A strabismus three-dimensional mapping method and system based on a knowledge graph

Abstract

The application relates to the technical field of knowledge graph, and particularly provides a strabismus three-dimensional mapping method and system based on a knowledge graph, which comprises the following steps: taking a starting point space coordinate, a patient corneal reflection point offset and an eyeball rotation center coordinate as attribute values of an entity node, establishing a directed connection edge between nodes according to a fixed attachment relationship between muscles and eyeballs and between muscles and bone walls in anatomy, and constructing an initial heterogeneous graph by all entity nodes and the directed connection edge; weighting and aggregating attribute values of neighborhood nodes connected by the directed connection edge of each node in the obtained initial heterogeneous graph to obtain an aggregated feature vector of the node; and mapping all node coordinates in each obtained sub-graph to a unified three-dimensional coordinate system, and adjusting the relative positions between the sub-graphs so that all positioning base points are located on a plane. The application realizes systematic mapping and pathological posture analysis of three-dimensional space configurations of eye external muscles of a strabismus patient.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a 3D mapping method and system for strabismus based on knowledge graphs. Background Technology

[0002] Strabismus is an ophthalmic disorder caused by the misalignment of the visual axes of both eyes. Its pathological basis usually involves abnormal spatial attachment of one or more extraocular muscles, changes in the curvature of the muscle belly, and a mismatch between the center of eye rotation and the direction of muscle tension. In clinical diagnosis and surgical planning, it is necessary to accurately map the three-dimensional spatial configuration of the patient's extraocular muscles to quantitatively assess the degree of abnormality and guide treatment.

[0003] Existing strabismus mapping methods mainly fall into the following categories: The first category is the method of measuring extraocular muscle attachment points based on single-angle or double-angle images. This method manually marks the projection positions of the extraocular muscles on the sclera and bony walls from two-dimensional images, and then uses a pre-set eyeball model to infer the three-dimensional coordinates. During the measurement process, the influence of corneal light-reflection point offset on the attachment point position is usually ignored, and the coordinates of the eyeball rotation center are not included in a unified spatial reference transformation. The measurement of the starting point spatial coordinates is not corrected for corneal light-reflection point offset, resulting in systematic deviations in the positioning of attachment points under actual eye position; the coordinates of the eyeball rotation center are not included in a unified spatial reference system, resulting in a lack of consistent rotational references for the relative positions of different extraocular muscles. The second category is the mechanical inversion method based on eye movement tracking. This method records the patient's eyeball rotation angle under different gaze directions and uses an extraocular muscle mechanical model to inversely calculate the estimated values ​​of muscle attachment points or tension parameters; however, it relies on pre-set initial values ​​of muscle parameters and cannot directly obtain the actual spatial path of the muscle belly midline. The estimation results of muscle parameters heavily rely on the assumptions of the mechanical model and cannot independently verify the authenticity of the path points of the muscle belly midline; the spatial configuration differences between different patients or different extraocular muscles lack quantitative and automatically classifiable criteria. The third category is three-dimensional reconstruction methods based on individual orbital images. This method segments the extraocular muscles and orbital bone walls from CT or MRI data and reconstructs the three-dimensional morphology of the muscles. This method requires additional tomographic imaging equipment, and the reconstructed muscle structure exists in the form of discrete point clouds or meshes, lacking explicit expression of the anatomical attachment relationships between muscles and the eyeball, and between muscles and the bone walls, making it difficult to directly use for subsequent spatial configuration classification and overall pose analysis. The reconstructed point cloud or mesh structure does not establish directed connections with the fixed attachment relationships in anatomy, and cannot form an atlas structure with mechanical transmission direction; the mapping results are usually output as the coordinates or angles of individual muscles, lacking a comprehensive description of the deflection angle and centroid offset distance of the entire extraocular muscle group in the spatial plane. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: The technical problem to be solved by this invention is: how to integrate the spatial coordinates of the origin, the spatial coordinates of the insertion, the path point of the muscle belly midline, the offset of the corneal light-reflecting point, and the coordinates of the eyeball rotation center with the anatomical fixation and attachment relationship of the extraocular muscles of strabismic patients into an initial heterogeneous atlas; how to automatically divide the sub-atlas with similar spatial configurations based on the aggregated feature vectors of the nodes in the atlas; and how to adjust the positioning base points of the sub-atlas to be coplanar, and output the spatial deflection angle and centroid offset distance of the plane relative to the standard eye position reference plane, thereby realizing the systematic mapping of the three-dimensional spatial configuration of the extraocular muscles of strabismic patients.

[0005] One aspect of the present invention provides a method for 3D mapping of strabismus based on knowledge graphs, comprising the following steps: Multi-angle image sequences of strabismus patients were acquired, and the origin spatial coordinates, insertion spatial coordinates, and midline path points of the muscle belly of each extraocular muscle were extracted. The origin spatial coordinates, the offset of the corneal light reflection point, and the coordinates of the eyeball rotation center were used as the attribute values ​​of the entity nodes. Based on the fixed attachment relationship between muscles and eyeballs and between muscles and bone walls in anatomy, directed connection edges between nodes were established. All entity nodes and directed connection edges constituted an initial heterogeneous atlas. For each node in the initial heterogeneous graph, the attribute values ​​of the neighboring nodes connected by the directed edges are weighted and aggregated to obtain the aggregated feature vector of the node. The nodes are then hierarchically divided according to the Euclidean distance between the aggregated feature vectors of all nodes. Nodes with an Euclidean distance less than a preset threshold are grouped into the same subgraph, so that the initial heterogeneous graph is divided into several non-overlapping subgraphs, and each subgraph corresponds to a type of extraocular muscle spatial configuration. The coordinates of all nodes in each sub-map are mapped to a unified three-dimensional coordinate system. The average spatial coordinates of all nodes in each sub-map are taken as the positioning base point of the sub-map. The normal deflection angle of each positioning base point relative to the horizontal and sagittal planes of the eyeball is calculated. The relative positions between the sub-maps are adjusted so that all positioning base points are in the same plane. The spatial deflection angle of the plane relative to the standard eye position reference plane and the centroid offset distance are used as outputs to obtain the three-dimensional mapping results of strabismus.

[0006] Another aspect of the present invention provides a knowledge graph-based 3D mapping system for strabismus, comprising: The initial heterogeneous atlas construction module is responsible for acquiring multi-angle image sequences of strabismus patients' eyes, extracting the origin spatial coordinates, insertion spatial coordinates, and muscle belly midline path points of each extraocular muscle; using the origin spatial coordinates, the offset of the patient's corneal light reflection point, and the coordinates of the eyeball rotation center as attribute values ​​of the entity nodes; establishing directed connection edges between nodes based on the fixed attachment relationships between muscles and eyeballs, and muscles and bone walls in anatomy; and all entity nodes and directed connection edges constitute the initial heterogeneous atlas. The graph segmentation module is responsible for weighting and aggregating the attribute values ​​of the neighboring nodes connected by the directed edges of each node in the initial heterogeneous graph to obtain the aggregated feature vector of the node; and dividing the nodes into hierarchical levels according to the Euclidean distance between the aggregated feature vectors of all nodes, and classifying the nodes with Euclidean distance less than a preset threshold into the same subgraph, so that the initial heterogeneous graph is divided into several non-overlapping subgraphs, each subgraph corresponding to a type of extraocular muscle spatial configuration. The positioning base point module is responsible for mapping the coordinates of all nodes in each sub-map to a unified three-dimensional coordinate system. It takes the average spatial coordinates of all nodes in each sub-map as the positioning base point of the sub-map, calculates the normal deflection angle of each positioning base point relative to the horizontal and sagittal planes of the eyeball anatomy, and adjusts the relative positions between sub-maps so that all positioning base points are in the same plane. Then, it outputs the spatial deflection angle of the plane relative to the standard eye position reference plane and the centroid offset distance to obtain the strabismus three-dimensional mapping result.

[0007] This invention extracts the spatial coordinates of extraocular muscles from multi-angle image sequences and constructs an initial heterogeneous atlas by combining corneal light-reflection point offset and eye rotation center coordinates. This transforms anatomical attachment relationships into directed connections between nodes, achieving a correlation modeling between extraocular muscle structure and eye movement mechanisms. Subsequently, node feature vectors are generated through weighted aggregation of neighboring node attributes. Hierarchical subgraph partitioning based on Euclidean distance automatically categorizes extraocular muscles with similar spatial configurations, forming a subgraph system that unifies anatomy and function. Finally, through coordinate system mapping and planar calibration of positioning base points, the spatial deflection angle and centroid offset of the subgraphs relative to the standard eye position reference plane are quantified, thereby completing a systematic mapping of the three-dimensional spatial configuration of extraocular muscles in strabismus patients and an analysis of their pathological posture. Attached Figure Description

[0008] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the knowledge graph-based 3D mapping method for strabismus provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the knowledge graph-based 3D mapping method for strabismus provided in Embodiment 1 of the present invention. Figure 3 This is a diagram illustrating the process of forming an initial heterogeneous graph from all entity nodes and directed connecting edges in Embodiment 2 of the present invention. Figure 4 This is a diagram illustrating the process of hierarchically dividing nodes based on the Euclidean distance between the aggregated feature vectors of all nodes, as provided in Embodiment 4 of the present invention. Figure 5This is a process diagram of obtaining the strabismus three-dimensional mapping result provided in Embodiment 7 of the present invention; Figure 6 This is a block diagram of the knowledge graph-based 3D mapping system for strabismus provided in Embodiment 10 of the present invention; Figure 7 A block diagram of the electronic device provided by the present invention; Figure 8 A block diagram of a computer-readable storage medium provided for this invention.

[0009] Reference numerals: 1. Initial heterogeneous map construction module; 2. Map segmentation module; 3. Locating base point module; 4. Central processing unit / microprocessor / main control chip; 5. Storage medium; 6. Data bus; 7. Input / output bus / external bus / device bus; 8. Display; 9. Input / output device; 10. Computer-readable instructions; 11. Non-transitory computer-readable storage medium. Detailed Implementation

[0010] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0011] Hereinafter, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0012] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components; it can also be understood as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in a non-contact manner, such as an electrical connection between two components using capacitive coupling to transmit electrical signals.

[0013] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation of the components in the accompanying drawings.

[0014] Example 1: As Figure 1 As shown, this embodiment of the invention provides a knowledge graph-based 3D mapping method for strabismus, comprising the following steps: Step S100: Acquire multi-angle image sequences of the eyes of strabismus patients, extract the origin spatial coordinates, insertion spatial coordinates, and midline path points of the muscle belly of each extraocular muscle; use the origin spatial coordinates, the offset of the corneal light reflection point, and the coordinates of the eyeball rotation center as attribute values ​​of the entity nodes; establish directed connection edges between nodes based on the fixed attachment relationships between muscles and eyeballs and between muscles and bone walls in anatomy; all entity nodes and directed connection edges constitute the initial heterogeneous atlas. Step S200: For each node in the obtained initial heterogeneous graph, the attribute values ​​of the neighboring nodes connected by its directed edges are weighted and aggregated to obtain the aggregated feature vector of the node; the nodes are hierarchically divided according to the Euclidean distance between the aggregated feature vectors of all nodes, and nodes with Euclidean distance less than a preset threshold are grouped into the same subgraph, so that the initial heterogeneous graph is divided into several non-overlapping subgraphs, and each subgraph corresponds to a type of extraocular muscle spatial configuration; Step S300: Map all node coordinates in each sub-map to a unified three-dimensional coordinate system. Take the average spatial coordinates of all nodes in each sub-map as the positioning base point of the sub-map. Calculate the normal deflection angle of each positioning base point relative to the horizontal and sagittal planes of the eyeball anatomy. Adjust the relative positions between sub-maps so that all positioning base points are in the same plane. Then, use the spatial deflection angle of the plane relative to the standard eye position reference plane and the centroid offset distance as the output to obtain the strabismus three-dimensional mapping result.

[0015] The multi-angle image sequence refers to a set of eye images acquired around the patient's head at different azimuth and elevation angles. Each angle corresponds to a set of light projection directions, used to eliminate the overlapping and obstruction of the spatial positions of extraocular muscles in single-angle imaging, so that the origin, insertion, and muscle belly path of each extraocular muscle can be completely observed from different perspectives. The origin spatial coordinates mark the boundary point of each extraocular muscle's attachment area on the scleral surface of the eyeball closest to the equator in three-dimensional space. These coordinates are used to determine the fixed end of the muscle that exerts tension on the eyeball. The insertion spatial coordinates mark the boundary point of each extraocular muscle's attachment area on the orbital bone wall surface closest to the orbital apex in three-dimensional space. These coordinates are used to determine the connection anchor point between the other end of the muscle and the bone wall. The muscle belly central axis path points connect the origin and insertion points of the same extraocular muscle, and are a set of three-dimensional points acquired sequentially along the geometric center line of the muscle belly. The direction of the line connecting each adjacent pair of points represents the main direction of the muscle fiber, used to describe the actual curvature of the muscle in space. Corneal light reflection offset: A three-dimensional deviation vector of the reflected image of the light source at the corneal apex relative to the center of the pupil. The offset reflects the degree of rotational tilt of the eyeball in a resting state and is used to correct the spatial coordinates of the extraocular muscle attachment points in the actual eye position. Eyeball rotation center coordinates: The fixed spatial position of the geometric center point around which the eyeball rotates in three-dimensional motion within the orbit. Calculated from multi-angle images, it is used to uniformly transform the origin coordinates of all extraocular muscles to a reference system with this center as the origin. Fixed attachment relationships between muscles and the eyeball, and between muscles and the orbital bone: Anatomically, every extraocular muscle's origin must be attached to the sclera of the eyeball, and its insertion must be attached to the orbital bone wall. This inherent connection rule determines the unique direction for establishing directed edges between nodes in the heterogeneous atlas, i.e., from the bone wall node to the muscle node and then to the eyeball node. In a heterogeneous graph, the attribute values ​​of neighboring nodes are the coordinates and offsets carried by all nodes directly and directed to a given node. For example, a muscle node connected to the origin node carries the coordinates of the muscle belly path point, and an eye node connected to the muscle node carries the coordinates of the rotation center. The aggregated feature vector is a weighted sequence of attribute values ​​from all neighboring nodes of a given node, ordered by the direction of the connecting edges. The weights are determined by the reciprocal of the spatial Euclidean distance between the node and its neighbors, and are used to quantify the local geometric configuration features of the node within the entire graph. The spatial configuration of extraocular muscles is the relative spatial layout pattern formed by the origin, insertion, muscle belly path point, and adjacent eye rotation centers and corneal light-reflecting points of each extraocular muscle. Differences in configuration between different patients or different extraocular muscles of the same patient are reflected in the differences in node sets between sub-graphs. The three-dimensional coordinate system takes the midpoint of the line connecting the midpoints of the patient's two external auditory canals as the origin, the direction from the origin to the root of the nose as the positive direction of the horizontal axis, the direction from the origin to the top of the skull as the positive direction of the vertical axis, and the direction from the origin to the left external auditory canal as the positive direction of the vertical axis. All spatial coordinates are measured within this system.A positioning reference point is a single spatial point obtained by taking the arithmetic mean of the coordinates of all nodes in a sub-map, including the origin, end, muscle belly path points, corneal light reflection points, and coordinates of the eye rotation center. This point represents the overall position of the sub-map and eliminates the influence of the discrete distribution of nodes within the sub-map on relative position comparisons. The standard eye position reference plane is the plane jointly determined by the corneal apex and nasal root point of both eyes when the normal human eye is looking straight ahead and the head is in a neutral posture. This plane serves as the reference zero point in strabismus mapping and is used to calculate the deflection angle and offset distance in the actual mapping results. Spatial deflection angle: The angle between the plane where all positioning reference points are located after adjustments and the standard eye position reference plane, decomposed into the pitch angle (rotation around the horizontal axis) and the yaw angle (rotation around the vertical axis); used to quantitatively describe the degree of torsion of the overall extraocular muscle attachment point distribution in strabismus patients relative to their normal position. The centroid offset distance is the straight-line distance between the geometric center of the plane where all positioning reference points are located in the three-dimensional coordinate system and the geometric center of the standard eye position reference plane. It is used to quantify the amount of translation of the overall deviation of the extraocular muscle spatial configuration of strabismus patients from their normal position.

[0016] In the above embodiments, the principle is as follows: Figure 2 This embodiment extracts the spatial coordinates of extraocular muscles from multi-angle image sequences and constructs an initial heterogeneous atlas by combining the offset of the corneal light-reflecting point with the coordinates of the eyeball rotation center. This transforms anatomical attachment relationships into directed connections between nodes, achieving a correlation modeling between extraocular muscle structure and eyeball movement mechanisms. Subsequently, node feature vectors are generated through weighted aggregation of neighborhood node attributes. Hierarchical subgraph partitioning is performed based on Euclidean distance, automatically classifying extraocular muscles with similar spatial configurations to form a subgraph system that unifies anatomy and function. Finally, through coordinate system mapping and planarization calibration of the positioning base point, the spatial deflection angle and centroid offset of the subgraph relative to the standard eye position reference plane are quantified, thereby completing a systematic mapping of the three-dimensional spatial configuration of the extraocular muscles in strabismus patients and an analysis of their pathological posture.

[0017] Example 2: Figure 3 As shown, based on Embodiment 1, the process of forming an initial heterogeneous graph by all entity nodes and directed connecting edges in step S100 of this embodiment of the invention specifically includes the following steps: Step S101: Spatial coordinates of the origin of each extraocular muscle extracted from the multi-angle image sequence are spatially aligned with the offset of the corneal light-reflection point and the coordinates of the eye rotation center corresponding to the same extraocular muscle, so that the starting point of the three-dimensional vector of the corneal light-reflection point offset is translated to the coordinates of the eye rotation center; then the starting point spatial coordinates are corrected in the opposite direction of the offset to obtain the corrected starting point spatial coordinates. Step S102: Store the corrected starting point spatial coordinates, corneal light reflection point offset, and eyeball rotation center coordinates as three parallel attribute values ​​of the same entity node to form a muscle starting point entity node; based on the fixed attachment relationship between muscles and eyeballs, and muscles and bone walls in anatomy, the starting point entity node of each extraocular muscle points to the node where the eyeball rotation center coordinates are located through directed connecting edges. Step S103: Send directed connecting edges from the orbital bone wall nodes to the muscle origin entity nodes, and connect all muscle origin entity nodes, eye rotation center nodes, and orbital bone wall nodes in sequence according to the direction rules. All established entity nodes and directed connecting edges constitute the initial heterogeneous graph.

[0018] In the above embodiments, this embodiment spatially aligns the spatial coordinates of the extraocular muscle origin extracted from the multi-angle image sequence with the corneal light-reflection point offset and the coordinates of the eyeball rotation center. By translating the three-dimensional vector origin of the corneal light-reflection point offset to the coordinates of the eyeball rotation center, and then correcting the origin coordinates in reverse, the precise positioning of the extraocular muscle origin in three-dimensional space is achieved. This eliminates measurement errors caused by changes in eyeball position and provides an accurate spatial reference for atlas construction. The corrected origin coordinates, corneal light-reflection point offset, and eyeball rotation center coordinates are stored as parallel attribute values ​​in the same entity node, forming a muscle origin entity node. Based on the anatomical attachment relationship, the origin node is pointed to the eyeball rotation center node through directed connecting edges, establishing a mechanical transmission path between the extraocular muscle and the eyeball. At the same time, the directed connecting edges from the orbital bone wall node to the muscle origin node fully express the fixed support relationship of the bone wall for the extraocular muscle. The initial heterogeneous atlas, formed by connecting all nodes according to directional rules, integrates the spatial positioning of extraocular muscles, eye movement parameters, and bony structural information. The atlas realizes the structured expression of the topological relationships and biomechanical connections between the origin and insertion points of extraocular muscles, the center of eye rotation, and the orbital bone wall, providing a calculable and scalable data foundation for analyzing extraocular muscle function, simulating eye movement, and assisting in clinical diagnosis.

[0019] Example 3: Based on Example 2, the process of sequentially connecting all muscle origin nodes, eyeball rotation center nodes, and orbital bone wall nodes according to directional rules in step S103 of this embodiment of the invention specifically includes the following steps: Step S1031: Perform a spatial difference operation between the corrected origin spatial coordinates stored in the obtained muscle origin entity node and the eye rotation center coordinates to obtain a vector pointing from the corrected origin spatial coordinates to the eye rotation center coordinates; use the direction of the vector as the direction of the directed connection edge, and use the magnitude of the vector as the weight value of the edge to establish a directed connection edge pointing from the muscle origin entity node to the node where the eye rotation center coordinates are located. One such directed connection edge is generated for each extraocular muscle. Step S1032: Spatially superimpose each formed muscle origin entity node with the opposite direction vector of the corneal light-reflecting point offset used to correct the origin spatial coordinates to obtain the offset virtual bone wall attachment point coordinates; search for the bone wall point with the closest spatial distance to the virtual bone wall attachment point coordinates in the pre-established orbital bone wall point cloud, set the bone wall point as the orbital bone wall node, and establish a directed connection edge from the orbital bone wall node to the corresponding muscle origin entity node; the direction is from the bone wall to the muscle, and the edge attribute records the residual vector between the virtual bone wall attachment point and the real bone wall point; Step S1033: All the directed edges generated from the muscle origin entity node to the eye rotation center node, all the directed edges generated from the orbital bone wall node to the muscle origin entity node, as well as all the muscle origin entity nodes, all the involved eye rotation center nodes, and all the determined orbital bone wall nodes, are gathered according to the direction of the arrows on the directed edges, so that every path starting from the orbital bone wall node, passing through the muscle origin entity node, and finally reaching the eye rotation center node remains continuous and does not form any reverse loops. The union of all paths constitutes the initial heterogeneous graph.

[0020] In the above embodiments, this embodiment aggregates all directed edges from the muscle origin entity node to the eye rotation center node, all directed edges from the orbital bone wall node to the muscle origin entity, and all related nodes according to the arrow direction. This ensures that each path from the bone wall node through the muscle origin node to the eye rotation center node remains continuous and without reverse loops. The initial heterogeneous graph formed by the union of all paths fully integrates the spatial origin and termination relationships of the extraocular muscles, mechanical vectors, bony attachment registration, and path continuity, forming a heterogeneous graph structure with clear biomechanical orientation and topological constraints. This provides a clear and comprehensive computational framework for motion simulation, mechanical analysis, and pathological reasoning.

[0021] Example 4: Figure 4 As shown, based on Example 1, the process of hierarchically dividing nodes according to the Euclidean distance between the aggregate feature vectors of all nodes in step S200 of this embodiment of the invention specifically includes the following steps: Step S201: Based on the obtained aggregate feature vector of each node, sort the aggregate feature vectors of all nodes in ascending order of vector magnitude, and take the aggregate feature vector of the node with the smallest magnitude as the first-level reference vector; calculate the Euclidean distance between the aggregate feature vector of each of the remaining nodes and the first-level reference vector, and include the nodes with Euclidean distance less than a preset threshold together with the nodes corresponding to the first-level reference vector into the first-level candidate set, and remove all nodes in the first-level candidate set from the full node set; Step S202: In the remaining node set after removing the first-level candidate set, sort the nodes again according to the magnitude of the aggregated feature vector from smallest to largest, and take the aggregated feature vector of the node with the smallest magnitude as the second-level reference vector; calculate the Euclidean distance between the aggregated feature vector of each node in the remaining nodes and the second-level reference vector, and include nodes whose Euclidean distance is less than the same preset threshold, together with the nodes corresponding to the second-level reference vector, into the second-level candidate set, and remove all nodes in the second-level candidate set from the remaining node set; repeat until the remaining node set is empty, and obtain several level candidate sets; Step S203: For each hierarchical candidate set, extract the directed edges of all nodes in the constructed initial heterogeneous graph. Check whether there is at least one path consisting of directed edges between any two nodes in the hierarchical candidate set that does not cross any node outside the set. If there is no path between any two nodes that does not cross any node outside the set, split the hierarchical candidate set into two subsets, such that there is at least one path consisting of directed edges between any two nodes in each subset that does not cross any node outside the subset. After all the splits are completed, each subset is a subgraph. All subgraphs do not overlap. Each subgraph corresponds to a type of extraocular muscle spatial configuration.

[0022] In the above embodiments, this embodiment sorts the nodes based on the magnitude of their aggregated feature vectors and selects a reference vector. It then calculates the Euclidean distance between the remaining nodes and the reference vector and filters them according to a preset threshold, achieving preliminary node clustering based on feature space proximity. Nodes with similar features are grouped into the same level candidate set and removed from the full set, ensuring mutual exclusion and progressive partitioning of the levels. The process of selecting a reference vector, calculating Euclidean distance, and classifying nodes is repeated in each round of the remaining node set until the node set is empty, resulting in multiple level candidate sets. This iterative partitioning method ensures that nodes within each level have high similarity in the feature space, and that different levels form an ordered hierarchical structure in terms of feature magnitude and vector distribution. For each level candidate set, directed connecting edges in the initial heterogeneous graph are extracted, and it is checked whether there is a path between any two nodes within the set that does not cross nodes outside the set. If no such path exists, the candidate set is further split into subsets that satisfy the internal path connectivity requirements. The resulting subgraphs are connected by direct or indirect directed edges, and the subgraphs do not overlap. Based on feature similarity, connectivity constraints on the graph topology are further introduced to ensure that each subgraph not only has similar features but also forms a relatively independent connected subgraph in terms of anatomical and functional relationships. Each subgraph corresponds to a class of extraocular muscle configurations that are consistent in spatial location, mechanical properties, and topological connections, providing a structurally clear and semantically explicit subgraph partitioning basis for configuration classification, functional analysis, and anomaly detection.

[0023] Example 5: Based on Example 4, the process of splitting the hierarchical candidate set into two subsets in step S203 of this embodiment of the invention specifically includes the following steps: Step S2031: Within the hierarchical candidate set, calculate the Euclidean distance between the aggregated feature vectors of every two nodes, and select the pair of nodes with the largest Euclidean distance as the first splitting node and the second splitting node; using the first splitting node and the second splitting node as the reference, calculate the Euclidean distance between the aggregated feature vector of each other node in the hierarchical candidate set and the first splitting node, and the Euclidean distance between the first splitting node and the second splitting node; assign the node that is closer to the first splitting node in Euclidean distance to the first temporary subset, assign the node that is closer to the second splitting node to the second temporary subset, and temporarily set aside the two nodes with equal Euclidean distances. Step S2032: Take out the temporarily shelved nodes one by one, and calculate the minimum value of the aggregate feature vector Euclidean distance between the node and all nodes in the first temporary subset, and the minimum value of the aggregate feature vector Euclidean distance between the node and all nodes in the second temporary subset. If the former is less than the latter, the node is assigned to the first temporary subset; otherwise, it is assigned to the second temporary subset. After all shelved nodes have been assigned, the updated first temporary subset and second temporary subset are obtained. Step S2033: Extract all directed edges of nodes in the updated first temporary subset from the initial heterogeneous graph, retain only directed edges whose two endpoints both belong to the first temporary subset, start from any node in the first temporary subset, and simultaneously expand along the direction and opposite direction of the retained directed edges, marking all other nodes that can be reached through the directed edge sequence; if the number of marked nodes is equal to the total number of nodes in the first temporary subset, then the first temporary subset satisfies the condition that an internal path exists, otherwise remove the unmarked nodes from the first temporary subset and assign them to the second temporary subset; perform the same operation on the second temporary subset; the final first temporary subset and the second temporary subset are the two subsets after splitting the hierarchical candidate set.

[0024] Euclidean distance refers to the linear spatial distance between the aggregated feature vectors of two nodes, used to quantify the degree of difference between two different extraocular muscles in the feature space. The aggregated feature vector of each node is obtained by weighted aggregation of the spatial coordinates of the origin, insertion, midline path point of the muscle belly, offset of the corneal light-reflecting point, and coordinates of the eye rotation center of the corresponding extraocular muscle. Therefore, the magnitude of the Euclidean distance directly reflects the overall similarity of the two extraocular muscles in terms of attachment position, curvature, and spatial layout relative to the eye rotation center. The smaller the Euclidean distance, the closer the spatial configuration of the two extraocular muscles is; the larger the Euclidean distance, the more significant the difference in the spatial configuration of the two extraocular muscles. In steps S201 and S202, by calculating the Euclidean distance between the remaining nodes and the reference vector and comparing it with a preset threshold, extraocular muscles with similar spatial configurations are grouped into the same candidate set at the same level. In step S2031, the pair of nodes with the largest Euclidean distance is selected as splitting nodes, which are used to separate the two extraocular muscles with the greatest internal differences, thereby ensuring that the extraocular muscle configurations within the two subsets after splitting have higher consistency. Throughout the mapping process, the Euclidean distance is the core metric tool connecting feature vectors and spatial configuration classification.

[0025] In step S201, the Euclidean distance between the aggregated feature vector of each remaining node and the first-level reference vector is first calculated. Each aggregated feature vector is generated by weighted aggregation of the spatial coordinates of the origin, the spatial coordinates of the insertion, the path point of the muscle belly midline, the offset of the corneal light-reflecting point, and the coordinates of the eye rotation center of the corresponding extraocular muscle. The magnitude of the Euclidean distance directly reflects the overall similarity between the spatial configuration of the extraocular muscle represented by the node and the spatial configuration of the extraocular muscle of the reference node. Nodes with Euclidean distances less than a preset threshold are grouped into the same level of candidate set, thereby automatically aggregating extraocular muscles with similar spatial configurations. In step S202, in each iteration, the node with the smallest aggregated feature vector magnitude in the current remaining node set is used as the new reference vector. The Euclidean distances between the remaining nodes and the reference vector are calculated again, and the same threshold filtering operation is repeated. At this time, the Euclidean distance is used to continue to extract extraocular muscles with similar configurations to the new reference vector in the current remaining set, while ensuring that the Euclidean distances between different level candidate sets are all greater than the threshold, thus forming a non-overlapping hierarchical structure. In step S2031, for all nodes within the same hierarchical candidate set, the Euclidean distance between the aggregated feature vectors of every two nodes is calculated, and the pair of nodes with the largest Euclidean distance is selected as the first splitting node and the second splitting node. The Euclidean distance is used to identify the two extraocular muscles with the greatest spatial configuration difference within the hierarchical candidate set, and these two are used as the two separation centers for subsequent splitting. Subsequently, the Euclidean distance between the aggregated feature vector of each of the remaining nodes and the two splitting nodes is calculated. Nodes closer to the first splitting node are assigned to the first temporary subset, and nodes closer to the second splitting node are assigned to the second temporary subset. Here, the Euclidean distance is used to assign each extraocular muscle to the subset containing the splitting node with the closest spatial configuration, thereby further improving the consistency of the extraocular muscle configuration within each of the two subsets after splitting. Throughout the mapping process, each calculation of Euclidean distance is based on aggregated feature vectors, which in turn originate from attribute values ​​in the initial heterogeneous atlas that are directly related to the anatomical coordinates and motion parameters of the extraocular muscles. Therefore, the value of Euclidean distance directly corresponds to the degree of similarity or difference in the spatial configuration between different extraocular muscles in strabismus patients, providing a quantitative classification basis for sub-atlas division and strabismus mapping results output.

[0026] In the above embodiments, this embodiment calculates the Euclidean distance between the aggregated feature vectors of every two nodes within the hierarchical candidate set, selecting the pair of nodes with the largest distance as the splitting benchmark. This ensures that the two initial splitting nodes have the greatest difference in the feature space, providing two representative types of feature centers for set partitioning. Using the two splitting nodes as benchmarks, the Euclidean distances between the remaining nodes in the set and these nodes are calculated, and the nodes are assigned to the closer temporary subsets, achieving a preliminary bipartition based on feature similarity. Nodes with equal distances are temporarily set aside to avoid situations where their assignment cannot be clearly defined at the feature distance level. The set-aside nodes are then sequentially retrieved, and their minimum Euclidean distances to all nodes in the two temporary subsets are calculated. Based on the minimum distance, they are assigned to the corresponding subsets. This further refines the node assignment judgment, considering not only the distance between the node and the splitting benchmark but also the proximity of the node to the subset as a whole in the feature space, enhancing the local consistency of the partitioning. For the updated temporary subset, the directed connecting edges that are completely located within the subset in the initial heterogeneous graph are extracted. Starting from any node, the graph is expanded bidirectionally along the direction of the directed edge and in the opposite direction, marking all nodes reachable through the edge sequence. If the number of marked nodes equals the total number of nodes in the subset, it indicates that all nodes within the subset are connected by directed edges, satisfying the condition for the existence of internal paths. Otherwise, unmarked nodes are removed and assigned to another subset. Based on feature partitioning, a graph topological connectivity constraint is forcibly introduced to ensure that each subset obtained in the end is not only similar in features but also constitutes a connected subgraph in terms of directed connections. The two subsets formed in the end maintain the clustering consistency of the feature space and satisfy the structural connectivity requirements of the heterogeneous graph, laying the foundation for constructing a subgraph with clear anatomical and functional boundaries.

[0027] Example 6: Based on Example 5, the process of simultaneously expanding along the direction and opposite direction of the retained directed connection edge in step S2033 of this embodiment of the invention specifically includes the following steps: Step S20331: Randomly select a node from the first temporary subset as the starting node, mark the starting node as a visited node, and record the magnitude of the aggregated feature vector of the starting node; extract all directed connection edges of the starting node in the initial heterogeneous graph, and filter out directed connection edges whose two endpoints both belong to the first temporary subset; add all the other end nodes of the directed connection edges that are directly connected to the starting node, regardless of whether the arrow of the directed connection edge points to the starting node or points out from the starting node, to the set of nodes to be expanded, and record the magnitude of the aggregated feature vector for each node in the set of nodes to be expanded; Step S20332: Extract the node with the smallest aggregate feature vector magnitude from the set of nodes to be expanded, mark the node as a visited node and remove it from the set of nodes to be expanded; extract all directed connections of the removed node in the initial heterogeneous graph; filter again for directed connections where both endpoints belong to the first temporary subset and the other end node is not marked as visited, add all the other end nodes of the directed connections that are directly connected to the removed node to the set of nodes to be expanded, and record the aggregate feature vector magnitude of each newly added node; Step S20333: Repeat step S20332 until the set of nodes to be expanded is empty. At this time, the set of all nodes marked as visited is the set of all nodes that can be reached from the starting node through the directed edge sequence. If multiple nodes in the set of nodes to be expanded have the same aggregate feature vector magnitude and are both the minimum value during the repetition process, then the nodes are taken out at the same time, and the extraction and addition operations in step S20332 are performed on each node respectively, and then the set of nodes to be expanded is updated uniformly.

[0028] In the above embodiments, this embodiment selects a node from the first temporary subset as the starting node and marks it as visited; extracts directed connecting edges that are completely located within the subset in the initial heterogeneous graph, and adds all nodes on the other end directly connected to the starting node through these edges to the set to be expanded, regardless of whether the edge direction points to or points out from the starting node; this achieves the initialization of bidirectional expansion along directed edges starting from the starting node, and records the aggregate feature vector magnitude of the nodes to be expanded, providing a basis for expansion in order of magnitude. The node with the smallest aggregate feature vector magnitude is taken from the set to be expanded, marked as visited, and removed from the set; directed connecting edges that are completely located within the subset in the initial heterogeneous graph and whose other end is not visited are extracted, and unvisited nodes directly connected through these edges are added to the set to be expanded; by prioritizing the expansion of nodes with smaller magnitudes, the expansion process is guided to gradually advance along the low-magnitude direction in the feature space, which helps to explore the connectivity structure of the subset while maintaining the consistency of local features. Repeat the expansion process until the set to be expanded is empty. If multiple nodes with the same minimum feature magnitude are encountered during the expansion, these nodes are simultaneously extracted and added, and the set to be expanded is updated uniformly. This ensures the parallelism and fairness of the expansion when feature magnitudes are the same, avoiding bias introduced by the order of selection. Ultimately, all nodes marked as visited constitute the set of all nodes reachable from the starting node via a sequence of directed edges. By performing bidirectional breadth-first expansion in ascending order of feature magnitude, connectivity is guaranteed while ensuring a certain degree of order in the feature space of the expansion path. This allows for more effective identification of connected components within a subset based on directed edges and provides an accurate node coverage range for determining whether a subset satisfies the condition for the existence of an internal path.

[0029] Example 7: Figure 5 As shown, based on Example 1, the process of obtaining the oblique three-dimensional mapping result in step S300 of this embodiment of the invention specifically includes the following steps: Step S301: Extract the coordinates of the positioning base points of each sub-map in sequence. Using the horizontal plane of ocular anatomy as the reference plane, calculate the normal deflection angle of each positioning base point relative to the horizontal plane of ocular anatomy. Then, using the sagittal plane of ocular anatomy as the reference plane, calculate the normal deflection angle of each positioning base point relative to the sagittal plane. Combine the two normal deflection angles corresponding to each positioning base point to form the dual-angle offset of that base point. Select the base point with the smallest dual-angle offset modulus among all positioning base points as the fixed base point and keep its spatial coordinates unchanged. Step S302: Rotate and correct each positioning base point except the fixed base point according to its normal deflection angle relative to the horizontal plane and the sagittal plane, so that the horizontal plane normal deflection angle and the sagittal plane normal deflection angle in the dual-angle offset of each positioning base point are both zero; after correction, a new set of positioning base point coordinates is obtained, and all corrected positioning base points are located on the intersection line of the horizontal plane and the sagittal plane of ocular anatomy; take the intersection line as the rotation axis and rotate all corrected positioning base points at the same time, so that the sum of the squares of the Euclidean distances between the positioning base points in the direction perpendicular to the rotation axis is minimized, and all positioning base points are in the same plane when the rotation stops; Step S303: Fit all the coplanar positioning base points obtained to a least squares plane, calculate the angle between the normal direction of the least squares plane and the normal direction of the standard eye reference plane, decompose the angle into two components around the horizontal and vertical axes of the three-dimensional coordinate system to obtain the spatial deflection angle; calculate the three-dimensional straight-line distance between the position of the geometric center point of the least squares plane in the three-dimensional coordinate system and the geometric center point of the standard eye reference plane to obtain the centroid offset distance; use the spatial deflection angle and the centroid offset distance as outputs to obtain the squinting three-dimensional mapping result.

[0030] The normal deflection angle of the positioning reference point relative to the anatomical horizontal plane of the eyeball refers to the angle between a perpendicular line drawn from the positioning reference point to the anatomical horizontal plane of the eyeball and the normal direction of the horizontal plane at the location of the positioning reference point. It describes the degree of deviation of the overall spatial position of the sub-map in the vertical tilt direction. In strabismus mapping, the angle reflects the pitch deviation of the corresponding extraocular muscle configuration relative to the normal horizontal reference plane. The normal deflection angle of the positioning reference point relative to the sagittal plane refers to the angle between a perpendicular line drawn from the positioning reference point to the anatomical sagittal plane of the eyeball and the normal direction of the sagittal plane at the location of the positioning reference point. It describes the degree of deviation of the overall spatial position of the sub-map in the horizontal tilt direction. In strabismus mapping, the angle reflects the yaw deviation of the corresponding extraocular muscle configuration relative to the normal sagittal reference plane. The three-dimensional straight-line distance between the geometric center of the least squares plane in the three-dimensional coordinate system and the geometric center of the standard eye position reference plane refers to the Euclidean distance between the geometric center of the least squares plane obtained by fitting all positioning base points after coplanar adjustment and considering it as a spatial point, and the geometric center of the standard eye position reference plane. This distance is used to quantify the translational displacement of the patient's overall extraocular muscle attachment point distribution relative to the normal eye position. In strabismus mapping results, distance and spatial deflection angle together constitute a dual quantitative description of the translational and rotational degree of strabismus. The process of obtaining the spatial deflection angle involves selecting three non-collinear positioning base points on the least squares plane. A spatial plane is defined using these three base points. A straight line perpendicular to the least squares plane is drawn through any one of the base points on this spatial plane, and this line is extracted as the normal direction line of the least squares plane. From the standard eye-position reference plane, three non-collinear fixed points are selected on the plane. A standard eye-position reference plane is defined using these three fixed points. A straight line perpendicular to the plane is drawn through any one of the fixed points on this standard eye-position reference plane, and this line is extracted as the normal direction line of the standard eye-position reference plane. The two normal direction lines are translated to the origin of the three-dimensional coordinate system, and the minimum spatial angle between the two normal direction lines is measured to obtain the total deflection angle. A line parallel to the horizontal axis of the three-dimensional coordinate system is drawn through the origin of the three-dimensional coordinate system. A straight line is used as the horizontal axis projection baseline. A straight line parallel to the vertical axis of the three-dimensional coordinate system is drawn through the origin of the three-dimensional coordinate system, and this line is used as the vertical axis projection baseline. The spatial cone surface containing the obtained total deflection angle is orthogonally decomposed into the vertical planes of the horizontal axis projection baseline and the vertical plane of the vertical axis projection baseline, respectively, to obtain the projection angle components of the total deflection angle in the vertical plane of the horizontal axis projection baseline and the projection angle components of the total deflection angle in the vertical plane of the vertical axis projection baseline. The projection angle components in the vertical plane of the horizontal axis projection baseline are used as the yaw angle components of the rotation around the vertical axis, and the projection angle components in the vertical plane of the vertical axis projection baseline are used as the pitch angle components of the rotation around the horizontal axis. The yaw angle components and the pitch angle components are combined as the spatial deflection angle and output to the mapping results.

[0031] In the above embodiments, this embodiment sequentially extracts the coordinates of the positioning base points of each sub-map, calculates their normal deflection angles relative to the horizontal and sagittal planes of the eyeball's anatomy, and combines them into a dual-angle offset. The base point with the smallest dual-angle offset magnitude among all base points is selected as the fixed base point, keeping its spatial coordinates unchanged; a corrected geometric reference datum is established, ensuring that at least one base point remains stable in spatial position during the transformation process, providing a fixed point for rotation and translation of the overall structure. Each remaining positioning reference point is rotated and corrected according to its normal deflection angle relative to the horizontal and sagittal planes, so that the dual-angle offset of each reference point is zero. After correction, all reference points are located on the intersection of the horizontal and sagittal planes of ocular anatomy, with the intersection line serving as the rotation axis. Simultaneously, all corrected reference points are rotated to minimize the sum of the squares of the Euclidean distances between them in the direction perpendicular to the rotation axis, and all positioning reference points are in a coplanar plane when the rotation stops. Through a two-step rotation operation, the reference points are first aligned to the intersection of the anatomical reference planes, and then optimized to make the reference points coplanar, thus achieving geometric regularization of the extraocular muscle spatial configuration and eliminating non-coplanarity caused by individual anatomical differences or measurement errors. All coplanar positioning reference points are fitted to a least-squares plane, and the angle between the normal direction of this plane and the normal direction of the standard eye position reference plane is calculated. The angle is decomposed into two components around the horizontal and vertical axes of the three-dimensional coordinate system to obtain the spatial deflection angle. At the same time, the three-dimensional straight-line distance between the geometric center point of the least-squares plane and the geometric center point of the standard eye position reference plane is calculated to obtain the centroid offset distance. The spatial deflection angle reflects the directional deviation of the overall plane of the extraocular muscle group relative to the standard eye position, while the centroid offset distance characterizes the positional shift of this plane in space. Together, they constitute the three-dimensional mapping results of strabismus, enabling a quantitative description of the overall pose of the extraocular muscle spatial configuration and providing precise geometric parameters for clinical diagnosis and surgical planning.

[0032] Example 8: Based on Example 7, the process of obtaining the centroid offset distance in step S303 of this embodiment of the invention specifically includes the following steps: Step S3031: From all the coplanar positioning base points obtained, extract the aggregated feature vector of all nodes in the obtained sub-map corresponding to each positioning base point, and calculate the sum of the magnitudes of the aggregated feature vectors of all nodes in each sub-map; use the sum as the spatial weight value of the positioning base point; multiply the horizontal, vertical, and ordinate of the spatial coordinates of each positioning base point by the spatial weight value of the positioning base point to obtain the weighted coordinate components of the positioning base point; accumulate the weighted coordinate components of all positioning base points horizontally, vertically, and vertically to obtain three sums; divide each sum by the sum of all spatial weight values ​​to obtain the horizontal, vertical, and ordinate of the weighted center point, and use the weighted center point as the geometric center point of the least squares plane; Step S3032: Select three pre-calibrated fixed points from the standard eye position reference plane, denoted as the first fixed point, the second fixed point, and the third fixed point, respectively; extract the dual-angle offset magnitude of all positioning base points and calculate the arithmetic mean of these magnitudes; multiply each component of the spatial coordinate of the first fixed point by the arithmetic mean to obtain the coordinates of the first offset point; multiply each component of the spatial coordinate of the second fixed point by the arithmetic mean to obtain the coordinates of the second offset point; multiply each component of the spatial coordinate of the third fixed point by the arithmetic mean to obtain the coordinates of the third offset point; add the corresponding components of the coordinates of the first, second, and third offset points and divide by three to obtain the coordinates of the offset center point; extract the dual-angle offset magnitude of the fixed base points as the reference coefficient, divide each coordinate component of the offset center point by the reference coefficient to obtain the corrected coordinates of the standard center point, and use the corrected standard center point as the geometric center point of the standard eye position reference plane; Step S3033: Subtract the x-coordinate of the geometric center point of the least squares plane from the x-coordinate of the geometric center point of the standard eye reference plane to obtain the x-coordinate deviation; subtract the y-coordinate to obtain the y-coordinate deviation; subtract the y-coordinate to obtain the y-coordinate deviation; subtract the y-coordinate to obtain the y-coordinate deviation; project the x-coordinate deviation, y-coordinate deviation, and y-coordinate deviation onto the three coordinate planes of the three-dimensional coordinate system to obtain three projection deviation vectors; calculate the magnitude of each projection deviation vector to obtain three projection distances; arrange the three projection distances in descending order of value, take the distance value in the middle position after arrangement, and output the distance value as the three-dimensional straight-line distance, which is the centroid offset distance.

[0033] In the above embodiments, this embodiment extracts the aggregated feature vectors of all nodes within the sub-map corresponding to each coplanar positioning base point, calculates the sum of the magnitudes of the aggregated feature vectors of all nodes within each sub-map, and uses this sum as the spatial weight value of the positioning base point. Each component of the spatial coordinates of each positioning base point is multiplied by its spatial weight value to obtain weighted coordinate components. All weighted coordinate components are accumulated in the horizontal, vertical, and longitudinal directions to obtain three sums. Each sum is then divided by the sum of all spatial weight values ​​to obtain the coordinates of the weighted center point, which serves as the geometric center point of the least squares plane. By introducing the sum of the magnitudes of the aggregated feature vectors of nodes within the sub-map as weights, the calculation of the geometric center point not only depends on the spatial position of the base point but also reflects the functional or structural strength of the anatomical region represented by each sub-map, thus obtaining a geometric center point that can characterize the weighted centroid of the spatial distribution of the extraocular muscle group. Three pre-calibrated fixed points are selected on the standard eye position reference plane. The magnitudes of the dual-angle offsets of all positioning base points are extracted and their arithmetic mean is calculated. Each component of the spatial coordinates of each fixed point is multiplied by this arithmetic mean to obtain the coordinates of the three offset points. The corresponding components of the three offset point coordinates are then added together and divided by three to obtain the coordinates of the offset center point. The magnitudes of the dual-angle offsets of the fixed base points are extracted as reference coefficients. Each coordinate component of the offset center point is divided by this reference coefficient to obtain the corrected coordinates of the standard center point, which is used as the geometric center point of the standard eye position reference plane. The standard reference point is scaled and corrected by the statistical measure of the magnitudes of the dual-angle offsets, so that the position of the standard center point can adapt to the overall offset degree of the individual extraocular muscle spatial configuration, realizing the dynamic adjustment of the center point of the standard reference plane relative to the individual anatomical features. The coordinate components of the least squares plane geometric center point and the standard eye position reference plane geometric center point are subtracted respectively to obtain the deviation components in the three coordinate axis directions. The three deviation components are projected onto the three coordinate planes of the three-dimensional coordinate system to obtain three projection deviation vectors. The magnitude of each projection deviation vector is calculated to obtain the three projection distances. The three projected distances are arranged in descending order of value, and the distance value in the middle position is taken as the three-dimensional straight-line distance output, that is, the centroid offset distance. By projecting the coordinate deviation and taking the median value of the sorting, the influence of extreme deviation in a single direction on the overall distance measurement is reduced, the robustness of the centroid offset distance is improved, and it can more stably represent the overall positional difference between two geometric center points in space.

[0034] Example 9: Based on Example 8, the process of obtaining three projection distances in step S3033 of this embodiment of the invention specifically includes the following steps: Step S30331: Take out the obtained horizontal coordinate deviation and vertical coordinate deviation, and take these two deviation values ​​as two mutually perpendicular line segments. Set the common starting point of the two line segments as the origin of the three-dimensional coordinate system; start from the origin and move the length of the horizontal coordinate deviation along the horizontal axis to reach the first intermediate point; then move the length of the vertical coordinate deviation along the vertical axis from the first intermediate point to reach the second intermediate point; connect the origin and the second intermediate point to obtain a straight line segment, and measure the length of the straight line segment as the first projected distance; Step S30332: Take out the obtained horizontal coordinate deviation and vertical coordinate deviation, and take the two deviation values ​​as two mutually perpendicular line segments. Set the common starting point of the two line segments as the origin of the three-dimensional coordinate system; start from the origin and move the length of the horizontal coordinate deviation along the horizontal axis to reach the third intermediate point; then move the length of the vertical coordinate deviation along the vertical axis from the third intermediate point to reach the fourth intermediate point; connect the origin and the fourth intermediate point to obtain a straight line segment, and measure the length of the straight line segment as the second projection distance; Step S30333: Take out the obtained ordinate deviation and vertical coordinate deviation, and take these two deviation values ​​as two mutually perpendicular line segments. Set the common starting point of the two line segments as the origin of the three-dimensional coordinate system; start from the origin and move the length of the ordinate deviation along the vertical axis to reach the fifth intermediate point; then move the length of the vertical coordinate deviation along the vertical axis from the fifth intermediate point to reach the sixth intermediate point; connect the origin and the sixth intermediate point to obtain a straight line segment, and measure the length of the straight line segment as the third projection distance.

[0035] In the above embodiments, this embodiment constructs the deviation components of the two geometric center points along the two coordinate axes into right-angled sides in three pairwise orthogonal coordinate planes, and calculates the length of their hypotenuses, thus systematically obtaining the projected distances of the two points in the horizontal, sagittal, and coronal planes. The three projected distances together constitute a multi-plane metric of the spatial position difference between the two center points, providing basic data for subsequently selecting the median value as a robust centroid offset distance.

[0036] Example 10: As Figure 6 As shown, based on Examples 1-9, the knowledge graph-based 3D mapping system provided in this embodiment of the invention includes: The initial heterogeneous atlas construction module 1 is responsible for acquiring multi-angle image sequences of strabismus patients' eyes, extracting the origin spatial coordinates, insertion spatial coordinates, and muscle belly midline path points of each extraocular muscle; using the origin spatial coordinates, the offset of the patient's corneal light reflection point, and the coordinates of the eyeball rotation center as attribute values ​​of the entity nodes; establishing directed connection edges between nodes based on the fixed attachment relationships between muscles and eyeballs, and muscles and bone walls in anatomy; and all entity nodes and directed connection edges constitute the initial heterogeneous atlas. The graph segmentation module 2 is responsible for weighting and aggregating the attribute values ​​of the neighboring nodes connected by the directed edges of each node in the initial heterogeneous graph to obtain the aggregated feature vector of the node; and dividing the nodes into hierarchical levels according to the Euclidean distance between the aggregated feature vectors of all nodes, and classifying the nodes with Euclidean distance less than a preset threshold into the same subgraph, so that the initial heterogeneous graph is divided into several non-overlapping subgraphs, each subgraph corresponding to a type of extraocular muscle spatial configuration. The positioning base point module 3 is responsible for mapping the coordinates of all nodes in each sub-map to a unified three-dimensional coordinate system. It takes the average spatial coordinates of all nodes in each sub-map as the positioning base point of the sub-map, calculates the normal deflection angle of each positioning base point relative to the horizontal and sagittal planes of the eyeball anatomy, and adjusts the relative positions between sub-maps so that all positioning base points are in the same plane. Then, it takes the spatial deflection angle of the plane relative to the standard eye position reference plane and the centroid offset distance as the output to obtain the strabismus three-dimensional mapping result.

[0037] In the above embodiments, this embodiment integrates the spatial coordinates of extraocular muscles, the offset of corneal light reflection points, and the coordinates of the eyeball rotation center extracted from multi-angle images into node attributes with anatomical attachment relationships through an initial heterogeneous atlas construction module. This establishes a directed connection structure between muscles and the eyeball and bone wall, forming a heterogeneous atlas basis that can characterize the spatial distribution and motion association of extraocular muscles. Subsequently, the atlas segmentation module generates feature vectors through weighted aggregation of neighborhood node attributes and performs node clustering based on Euclidean distance thresholds, dividing the heterogeneous atlas into mutually exclusive sub-atlases corresponding to different extraocular muscle spatial configurations, thus completing the automatic classification of anatomical structures and functional features. Finally, the positioning base point module quantifies the spatial deflection angle and centroid offset distance of each sub-atlas relative to the standard eye position reference plane through coordinate system-one mapping and positioning base point planarization calibration, thereby systematically outputting the pathological posture mapping results of the three-dimensional spatial configuration of extraocular muscles in strabismic patients.

[0038] Figure 7 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0039] The electronic device may include a central processing unit / microprocessor / main control chip 4; and a storage medium 5 coupled to the central processing unit / microprocessor / main control chip 4 and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by the processor.

[0040] The central processing unit / microprocessor / main control chip 4 may include, but is not limited to, one or more processors or microprocessors.

[0041] Storage medium 5 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0042] In addition, the electronic device may include (but is not limited to) a data bus 6, an input / output bus / external bus / device bus 7, a display 8, and input / output devices 9 (e.g., keyboard, mouse, speaker, etc.).

[0043] The central processing unit / microprocessor / main control chip 4 can communicate with external devices (8, 9, etc.) via wired or wireless networks (not shown) through the input / output bus / external bus / device bus 7.

[0044] The storage medium 5 may also store at least one computer-executable instruction for performing the steps of various functions and / or methods in the embodiments described herein when the central processing unit / microprocessor / main control chip 4 is running.

[0045] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0046] Figure 8 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0047] like Figure 8 As shown, the non-transitory computer-readable storage medium 11 stores instructions, such as computer-readable instructions 10. When the computer-readable instructions 10 are executed by a processor, the various methods described above can be performed. The non-transitory computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium 11 can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions 10 stored on the non-transitory computer-readable storage medium 11, the various methods described above can be performed.

[0048] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0049] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0050] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0051] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods of the various embodiments of this invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0052] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge graph-based method for 3D mapping of strabismus, characterized in that, Includes the following steps: The coordinates of all nodes in each sub-atlas corresponding to a certain type of extraocular muscle spatial configuration are mapped to a unified three-dimensional coordinate system. The average spatial coordinates of all nodes in each sub-atlas are taken as the positioning base point of the sub-atlas. The normal deflection angle of each positioning base point relative to the horizontal plane and sagittal plane of the eyeball is calculated. The relative positions between the sub-atlases are adjusted so that all positioning base points are in the same plane. Then, the spatial deflection angle of the plane relative to the standard eye position reference plane and the centroid offset distance are used as outputs to obtain the strabismus three-dimensional mapping results.

2. The knowledge graph-based 3D mapping method for strabismus as described in claim 1, characterized in that, The process of obtaining strabismus 3D mapping results includes the following steps: The coordinates of the positioning base points of each sub-map are extracted sequentially. Using the horizontal plane of ocular anatomy as the reference plane, the normal deflection angle of each positioning base point relative to the horizontal plane of ocular anatomy is calculated. Then, using the sagittal plane of ocular anatomy as the reference plane, the normal deflection angle of each positioning base point relative to the sagittal plane is calculated. The two normal deflection angles corresponding to each positioning base point are combined to form the dual-angle offset of that base point. The base point with the smallest dual-angle offset modulus among all positioning base points is selected as the fixed base point, and its spatial coordinates remain unchanged. Rotation correction is performed on the deflection angle to make the horizontal normal deflection angle and the sagittal normal deflection angle of the dual-angle offset of each positioning base point zero. After correction, a new set of positioning base point coordinates is obtained, and all the corrected positioning base points are located on the intersection of the horizontal plane and the sagittal plane of the eyeball anatomy. Using the intersection line as the rotation axis, all the corrected positioning base points are rotated at the same time to minimize the sum of the squares of the Euclidean distances between the positioning base points in the direction perpendicular to the rotation axis. When the rotation stops, all positioning base points are located in the same plane. All coplanar positioning base points are fitted to a least-squares plane. The angle between the normal direction of the least-squares plane and the normal direction of the standard eye-position reference plane is calculated. The angle is decomposed into two components around the horizontal and vertical axes of the three-dimensional coordinate system to obtain the spatial deflection angle. The three-dimensional straight-line distance between the position of the geometric center point of the least-squares plane in the three-dimensional coordinate system and the geometric center point of the standard eye-position reference plane is calculated to obtain the centroid offset distance. The spatial deflection angle and the centroid offset distance are used as outputs to obtain the 3D mapping result of the strabismus.

3. The knowledge graph-based 3D mapping method for strabismus as described in claim 2, characterized in that, The process of obtaining the centroid offset distance includes the following steps: From all the coplanar positioning base points, extract the aggregated feature vector of all nodes in the sub-map corresponding to each positioning base point, and calculate the sum of the magnitudes of the aggregated feature vectors of all nodes in each sub-map. Use the sum as the spatial weight value of the positioning base point. Multiply the horizontal, vertical, and ordinate of the spatial coordinates of each positioning base point by the spatial weight value of the positioning base point to obtain the weighted coordinate components of the positioning base point. Accumulate the weighted coordinate components of all positioning base points in terms of horizontal, vertical, and ordinate respectively to obtain three sums. Divide each sum by the sum of all spatial weight values ​​to obtain the horizontal, vertical, and ordinate of the weighted center point. Use the weighted center point as the geometric center point of the least squares plane. Three pre-calibrated fixed points are selected on the standard eye position reference plane and designated as the first fixed point, the second fixed point, and the third fixed point, respectively. The magnitudes of the dual-angle offsets of all positioning base points are extracted, and the arithmetic mean of these magnitudes is calculated. Each component of the spatial coordinates of the first fixed point is multiplied by the arithmetic mean to obtain the coordinates of the first offset point. Each component of the spatial coordinates of the second fixed point is multiplied by the arithmetic mean to obtain the coordinates of the second offset point. Each component of the spatial coordinates of the third fixed point is multiplied by the arithmetic mean to obtain the coordinates of the third offset point. The corresponding components of the coordinates of the first, second, and third offset points are added together and divided by three to obtain the coordinates of the offset center point. The magnitudes of the dual-angle offsets of the fixed base points are extracted as reference coefficients. Each coordinate component of the offset center point is divided by the reference coefficients to obtain the corrected coordinates of the standard center point. The corrected standard center point is used as the geometric center point of the standard eye position reference plane. Subtract the x-coordinate of the geometric center point of the least squares plane from the x-coordinate of the geometric center point of the standard eye reference plane to obtain the x-coordinate deviation; subtract the y-coordinates to obtain the y-coordinate deviation; subtract the y-coordinates to obtain the y-coordinate deviation; subtract the y-coordinates to obtain the y-coordinate deviation. Project the horizontal, vertical, and triangular coordinate deviations onto the three coordinate planes of the three-dimensional coordinate system to obtain three projection deviation vectors. Calculate the magnitude of each projection deviation vector to obtain three projection distances. Arrange the three projection distances in descending order of value, and take the distance value in the middle position after arrangement. Output this distance value as the three-dimensional straight-line distance, which is the centroid offset distance.

4. The knowledge graph-based 3D mapping method for strabismus as described in claim 3, characterized in that, The process of obtaining the three projected distances includes the following steps: Extract the obtained x-coordinate deviation and y-coordinate deviation, and treat these two deviation values ​​as two mutually perpendicular line segments. Set the common starting point of the two line segments as the origin of the three-dimensional coordinate system. Move the length of the x-coordinate deviation along the x-axis from the origin to reach the first intermediate point. Then move the length of the y-coordinate deviation along the y-axis from the first intermediate point to reach the second intermediate point. Connect the origin and the second midpoint to obtain a straight line segment. Measure the length of the straight line segment and use it as the first projected distance. Extract the obtained horizontal and vertical coordinate deviations, and treat the two deviation values ​​as two mutually perpendicular line segments. Set the common starting point of the two line segments as the origin of the three-dimensional coordinate system. Move the length of the horizontal coordinate deviation along the horizontal axis from the origin to reach the third intermediate point. Then move the length of the vertical coordinate deviation along the vertical axis from the third intermediate point to reach the fourth intermediate point. Connect the origin and the fourth intermediate point to obtain a straight line segment. Measure the length of the straight line segment as the second projected distance. Extract the obtained ordinate deviation and vertical coordinate deviation, and treat these two deviation values ​​as two mutually perpendicular line segments. Set the common starting point of the two line segments as the origin of the three-dimensional coordinate system. Starting from the origin, move the length of the ordinate deviation along the ordinate axis to reach the fifth intermediate point. Then, move the length of the vertical coordinate deviation along the vertical axis from the fifth intermediate point to reach the sixth intermediate point. Connect the origin and the sixth midpoint to obtain a straight line segment. Measure the length of the straight line segment and use it as the third projected distance.

5. The knowledge graph-based 3D mapping method for strabismus as described in claim 1, characterized in that, It also includes acquiring multi-angle image sequences of the eyes of strabismus patients, extracting the origin spatial coordinates, insertion spatial coordinates, and the path point of the muscle belly midline of each extraocular muscle; using the origin spatial coordinates, the offset of the patient's corneal light reflection point, and the coordinates of the eyeball rotation center as attribute values ​​of the entity nodes, and establishing directed connection edges between nodes based on the fixed attachment relationships between muscles and eyeballs and between muscles and bone walls in anatomy, and all entity nodes and directed connection edges constitute an initial heterogeneous atlas.

6. The knowledge graph-based 3D mapping method for strabismus as described in claim 5, characterized in that, The process of forming an initial heterogeneous graph from all entity nodes and directed connecting edges includes the following steps: The spatial coordinates of the origin of each extraocular muscle extracted from the multi-angle image sequence are spatially aligned with the offset of the corneal light-reflection point and the coordinates of the eye rotation center corresponding to the same extraocular muscle, so that the starting point of the three-dimensional vector of the corneal light-reflection point offset is translated to the coordinates of the eye rotation center; then the spatial coordinates of the origin are corrected in the opposite direction of the offset to obtain the corrected spatial coordinates of the origin. The corrected starting point spatial coordinates, corneal light reflection point offset, and eyeball rotation center coordinates are stored as three parallel attribute values ​​of the same entity node, forming a muscle starting point entity node; based on the fixed attachment relationship between muscles and eyeballs, and muscles and bone walls in anatomy, the starting point entity node of each extraocular muscle points to the node where the eyeball rotation center coordinates are located through directed connecting edges. Directed connecting edges are sent from the orbital bone wall nodes to the muscle origin entity nodes. All muscle origin entity nodes, eye rotation center nodes, and orbital bone wall nodes are connected sequentially according to the direction rules. All established entity nodes and directed connecting edges constitute the initial heterogeneous graph.

7. The knowledge graph-based 3D mapping method for strabismus as described in claim 6, characterized in that, The process of connecting all muscle origin nodes, eye rotation center nodes, and orbital bone wall nodes sequentially according to directional rules includes the following steps: The corrected origin spatial coordinates stored in the obtained muscle origin entity node are subjected to spatial difference operation with the eye rotation center coordinates to obtain a vector pointing from the corrected origin spatial coordinates to the eye rotation center coordinates; the direction of the vector is used as the direction of the directed connection edge, and the magnitude of the vector is used as the weight value of the edge to establish a directed connection edge pointing from the muscle origin entity node to the node where the eye rotation center coordinates are located. One directed connection edge is generated for each extraocular muscle. Each muscle origin entity node is spatially superimposed with the opposite direction vector of the corneal light-reflecting point offset used to correct the origin spatial coordinates to obtain the offset virtual bone wall attachment point coordinates; the bone wall point with the closest spatial distance to the virtual bone wall attachment point coordinates is searched in the pre-established orbital bone wall point cloud, and the bone wall point is set as the orbital bone wall node; a directed connection edge is established from the orbital bone wall node to the corresponding muscle origin entity node; the direction is from the bone wall to the muscle, and the edge attribute records the residual vector between the virtual bone wall attachment point and the real bone wall point; All the directed edges generated from the muscle origin entity node to the eye rotation center node, all the directed edges generated from the orbital bone wall node to the muscle origin entity node, as well as all the muscle origin entity nodes, all the involved eye rotation center nodes, and all the determined orbital bone wall nodes, are gathered together according to the direction of the directed edges. This ensures that every path starting from the orbital bone wall node, passing through the muscle origin entity node, and finally reaching the eye rotation center node remains continuous and does not form any reverse loops. The union of all paths constitutes the initial heterogeneous graph.

8. The knowledge graph-based 3D mapping method for strabismus as described in claim 1, characterized in that, It also includes weighted aggregation of the attribute values ​​of the neighboring nodes connected by the directed edges of each node in the obtained initial heterogeneous graph to obtain the aggregated feature vector of the node; and hierarchical division of the nodes according to the Euclidean distance between the aggregated feature vectors of all nodes, and grouping the nodes with Euclidean distance less than a preset threshold into the same subgraph, so that the initial heterogeneous graph is divided into several non-overlapping subgraphs, each subgraph corresponding to a type of extraocular muscle spatial configuration.

9. The knowledge graph-based 3D mapping method for strabismus as described in claim 8, characterized in that, The process of hierarchically partitioning nodes based on the Euclidean distance between the aggregate feature vectors of all nodes includes the following steps: Based on the aggregated feature vector of each node, sort all the aggregated feature vectors of the nodes in ascending order of vector magnitude, and take the aggregated feature vector of the node with the smallest magnitude as the first-level reference vector. Calculate the Euclidean distance between the aggregated feature vector of each of the remaining nodes and the first-level reference vector. Nodes with an Euclidean distance less than a preset threshold, along with the nodes corresponding to the first-level reference vector, are included in the first-level candidate set. All nodes in the first-level candidate set are removed from the full set of nodes. In the remaining node set after removing the first-level candidate set, sort the nodes again according to the magnitude of the aggregated feature vector from smallest to largest, and take the aggregated feature vector of the node with the smallest magnitude as the second-level reference vector; calculate the Euclidean distance between the aggregated feature vector of each remaining node and the second-level reference vector, and include nodes whose Euclidean distance is less than the same preset threshold, along with the nodes corresponding to the second-level reference vector, into the second-level candidate set, and remove all nodes in the second-level candidate set from the remaining node set; repeat until the remaining node set is empty, and obtain several level candidate sets; For each hierarchical candidate set, extract the directed edges of all nodes in the hierarchical candidate set in the constructed initial heterogeneous graph, and check whether there is at least one path consisting of directed edges between any two nodes in the hierarchical candidate set that does not cross nodes outside the set. If there exists a path between two nodes that does not cross any node outside the set, then the hierarchical candidate set is split into two subsets, such that there exists at least one path consisting of directed edges between any two nodes within each subset that does not cross any node outside the subset. After all the splits are completed, each subset is a subgraph, and all subgraphs do not overlap. Each subgraph corresponds to a type of extraocular muscle spatial configuration.

10. A knowledge graph-based 3D strabismus mapping system, used to implement the knowledge graph-based 3D strabismus mapping method as described in any one of claims 1 to 9, characterized in that, include: The initial heterogeneous atlas construction module is responsible for acquiring multi-angle image sequences of strabismus patients' eyes, extracting the origin spatial coordinates, insertion spatial coordinates, and muscle belly midline path points of each extraocular muscle; using the origin spatial coordinates, the offset of the patient's corneal light reflection point, and the coordinates of the eyeball rotation center as attribute values ​​of the entity nodes; establishing directed connection edges between nodes based on the fixed attachment relationships between muscles and eyeballs, and muscles and bone walls in anatomy; and all entity nodes and directed connection edges constitute the initial heterogeneous atlas. The graph segmentation module is responsible for weighting and aggregating the attribute values ​​of the neighboring nodes connected by the directed edges of each node in the initial heterogeneous graph to obtain the aggregated feature vector of the node; and dividing the nodes into hierarchical levels according to the Euclidean distance between the aggregated feature vectors of all nodes, and classifying the nodes with Euclidean distance less than a preset threshold into the same subgraph, so that the initial heterogeneous graph is divided into several non-overlapping subgraphs, each subgraph corresponding to a type of extraocular muscle spatial configuration. The positioning base point module is responsible for mapping the coordinates of all nodes in each sub-map to a unified three-dimensional coordinate system. It takes the average spatial coordinates of all nodes in each sub-map as the positioning base point of the sub-map, calculates the normal deflection angle of each positioning base point relative to the horizontal and sagittal planes of the eyeball anatomy, and adjusts the relative positions between sub-maps so that all positioning base points are in the same plane. Then, it outputs the spatial deflection angle of the plane relative to the standard eye position reference plane and the centroid offset distance to obtain the strabismus three-dimensional mapping result.

Images