregularities on the skeleton surface, the system merges bifurcation nodes with a spatial distance of less than 2 millimeters, taking the average of their coordinates as a new single node. Subsequently, the system performs skeleton path tracing to generate a set of topological edges. This process employs a depth-first search strategy: starting from an unvisited endpoint or branch node, it explores forward along its connected skeleton path. During the exploration, the coordinates of all voxel points traversed by the path are recorded until another branch or endpoint node is reached, at which point a topological edge is discovered. The system records the starting and ending node numbers of this edge and stores the recorded ordered coordinate sequence as the spatial path of the edge. This process traverses all nodes, ensuring that each connecting path is recorded only once, ultimately generating a complete set of topological edges. Next, the system calculates the spatial geometric properties and medium density properties for each topological edge. For the length in the spatial geometric properties, the system accumulates the Euclidean distances between adjacent points in the ordered coordinate sequence corresponding to the edge to obtain the total length in millimeters. For the average curvature, the system uses the ordered coordinate sequence to calculate the local radius of curvature at each interior point on the path by defining a circle using three adjacent points, and then calculates the average of its reciprocals as the average curvature of the edge. For the medium density attribute, the system maps the ordered coordinate sequence corresponding to an edge back to the multidimensional fusion dataset. For each coordinate point in the sequence, the system obtains its basic numerical attribute (i.e., the normalized medium density value) in the first channel of the dataset through trilinear interpolation. Then, it calculates the average value of all interpolation points on the entire path as the medium density attribute of that edge.Finally, the system constructs the final target topology graph. All the branching nodes and endpoint nodes identified and deduplicated in the previous steps are used as the vertex set of the graph, and all the traced topological edges are used as the connecting edge set of the graph. The length, average curvature, and average medium density values ​​calculated for each edge are used together as the multidimensional weight vector of the connecting edge. The system integrates this vertex set, edge set, and edge weight vector to form a complete mathematical graph model with attributes, namely the target topology graph. The target topology accurately represents the topological connection relationship and local physical geometric characteristics of the tubular structure in the target area.

[0029] The present invention is further configured such that the matching comparison in S3 specifically includes: A preset standard topology database is invoked, which contains a reference topology model generated based on statistics of a large number of historical samples; Using edge weight information as the similarity metric, subgraph isomorphic matching operations are performed on the target topology graph and the reference topology model to establish node mapping relationships; Based on node mapping relationships, redundant nodes that cannot be mapped in the target topology graph are identified and marked as topological variation components. Specifically, firstly, the system calls a preset standard topology database, which is stored in the form of graph structure files. Each reference topology model file contains the following information: a graph structure consisting of reference nodes and reference edges, a standard weight vector for each reference edge (such as standard length and standard density range), and prior statistical spatial distribution of nodes. Based on the macroscopic features of the input target topology graph, the system intelligently selects one to three closest reference topology models from the database and loads them into memory as the current comparison benchmark. These macroscopic features include, for example, comparing the total number of nodes and the number of first-level branches. Next, the system performs subgraph isomorphic matching operations driven by edge weight information. The specific process is as follows: First, for each edge in the target topology graph and each reference edge in the reference model, the cosine similarity of their weight vectors is calculated to form an initial edge similarity matrix. Then, the VF2 algorithm or its improved version is used as the core matching engine. The VF2 algorithm starts from an initial, possibly empty, node mapping and iteratively attempts to pair a node in the target graph with a node in the reference model. In each attempt to expand the mapping, the algorithm not only checks the consistency of the topology (i.e., whether the connection relationships are compatible), but more importantly, it checks whether the weight similarity of the associated edges is higher than a preset threshold (e.g., requiring a cosine similarity greater than 0.85). Through this recursive search with dual constraints, one or more subgraph isomorphic mappings that meet the conditions are finally found. The system selects the mapping scheme with the most matched nodes and the highest cumulative edge similarity as the optimal node mapping relationship. Subsequently, based on the established optimal node mapping relationship, the system performs mutation identification. The system traverses all nodes in the target topology graph. For each node, it checks if it exists in the "key" set of the aforementioned mapping relationship. If it exists, it means the node has successfully matched a standard reference node and belongs to the normal structure; if it does not exist, the node is marked as a "redundant node." The system further uses these redundant nodes as seed points and utilizes the graph's connectivity to completely extract their directly connected edges and other local subgraphs formed only by redundant nodes. Each such independently connected maximum subgraph composed of redundant nodes and their associated edges is defined and marked as an independent "topological mutation component." Finally, the system records the specific nodes and edges contained in each topological mutation component, providing basic information such as its spatial location and scale for subsequent analysis. At this point, the matching and comparison process is complete, successfully separating the parts of the target structure that conform to the standard model from the abnormal parts representing mutations.

[0030] The present invention is further configured such that the computational topology compliance parameters in S3 specifically include: Based on the graph nodes with established mapping relationships in the target topology graph, calculate the spatial displacement vectors of the corresponding reference nodes in the reference topology model; Extract the local gradient magnitude of the corresponding graph node position from the second dimension of the multidimensional fusion dataset, and use it as the physical rigidity weight of the graph node; Calculate the product of the magnitude of the spatial displacement vector and the physical rigidity weight to generate the geometric strain energy component; Preset structural penalty weights are assigned to the topological variation components, and these weights are then weighted and fused with the geometric strain energy components to obtain the topological compliance parameters. Specifically, firstly, based on the established optimal node mapping relationship, for each mapped graph node in the target topology graph, the spatial displacement vector between its spatial coordinates and the coordinates of the corresponding reference node in the reference topology model is calculated. The specific calculation method is to take the difference between two points on each coordinate axis to form a vector, and then obtain the Euclidean distance of this vector as the displacement modulus by calculating the square root of the sum of squares of each component. At the same time, the coordinates of each graph node are mapped back to the multidimensional fused dataset. From the numerical distribution of the rate of change of the second dimension data, a third-order cubic neighborhood is taken with the coordinate as the center. The local gradient modulus of the point is accurately calculated by trilinear interpolation, and this gradient modulus is directly defined as the physical stiffness weight of the node. Then, the geometric strain energy component corresponding to each node is calculated, and its value is the product of the displacement modulus and the physical stiffness weight. After traversing all mapped nodes, the arithmetic mean of these geometric strain energy components is calculated as the average energy estimate of the overall geometric deformation. Next, a preset structural penalty weight is assigned to each topological variant component identified in the matching and comparison phase. The assignment rule uses a clear calculation formula based on the component type and scale: for redundant branch variants, the base penalty weight is 1.0 and increases linearly with the number of edges it contains, specifically calculated as: penalty weight = 1.0 + 0.1 × number of edges in the branch; for missing branch variants, a fixed penalty value of 2.5 is uniformly assigned. The geometric mean of the penalty weights of all variant components is calculated as the overall penalty coefficient for topological anomalies. The geometric mean is calculated by multiplying all penalty weights and taking the Mth root, where M is the total number of variant components. Finally, weighted fusion is performed, multiplying the overall average geometric strain energy by the overall topological penalty coefficient to obtain the final topological compliance parameters. Simultaneously, the system generates a refined parameter spatial distribution: for mapped node regions, if the node belongs to a certain variation component, its local parameter value is determined by the product of the strain energy component at that point and the penalty weight of that variation component; if it does not belong to any variation component, it is determined by the product of the strain energy component at that point and the overall penalty coefficient; for unmapped pure variation regions, a parameter value equal to the penalty weight of that variation component is directly assigned. Thus, the system outputs a comprehensive quantitative topological compliance parameter and its spatial distribution field that quantifies the degree of deviation between the overall and local structures.

[0031] The present invention is further configured such that S4 specifically includes: The topological compliance parameters are mapped to the three-dimensional logical space corresponding to the multidimensional fusion dataset to construct the potential energy density distribution field. Traverse the potential energy density distribution field and define the spatial region where the value exceeds the preset deviation threshold as the parameter abnormal region; Based on the parameter anomaly region, the corresponding local gradient principal direction is extracted from the first and second dimensions of the multidimensional fusion dataset and used as the directional difference guiding vector. Centered on the region of parameter anomalies, spatial asymmetric expansion is performed along the direction of the directional difference guiding vector to generate an anisotropic bounding box set; By utilizing spatial set union operations and collision detection algorithms, the mutually exclusive influence range of each anisotropic bounding box in the three-dimensional logical space is calculated, constructing a spatial safety constraint model. Specifically, firstly, the calculated topological compliance parameter values ​​are mapped to the three-dimensional logical space where the multidimensional fusion dataset resides based on their corresponding spatial coordinates. Each voxel location is assigned a topological compliance parameter value, thereby constructing a continuous three-dimensional potential energy density distribution field. Here, the topological compliance parameter value represents the degree of structural deviation risk at the voxel location. Next, each voxel in the three-dimensional potential energy density distribution field is traversed, and voxels whose topological compliance parameter values ​​exceed a preset deviation threshold are marked as outliers. The preset deviation threshold is obtained by adding twice the standard deviation to the average value of all voxel parameter values ​​measured in the initial experimental phase. Spatially adjacent outliers are clustered to form one or more continuous parameter anomalous regions. Each region must contain more than 50 voxels to filter noise. Then, based on each parameter anomaly region, local gradient information is extracted from the first and second dimensions of the multidimensional fusion dataset to determine the directional difference guidance vector. Specifically, within each anomaly region, the gradient vectors of all voxels in the first and second dimensions are calculated, and their average gradient vectors are obtained. These two average gradient vectors are then weighted and synthesized, with the weights being the confidence weights of the corresponding dimensions. After normalization, the synthesized vector yields the directional difference guidance vector for that anomaly region, indicating the main orientation of the local tissue structure. Subsequently, with each parameter anomaly region as the center, spatial asymmetric expansion is performed along its directional difference guidance vector to generate an anisotropic bounding box. The expansion rules are: an expansion distance of 10 mm along the direction of the directional difference guidance vector, an expansion distance of 5 mm in the reverse direction, and an expansion distance of 3 mm in each of the two orthogonal directions perpendicular to the vector, thus forming an ellipsoidal bounding box centered on the anomaly region with its major axis along the guidance vector direction. After processing all anomaly regions, an anisotropic bounding box set is formed. Finally, spatial set union and collision detection algorithms are used to process the anisotropic bounding box set to construct a spatial safety constraint model. The specific process is as follows: First, bounding boxes with overlapping spatial positions or a distance of less than 2 mm are merged to form a larger connected constraint region; then, collision detection is performed on the merged bounding box set to calculate the minimum interval distance between each bounding box. If the interval distance is less than 5 mm, these bounding boxes are regarded as a whole constraint region; finally, all constraint regions are converted into binary mask data in three-dimensional logical space, where voxels with a mask value of one represent impassable regions and voxels with a mask value of zero represent safe passage regions, thus completing the construction of the spatial safety constraint model.

[0032] The present invention is further configured such that S5 specifically includes: Using the spatial security constraint model as an impassable constraint, a global energy potential field is constructed in the three-dimensional logical space with the cumulative integral of the topological compliance parameter as the cost function; Based on the preset starting coordinates and target coordinates, a heuristic search algorithm is used to retrieve the discrete point sequence with the minimum cost value in the global energy potential field, which is used as the original planned path. Curve fitting is performed on the original planned path to extract the coordinates of key control points and the corresponding directional difference guidance vectors, and to generate optimized path planning parameters. Extract the topological features and corresponding topological compliance parameters of the parameter anomaly regions and encapsulate them as structural anomaly information; Establish the correlation between structural anomaly information and optimized path planning parameters, and execute asynchronous output in the form of structured data reports. Specifically, firstly, the spatial safety constraint model is used as an impassable constraint. A global energy potential field is constructed in the three-dimensional logical space with the cumulative integral of the topology compliance parameter as the cost function. The specific construction method is as follows: traverse each voxel in the three-dimensional logical space. If the voxel is marked as an impassable region in the spatial safety constraint model, its cost is set to a maximum value of 1,000,000; otherwise, read the topology compliance parameter value corresponding to the voxel position and use this parameter value as the basic cost value. To smooth the potential field, the basic cost value is processed by a three-dimensional Gaussian filter using a third-order cubic kernel function with a standard deviation of 0.5 mm. Finally, a continuous and smooth global energy potential field is generated, where the energy value of each voxel position represents the cost of passing through that location. Next, based on the preset starting and target coordinates, the A* heuristic search algorithm is used to retrieve the discrete point sequence with the minimum cost value in the global energy potential field. The search process is conducted in a voxel grid using a 26-neighborhood connection method. Starting from the voxel containing the starting coordinates, each iteration selects the node with the minimum sum of cost value and heuristic function value from the open list for expansion. The heuristic function uses Euclidean distance to calculate the straight-line distance from the current node to the target coordinates. When the search reaches the voxel containing the target coordinates, the parent node pointer is backtracked to obtain an original planned path composed of the voxel center coordinates. Then, curve fitting is performed on the original planned path using a cubic B-spline curve fitting algorithm, with the original path points as control points to generate a smooth spatial curve. Ten points are sampled at equal intervals along the fitted curve as key control points. For each key control point, gradient information is extracted from the first and second dimensions of the multidimensional fusion dataset within a sphere with a radius of three millimeters around it. A weighted average gradient vector is calculated, and after normalization, the directional difference guidance vector for that point is obtained. Finally, optimized path planning parameters containing the key control point coordinate sequence and corresponding direction vectors are generated. Simultaneously, the system extracts complete annotation information for parameter anomaly regions. For each parameter anomaly region, it calculates the geometric center coordinates as the mutation location and determines the mutation type based on the corresponding topological mutation component type: if it is a redundant branch, it is marked as "structural hyperplasia"; if it is a missing branch, it is marked as "structural defect"; if it is a morphological distortion, it is marked as "trajectory anomaly". The average value of the topological compliance parameter in the region is mapped to a severity level: a topological compliance parameter value less than 1.0 is mild, 1.0 to 2.0 is moderate, and greater than 2.0 is severe. The mutation location, type, and severity of all anomaly regions are encapsulated into structured annotation results.Next, a visualized risk distribution map is generated. Based on the three-dimensional logical space, the topological compliance parameter values ​​at each voxel location are converted to RGB color values ​​through color mapping: parameter values ​​from 0 to 1 are mapped from dark blue to light blue, 1 to 2 from yellow to orange, and values ​​greater than 2 from red to dark red. This color mapping result is overlaid and fused with the first-dimensional base image data of the multidimensional fusion dataset at 30% transparency to generate a three-dimensional risk heatmap. On this heatmap, the planned path curve is displayed in bright green, and the boundary contours of abnormal regions for each parameter are marked with purple borders. Finally, the correlation between the annotation results and the planning parameters is established. In the structured annotation results, the nearest critical control point number is added to each abnormal region. In the optimized path planning parameters, an abnormal region identifier within a 5 mm radius is added to each critical control point. The complete annotation results, optimized path planning parameters, and the three-dimensional rendering data of the risk distribution map are packaged into a standardized data package. The annotation results are stored in JSON format, the planning parameters are stored in CSV format, and the risk distribution map is stored in DICOM Supplement 172 structured report format. The system uses an asynchronous transmission protocol to simultaneously output data packets to the surgical planning system database and the visualization terminal, completing the entire processing flow.

[0033] Example 2: Please refer to Figure 2 This exemplary preoperative anatomical variation identification system for breast endoscopic surgery includes: Multi-source data fusion module: acquires at least two types of heterogeneous source data for the target area, performs spatial registration and data standardization fusion based on a preset global reference coordinate system, and generates a multi-dimensional fused dataset; Topology modeling module: Based on a multi-dimensional fusion dataset, basic topological features are extracted, and edge weight information is calculated to construct a target topology graph representing the structural connectivity relationships within the target region; The variation quantification and comparison module matches and compares the target topology map with a preset standard topology database. Based on the comparison differences, it calculates the topology compliance parameter that reflects the degree of structural deviation of the target region. Risk Space Constraint Module: Identifies abnormal parameter regions based on the spatial distribution characteristics of topology compliance parameters, and generates a spatial safety constraint model with directional differences by combining local spatial medium properties; Intelligent planning output module: Based on the spatial safety constraint model, it outputs structural anomaly information and corresponding optimized path planning parameters.

[0034] It should be noted that the preoperative anatomical variation identification system for breast endoscopic surgery provided in the above embodiments and the preoperative anatomical variation identification method for breast endoscopic surgery provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the preoperative anatomical variation identification system for breast endoscopic surgery provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0035] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying anatomical variations before laparoscopic breast surgery, characterized in that, include: S1: Acquire at least two types of heterogeneous source data for the target region, perform spatial registration and data standardization fusion based on a preset global reference coordinate system, and generate a multidimensional fused dataset; S2: Based on the multi-dimensional fusion dataset, basic topological features are extracted, and edge weight information is calculated to construct a target topological graph representing the structural connectivity within the target region; S3: Match and compare the target topology map with the preset standard topology database, and calculate the topology compliance parameter that reflects the degree of structural deviation of the target area based on the comparison difference; S4: Identify abnormal regions of parameters based on the spatial distribution characteristics of topology compliance parameters, and generate a spatial safety constraint model with directional differences by combining local spatial medium properties; S5: Based on the spatial safety constraint model, output structural anomaly information and corresponding optimized path planning parameters.

2. The method for identifying anatomical variations before breast endoscopic surgery according to claim 1, characterized in that, The spatial registration in S1 specifically includes: The first static discrete data matrix of the same target object is obtained as the first mode and the second dynamic time-series data stream is obtained as the second mode through the data transmission interface. Parse the file header information of the first and second modes to extract the original resolution parameters and physical space mapping parameters; The weight allocation logic based on resolution numerical comparison is executed to generate corresponding data source confidence weight values ​​for the first and second modes respectively. Data with higher original resolution parameters are assigned higher source confidence weight values. Using the neighborhood numerical difference calculation rule, the numerical distribution of the data change rate of the first mode and the second mode in the three-dimensional logical space is calculated by traversing them respectively, and discrete coordinate points with change rate values ​​exceeding a preset threshold are selected to construct the first feature coordinate set and the second feature coordinate set. Based on the minimum distance iterative optimization strategy, the Euclidean distance between the first feature coordinate set and the second feature coordinate set is calculated, and the spatial transformation matrix that minimizes the Euclidean distance is used as the transformation operator for spatial dimension alignment.

3. The method for identifying anatomical variations before breast endoscopic surgery according to claim 2, characterized in that, The data standardization and fusion in S1 specifically includes: The second mode is mapped to the coordinate system of the first mode using a transformation operator, and numerical interpolation based on spatial neighborhood is performed on the mapped discrete data points to generate a normalized data volume. Construct a multidimensional composite data tensor, map the basic numerical attributes of the standardized data volume to the first dimension, map the numerical distribution of the data change rate to the second dimension, and map the corresponding source confidence weights to the third dimension to generate a multidimensional fused dataset.

4. The method for identifying anatomical variations before breast endoscopic surgery according to claim 1, characterized in that, The basic topological feature extraction in S2 specifically includes: Based on a multidimensional fusion dataset, continuous high-dimensional data that meets preset connectivity conditions is extracted by utilizing the numerical distribution features mapped to the first and second dimensions. Three-dimensional morphological thinning operations are performed on continuous high-dimensional data. While keeping the topological Euler number unchanged, boundary voxel points are iteratively removed to generate a set of central geometric skeletons of voxel width as the basic topological features of the target region.

5. The method for identifying anatomical variations before breast endoscopic surgery according to claim 4, characterized in that, The target topology graph to be constructed in S2 specifically includes: Traverse each voxel in the basic topological features and calculate the number of connected components in the three-dimensional spatial neighborhood of each voxel. Mark voxels with a number of connected components greater than a preset threshold as branch nodes and voxels with a number of connected components equal to 1 as endpoint nodes. Tracing the skeleton path connecting each branch node and endpoint node generates a set of topological edges, and calculating the spatial geometric properties and medium density properties corresponding to each topological edge based on the multidimensional fusion dataset; Using branch nodes and endpoint nodes as graph nodes, and topological edge sets as graph connecting edges, the target topological graph is constructed by using spatial geometric properties and medium density properties as edge weight information.

6. The method for identifying anatomical variations before breast endoscopic surgery according to claim 1, characterized in that, The matching comparison in S3 specifically includes: A preset standard topology database is invoked, which contains a reference topology model generated based on statistics of a large number of historical samples; Using edge weight information as the similarity metric, subgraph isomorphic matching operations are performed on the target topology graph and the reference topology model to establish node mapping relationships; Based on node mapping relationships, redundant nodes that cannot be mapped in the target topology graph are identified and marked as topology variation components.

7. The method for identifying anatomical variations before breast endoscopic surgery according to claim 6, characterized in that, The calculation of topology compliance parameters in S3 specifically includes: Based on the graph nodes with established mapping relationships in the target topology graph, calculate the spatial displacement vectors of the corresponding reference nodes in the reference topology model; Extract the local gradient magnitude of the corresponding graph node position from the second dimension of the multidimensional fusion dataset, and use it as the physical rigidity weight of the graph node; Calculate the product of the magnitude of the spatial displacement vector and the physical rigidity weight to generate the geometric strain energy component; Preset structural penalty weights are assigned to the topological variation components, and the structural penalty weights are weighted and fused with the geometric strain energy components to obtain the topological compliance parameters.

8. The method for identifying anatomical variations before breast endoscopic surgery according to claim 1, characterized in that, S4 specifically includes: The topological compliance parameters are mapped to the three-dimensional logical space corresponding to the multidimensional fusion dataset to construct the potential energy density distribution field. Traverse the potential energy density distribution field and define the spatial region where the value exceeds the preset deviation threshold as the parameter abnormal region; Based on the parameter anomaly region, the corresponding local gradient principal direction is extracted from the first and second dimensions of the multidimensional fusion dataset and used as the directional difference guiding vector. Centered on the region of parameter anomalies, spatial asymmetric expansion is performed along the direction of the directional difference guiding vector to generate an anisotropic bounding box set; By utilizing spatial set union operations and collision detection algorithms, the mutual exclusion influence range of each anisotropic bounding box in the three-dimensional logical space is calculated, and a spatial safety constraint model is constructed.

9. The method for identifying anatomical variations before breast endoscopic surgery according to claim 1, characterized in that, S5 specifically includes: Using the spatial security constraint model as an impassable constraint, a global energy potential field is constructed in the three-dimensional logical space with the cumulative integral of the topological compliance parameter as the cost function; Based on the preset starting coordinates and target coordinates, a heuristic search algorithm is used to retrieve the discrete point sequence with the minimum cost value in the global energy potential field, which is used as the original planned path. Curve fitting is performed on the original planned path to extract the coordinates of key control points and the corresponding directional difference guidance vectors, and to generate optimized path planning parameters. Extract the topological features and corresponding topological compliance parameters of the parameter anomaly regions and encapsulate them as structural anomaly information; Establish the correlation between structural anomaly information and optimized path planning parameters, and execute asynchronous output in the form of structured data reports.

10. A preoperative anatomical variation identification system for breast endoscopic surgery, used to implement the preoperative anatomical variation identification method for breast endoscopic surgery as described in any one of claims 1-9, characterized in that, include: Multi-source data fusion module: acquires at least two types of heterogeneous source data for the target area, performs spatial registration and data standardization fusion based on a preset global reference coordinate system, and generates a multi-dimensional fused dataset; Topology modeling module: Based on a multi-dimensional fusion dataset, basic topological features are extracted, and edge weight information is calculated to construct a target topology graph representing the structural connectivity relationships within the target region; The variation quantification and comparison module matches and compares the target topology map with a preset standard topology database. Based on the comparison differences, it calculates the topology compliance parameter that reflects the degree of structural deviation of the target region. Risk Space Constraint Module: Identifies abnormal parameter regions based on the spatial distribution characteristics of topology compliance parameters, and generates a spatial safety constraint model with directional differences by combining local spatial medium properties; Intelligent planning output module: Based on the spatial safety constraint model, it outputs structural anomaly information and corresponding optimized path planning parameters.

Images