Method and system for automatic labeling of bone and joint lesion areas based on image processing

Abstract

The present application discloses an automatic labeling method and system for bone and joint lesion areas based on image processing, which belongs to the field of bone and joint lesion area labeling. The present application obtains bone and joint reflection signals through a polarization-adjustable laser light source combined with polarization-sensitive optical coherence tomography, and generates a three-dimensional birefringence distribution map through matrix analysis. After the distribution map is subjected to non-local mean filtering, the lesion edge contour is extracted based on the dynamic gradient amplitude to generate a binary mask. The mask pixels are converted into network nodes containing spatial coordinates, gradient direction and birefringence, and a topological network is constructed based on the Euclidean distance of adjacent pixels. The weights are calculated by node intersection density and birefringence, and the graph attention network is used to aggregate multi-layer features to extract lesion topological features. The features are mapped to three-dimensional space to construct a lesion probability distribution field, and interactive labeling results are generated through isosurface extraction and transparency mixed rendering. The present application can improve the accuracy of labeling complex bone and joint lesion areas.

Description

Technical Field

[0001] The present application belongs to the field of bone and joint lesion area annotation, and in particular relates to a method and system for automatic bone and joint lesion area annotation based on image processing. Background Art

[0002] Accurate annotation of bone and joint lesions is a crucial foundation for diagnosis and treatment planning, particularly in personalized medicine and surgical navigation. Traditional annotation methods rely on manual interpretation of imaging features or semi-automatic segmentation, but their efficiency and consistency struggle to meet the demands of clinical refinement. There is an urgent need to develop high-precision, automated annotation technologies.

[0003] Existing technologies for annotating bone and joint lesions primarily rely on threshold segmentation, region growing algorithms, and machine learning models. For example, imaging technologies such as X-rays and MRIs often extract lesion regions using grayscale threshold segmentation or utilize morphological operations to enhance feature contrast. Deep learning methods have also been used for bone and joint image segmentation.

[0004] However, existing methods have significant drawbacks in labeling complex lesions such as multi-branched fissures and three-dimensional osteophytes. Threshold segmentation is susceptible to noise interference and has difficulty distinguishing the gradient boundary between lesions and normal tissue. Edge detection is insufficiently sensitive to the texture features of microcracks, resulting in missed detections or fractures. Deep learning methods rely on large-scale annotated data and have limited modeling capabilities for complex topological structures. Therefore, existing methods for labeling bone and joint lesion regions suffer from the technical problem of poor labeling accuracy when labeling complex lesion regions. Summary of the Invention

[0005] The present application provides a method and system for automatic labeling of bone and joint lesion areas based on image processing, which can improve the accuracy of labeling complex bone and joint lesion areas.

[0006] In a first aspect, the present application provides an automatic annotation method for bone and joint lesion areas based on image processing, which is applicable to bone and joint lesion areas with osteophytes or cartilage fissures. The method comprises:

[0007] A polarization-tunable laser light source is used to emit light signals into the bone joint area, and polarization-sensitive optical coherence tomography is used to receive the reflected signals. The birefringence distribution is calculated using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information.

[0008] The birefringence distribution map is subjected to non-local mean filtering, and the edge contour of the target lesion area is extracted based on the gradient amplitude range, generating binary mask data that retains the microcrack texture characteristics. The gradient amplitude range is dynamically adjusted based on the local structural characteristics of the birefringence distribution map.

[0009] The pixels in the binary mask data are converted into nodes, and a topological network representing the fissure branching and osteophyte morphology is constructed through the connection relationship between nodes. The node attributes include spatial coordinates, gradient direction, and birefringence. The connection relationship between nodes is determined by the Euclidean distance between adjacent pixels in the binary mask data.

[0010] Based on the connection relationship between nodes, the node weight is determined according to the node intersection density and birefringence, and a graph attention network is used to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area;

[0011] The lesion topological features are mapped to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field. Based on the lesion probability distribution field, isosurface extraction and transparency mixed rendering are performed to generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display.

[0012] In one feasible implementation, based on the connection relationship between nodes, the node weight of the node is determined according to the intersection density and birefringence of the node, and a graph attention network is used to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area, including:

[0013] Based on the connection relationship between nodes, the node weight is determined according to the node's intersection density, birefringence and gradient direction consistency, and a dual-channel graph attention network is used to aggregate node attributes and geometric relationship features respectively. After fusion, the lesion topology characteristics of the target lesion area are extracted. The intersection density is the ratio of the number of node adjacent edges to the preset maximum number of connection edges.

[0014] In one feasible implementation, based on the connection relationship between nodes, the node weight of the node is determined according to the node intersection density, birefringence and gradient direction consistency, and a dual-channel graph attention network is used to aggregate node attributes and geometric relationship features respectively. After fusion, the lesion topology features of the target lesion area are extracted, including:

[0015] Calculate the node's gradient direction consistency parameter based on the angle between the node's gradient direction and the gradient direction of the adjacent node;

[0016] Based on the node intersection density, birefringence and gradient direction consistency parameters, the initial node weight of the node is generated by weighted summation;

[0017] Based on the initial node weights, the spatial coordinates, birefringence and gradient direction of the nodes are weighted aggregated to generate node attribute features that characterize the biochemical characteristics of the lesion;

[0018] According to the ratio of the Euclidean distance of the connection relationship between nodes and the initial node weight, the geometric relationship strength of the nodes is constructed to generate geometric relationship features that characterize the morphological distribution of lesions;

[0019] The node attribute features and geometric relationship features are input into the dual-channel graph attention network, and the node attribute features and geometric relationship features are weightedly fused according to the initial node weights to generate lesion topology features including fissure branch topology structure and osteophyte morphology.

[0020] In one feasible embodiment, a polarization-tunable laser light source is used to transmit a light signal to the bone joint region, a reflected signal is received by polarization-sensitive optical coherence tomography, and a birefringence distribution is calculated using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information, including:

[0021] By adjusting the wavelength and polarization direction parameters of the polarization-tunable laser light source, multi-angle polarized light signals are emitted to the surface and deep layers of the cartilage in the bone joint area. The multi-angle polarized light signals cover the collagen fiber arrangement direction in the surface and deep layers of the cartilage.

[0022] Receive interference signals from multi-angle polarized light signals reflected at different depths in the bone joint area, and analyze the local birefringence value of the cartilage based on the phase delay and polarization state changes of the interference signals. The phase delay and polarization state changes are caused by the differences in the orientation of the collagen fibers in the cartilage.

[0023] The local birefringence value is mapped to the three-dimensional spatial coordinate system of the bone joint area to generate a birefringence gradient distribution map associated with the collagen fiber orientation to obtain a three-dimensional birefringence distribution map, in which the gradient direction is orthogonal to the collagen fiber arrangement direction.

[0024] In one feasible embodiment, non-local mean filtering is performed on the birefringence distribution map, and the edge contour of the target lesion area is extracted based on the gradient amplitude range to generate binary mask data that retains the microcrack texture features, including:

[0025] According to the consistency of collagen fiber arrangement direction in adjacent areas of the birefringence distribution map, the local structural uniformity parameter of the birefringence distribution map is calculated. The local structural uniformity parameter is used to characterize the fiber continuity in the bone joint area.

[0026] Adjusting the gradient amplitude range based on the local structural uniformity parameter, expanding the gradient amplitude range in areas where the local structural uniformity parameter is lower than a set threshold, and narrowing the gradient amplitude range in areas where the local structural uniformity parameter is higher than the set threshold;

[0027] The birefringence distribution map is traversed according to the adjusted gradient amplitude range, and the pixel points within the gradient amplitude range and matching the gradient direction are marked as edge contours to generate binary mask data including the topological connection relationship of microcracks.

[0028] In one feasible embodiment, pixels in the binary mask data are converted into nodes, and a topological network representing the fissure branches and osteophyte morphology is constructed through the connection relationship between the nodes, including:

[0029] Traverse the pixels in the binary mask data and mark the pixels that meet the preset gradient direction continuity as initial nodes. The preset gradient direction continuity is determined by the angle between the gradient direction of the pixel and the gradient direction of the adjacent pixel and the preset angle threshold;

[0030] Based on the spatial coordinates, gradient direction and birefringence of the initial nodes, the spatial distribution density and birefringence difference of the initial nodes are calculated, and the initial nodes with birefringence difference greater than a preset difference threshold are selected as target nodes;

[0031] With the target node as the center, connecting edges are generated according to the Euclidean distance of adjacent pixels in the binary mask data to construct a topological network representing the fissure branches and osteophyte morphology. The maximum connection distance of the connecting edges is determined by the difference in birefringence.

[0032] In one feasible embodiment, the lesion topological features are mapped to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field. Based on the lesion probability distribution field, isosurface extraction and transparency mixed rendering are performed to generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display, including:

[0033] According to the target node weights in the lesion topological features and the spatial coordinates of the three-dimensional birefringence distribution map, a probability mapping relationship between the target node weights and the spatial coordinates is established to generate a three-dimensional probability distribution field including the lesion probability gradient;

[0034] Based on the lesion probability gradient direction of the three-dimensional probability distribution field and the collagen fiber orientation information of the birefringence distribution map, the isosurface boundary of the lesion probability distribution field is determined. The extraction range of the isosurface boundary is determined by the product of the lesion probability gradient and the birefringence difference.

[0035] The isosurface boundary is aligned with the spatial coordinates of the three-dimensional birefringence distribution map, transparency parameters are assigned according to the birefringence difference, and transparency blending rendering is performed on the area covered by the isosurface boundary to generate interactive annotation results that support multi-plane slice display.

[0036] In a second aspect, the present application provides an automatic annotation system for bone and joint lesion areas based on image processing, which is applicable to bone and joint lesion areas with osteophytes or cartilage fissures. The system includes:

[0037] The transmitting module is used to transmit light signals to the bone joint area through a polarization-tunable laser light source, receive the reflected signals through polarization-sensitive optical coherence tomography, and calculate the birefringence distribution using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information;

[0038] An extraction module, which performs non-local mean filtering on the birefringence distribution map and extracts the edge contour of the target lesion area based on the gradient amplitude range, generating binary mask data that retains the texture characteristics of microcracks. The gradient amplitude range is dynamically adjusted based on the local structural characteristics of the birefringence distribution map;

[0039] A construction module is used to convert pixels in the binary mask data into nodes and construct a topological network representing the fissure branching and osteophyte morphology through the connection relationship between nodes. The node attributes include spatial coordinates, gradient direction, and birefringence. The connection relationship between nodes is determined by the Euclidean distance between adjacent pixels in the binary mask data.

[0040] The extraction module is also used to determine the node weight of the node based on the connection relationship between the nodes, the intersection density and the birefringence of the nodes, and use the graph attention network to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area;

[0041] A generation module is used to map the lesion topological features to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field. Based on the lesion probability distribution field, isosurface extraction and transparency mixed rendering are performed to generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display.

[0042] In a third aspect, the present application provides an electronic device comprising: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement a method for automatic labeling of bone and joint lesion areas based on image processing as in any embodiment of the first aspect.

[0043] In a fourth aspect, the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, an automatic labeling method for bone and joint lesion areas based on image processing as in any one of the embodiments of the first aspect is implemented.

[0044] This application implements an automatic labeling method and system for bone and joint lesion areas based on image processing. It obtains a three-dimensional birefringence distribution map through polarization-sensitive optical coherence tomography, which can accurately capture subtle changes in collagen fiber orientation and effectively distinguish the gradual boundary between diseased tissue and normal tissue. A dynamic gradient amplitude adjustment strategy is combined with non-local mean filtering to suppress artifact interference while retaining the continuous texture features of microcracks, solving the problem of insufficient sensitivity of edge detection to subtle structures. By constructing a topological network and fusing a dual-channel graph attention mechanism that integrates node attributes and geometric relationships, adaptive modeling of complex structures such as multi-branched fissures and three-dimensional osteophytes is achieved, breaking through the limitations of traditional deep learning models on data scale and topological modeling capabilities. Based on the three-dimensional visualization technology of the lesion probability distribution field, the pathological features are spatially associated with the anatomical structure, providing interactive labeling results for clinical use. While reducing the need for manual intervention, this application improves the accuracy of labeling complex lesion areas.

[0045] Furthermore, based on the joint calculation of gradient direction consistency parameters and intersection density, the continuous direction of microcracks and their branch intersection characteristics can be effectively identified. Secondly, by introducing a dual-channel graph attention network to aggregate node attribute features and geometric relationship features respectively, the collaborative modeling of the biochemical characteristics and morphological distribution of the lesion area is achieved, breaking through the one-sidedness of single feature analysis in the representation of complex structures. Through the adaptive fusion mechanism of geometric relationship strength and node weight, the network's ability to discriminate the three-dimensional morphology of osteophytes and the spatial distribution of fracture branches is enhanced. Through refined feature extraction and multi-level topological association analysis, this method significantly improves the ability to analyze gradient boundaries and complex lesion structures while reducing dependence on large-scale annotated data, and improves the annotation accuracy of bone and joint lesion areas with multi-branched fractures and osteophyte morphology. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0047] Figure 1 This is a flowchart of a method for automatically labeling bone and joint lesion areas based on image processing provided by an embodiment of the present application;

[0048] Figure 2 This is a flowchart of a method for automatically labeling bone and joint lesion areas based on image processing provided by an embodiment of the present application;

[0049] Figure 3 This is a flowchart of a method for automatically labeling bone and joint lesion areas based on image processing provided by an embodiment of the present application;

[0050] Figure 4 This is a schematic diagram of the structure of an automatic labeling system for bone and joint lesion areas based on image processing provided by an embodiment of the present application;

[0051] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0052] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0053] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0054] In existing technologies, the small, irregular branching structure of multi-branch fissures easily leads to the failure of traditional algorithms to capture local features. The heterogeneous calcification morphology of 3D osteophytes and the low contrast between them and the surrounding tissues lead to blurred boundaries in threshold-based segmentation. Deep learning methods are limited by their 3D spatial topological modeling capabilities and the generalization of complex lesion features under small sample sizes, making it difficult to accurately analyze the multi-scale geometric characteristics and spatial continuity of such lesions. Therefore, existing bone and joint lesion region annotation methods have a technical problem of poor annotation accuracy when annotating complex lesion regions.

[0055] To solve the problems of the prior art, the present invention provides an automatic labeling method, system, device and computer storage medium for bone and joint lesion regions based on image processing. The following first introduces the automatic labeling method for bone and joint lesion regions based on image processing provided by the present invention.

[0056] Figure 1 FIG. 1 is a flow chart of an automatic labeling method for bone and joint lesion areas based on image processing provided by an embodiment of the present application. Figure 1 As shown, the method includes steps S110 to S150. The method is applicable to bone and joint lesion areas with osteophytes or cartilage fissures, and the method includes:

[0057] S110: A light signal is emitted to the bone joint area through a polarization-tunable laser light source, a reflected signal is received through polarization-sensitive optical coherence tomography, and the birefringence distribution is calculated using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information.

[0058] Osteophytes or cartilage fissures are abnormal structures formed in joints due to degeneration or damage of the articular cartilage or bone. Osteophytes appear as abnormal growths at the edges of the bone, while cartilage fissures are microscopic cracks within the cartilage tissue. A polarization-tunable laser light source is an optical device capable of adjusting the polarization direction of the output light wave. This allows for multi-angle detection by varying the relative angle between the polarization state of the incident light and the orientation of the cartilage fibers. Polarization-sensitive optical coherence tomography (PS-OCT) is a technique that combines optical coherence tomography with polarization analysis. It analyzes the birefringence properties of a sample by detecting changes in the polarization state of reflected light. The birefringence distribution describes the differences in birefringence values at different locations in the tissue, reflecting the degree of order in the collagen fiber arrangement. Collagen fiber orientation information refers to fiber orientation data derived from the direction of the birefringence gradient. A three-dimensional birefringence distribution map is a stereoscopic image created by mapping the birefringence values to a three-dimensional coordinate system. It is used to characterize three-dimensional abnormalities in fiber arrangement.

[0059] First, the laser light source is controlled to output a probe beam with a preset wavelength and polarization direction, performing multi-angle scanning of the bone joint region. The reflected light signal is transmitted to an interferometer via a polarization-maintaining fiber, where it interferes with a reference beam to generate a time-domain signal carrying polarization information. A polarization-resolution algorithm based on the Jones matrix is then used to calculate the birefringence value at each sampling point based on the phase difference of the interference signal and the change in polarization state. A three-dimensional scanning mechanism controls the lateral and axial movement of the probe, mapping the birefringence values measured point by point to a spatial coordinate system. Finally, the orthogonality of the birefringence gradient direction to the collagen fiber alignment direction is utilized to generate a three-dimensional distribution map showing the fiber orientation. The phase difference is calculated by comparing the interference fringe displacement of the incident and reflected light with different polarization states. The birefringence gradient direction is determined by the direction of maximum birefringence change between adjacent voxels.

[0060] S120: Performing non-local mean filtering on the birefringence distribution map, and extracting the edge contour of the target lesion area based on the gradient amplitude range to generate binary mask data that retains microcrack texture features. The gradient amplitude range is dynamically adjusted based on the local structural characteristics of the birefringence distribution map.

[0061] Non-local mean filtering is a method for noise reduction by calculating the weighted average of similar regions in an image. Its core lies in utilizing the global structural similarity of the image rather than local neighborhood information. The gradient amplitude range refers to the effective interval of the gradient intensity value of the pixel point in the image, which is used to distinguish edges from background noise. The microcrack texture feature characterizes the fiber fracture area with a width less than millimeters in bone and joint lesions, which is manifested as a sudden change in the local gradient direction and numerical discontinuity in the birefringence distribution map. Binarized mask data is a binary image that classifies image pixels into edge and non-edge areas, where edge pixels are marked as foreground. Local structural characteristics refer to the spatial correlation between the arrangement direction and density of collagen fibers in adjacent areas in the birefringence distribution map, reflecting the continuity of tissue structure.

[0062] First, a non-local mean filter is applied to the birefringence distribution map. By calculating the weight coefficients between all similar blocks in the image, a weighted average is performed on the current pixel, suppressing speckle noise while preserving the texture details of microcracks. Subsequently, the gradient amplitude range is dynamically adjusted based on the local structural characteristics of the filtered image: the gradient threshold is reduced in areas with uniform collagen fibers to reduce noise interference, while the gradient threshold is increased in areas with broken or disordered fibers to capture the edges of microcracks. By traversing the entire image, pixels that meet the dynamic gradient threshold and have continuous direction are extracted as edge contours. Finally, the edge pixels are mapped to binary mask data, where the topological connectivity of microcracks is maintained by the consistency of the gradient direction of adjacent edge pixels.

[0063] S130: Convert the pixels in the binary mask data into nodes, and construct a topological network representing the fissure branches and osteophyte morphology through the connection relationship between the nodes. The node attributes of the nodes include spatial coordinates, gradient direction and birefringence. The connection relationship between the nodes is determined by the Euclidean distance between adjacent pixels in the binary mask data.

[0064] Spatial coordinates represent the location of a node in a three-dimensional coordinate system. The gradient direction represents the direction orthogonal to the collagen fiber alignment at that node in the birefringence distribution, characterizing the anisotropic characteristics of the local tissue. The Euclidean distance between adjacent pixels refers to the geometric distance between adjacent edge pixels in the binary mask data in three-dimensional space and is used to quantify the strength of spatial associations between nodes.

[0065] First, the edge pixels in the binary mask data are traversed, and each valid pixel is converted into a graph node. The node attributes inherit the spatial coordinates, gradient direction, and birefringence value of the corresponding position in the original birefringence distribution map. Subsequently, the connection relationship is established based on the Euclidean distance of adjacent pixels. The three-dimensional spatial distance between each node and its neighboring pixels is calculated. If the distance is less than a preset threshold, a connecting edge is established. The connection relationship is further optimized. The connection strength is strengthened for adjacent nodes with continuous gradient directions and significant birefringence differences, while the connection is weakened for nodes with sudden gradient changes or similar birefringence. By iteratively traversing all nodes, a topological network representing the morphology of fissure branches and osteophytes is formed. Among them, the fissure area appears as a long chain-like connection structure, while the osteophyte area presents a dense mesh connection pattern. The threshold setting of the connecting edge needs to meet the spatial scale smaller than the typical fissure width. It can be dynamically adjusted according to the spatial resolution of the birefringence distribution map to ensure that the topological network can accurately reflect the lesion morphology.

[0066] S140: Based on the connection relationship between nodes, the node weight of the node is determined according to the intersection density and birefringence of the node, and a graph attention network is used to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area.

[0067] Intersection density is the ratio of the number of edges connecting a node in a topological network to the preset maximum number of possible edges, reflecting the criticality of the node in the lesion structure. The graph attention network is a deep learning model based on graph structures that focuses on key node features through an adaptive weight distribution mechanism. Lesion topology features refer to high-dimensional feature vectors that represent the topological structure of fissure branches and the three-dimensional morphology of osteophytes, generated by fusing node attributes and geometric relationships through a dual-channel graph attention network. These vectors are obtained by weighted aggregation of key node features through a self-attention mechanism and can include information on the geometric connection patterns of fissure branches and the density distribution of osteophytes.

[0068] First, the intersection density is calculated based on the number of adjacent edges of the node, and the initial weight is generated by combining the birefringence value of the node. Nodes with high intersection density and abnormal birefringence are given higher weights to reflect their key role in the lesion structure. Subsequently, a multi-layer graph attention network is used for feature aggregation. Each layer calculates the feature correlation between adjacent nodes through the attention mechanism and dynamically adjusts the feature propagation weight. Through layer-by-layer iteration, the node features gradually fuse local neighborhood information with the global topological structure. Finally, high-order topological features including the extension trend of crack branches and the growth pattern of osteophytes are extracted. Among them, the birefringence difference is used to enhance the characterization of the biochemical characteristics of the lesion area, and the intersection density strengthens the feature expression of key nodes of topological connection.

[0069] S150: Mapping the lesion topological features to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field. Based on the lesion probability distribution field, isosurface extraction and transparency mixed rendering are performed to generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display.

[0070] The lesion probability distribution field is a probability density field formed by mapping the topological features of the lesion to three-dimensional space, which represents the possibility that each spatial location belongs to the lesion area. Isosurface extraction extracts surface boundaries with the same probability value from the three-dimensional field by setting a probability threshold, which is used to define the spatial range of the lesion area. Transparency blending rendering is a technology that overlays the three-dimensional annotation results with the original image data, achieving multi-level visualization by adjusting the transparency of different areas. The interactive annotation result is a visualization model that supports three-dimensional rotation and slicing operations, allowing users to observe the lesion morphology from multiple angles.

[0071] First, the node weights in the lesion topology are associated with three-dimensional spatial coordinates, and a continuous probability distribution field is generated through an interpolation algorithm. Based on the correlation between the probability gradient direction and collagen fiber orientation, an isosurface extraction algorithm is used to determine the lesion boundary, where the probability gradient region corresponds to the crack tip or osteophyte edge. Subsequently, the extracted isosurface is spatially aligned with the original birefringence distribution map, and a transparency parameter is assigned based on the birefringence difference. A gradient transparency is used at the junction of the lesion and normal tissue to highlight the boundary, while the transparency is reduced in the core of the lesion to enhance the display. Finally, a graphics rendering engine is used to enable interactive operation of the 3D model, supporting multi-plane slices to view the internal structure of the lesion.

[0072] This embodiment obtains a three-dimensional birefringence distribution map through polarization-sensitive optical coherence tomography, which can accurately capture subtle changes in collagen fiber orientation and effectively distinguish the gradual boundary between diseased tissue and normal tissue. A dynamic gradient amplitude adjustment strategy is combined with non-local mean filtering to suppress artifact interference while retaining the continuous texture features of microcracks, solving the problem of insufficient sensitivity of edge detection to subtle structures. By constructing a topological network and integrating a dual-channel graph attention mechanism with node attributes and geometric relationships, adaptive modeling of complex structures such as multi-branched cracks and three-dimensional osteophytes is achieved, breaking through the limitations of traditional deep learning models on data scale and topological modeling capabilities. Based on the three-dimensional visualization technology of the lesion probability distribution field, pathological features are spatially associated with anatomical structures, providing interactive annotation results for clinical use. While reducing the need for manual intervention, this application improves the accuracy of annotation of complex lesion areas.

[0073] Existing technologies cannot effectively distinguish the gradient boundary between lesion areas and normal tissues, and their ability to model complex topological structures such as multi-branched fissures is insufficient, resulting in missed detection or fractures. To address the above defects, a gradient direction consistency parameter is introduced to quantify the fiber fracture pattern, and a dual-channel graph attention network is used to parallelly process biochemical characteristics and spatial distribution characteristics to enhance the accurate modeling of complex lesion topological structures.

[0074] In one feasible embodiment, step S140: determining the node weight of the node based on the node connection relationship and the node intersection density and birefringence, and using a graph attention network to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area, including:

[0075] Based on the connection relationship between nodes, the node weight is determined according to the node's intersection density, birefringence and gradient direction consistency, and a dual-channel graph attention network is used to aggregate node attributes and geometric relationship features respectively. After fusion, the lesion topology characteristics of the target lesion area are extracted. The intersection density is the ratio of the number of node adjacent edges to the preset maximum number of connection edges.

[0076] Gradient direction consistency refers to the cosine of the gradient angle between a node and its neighbors, and is used to quantify the local directional continuity of fiber fracture regions. The dual-channel graph attention network is a deep learning model that processes node attributes and geometric relationships in parallel. The attribute channel focuses on biochemical characteristics, while the geometric channel learns spatial distribution patterns, dynamically adjusting feature contributions through an attention mechanism. Node attributes include inherent features such as spatial coordinates, gradient direction, and birefringence, reflecting the physical properties of local tissue. Geometric relationship features are constructed using the distance and weight of the edges connecting nodes to characterize the spatial distribution patterns of lesion morphology.

[0077] First, the gradient direction angle between each node and its adjacent nodes is calculated to generate a directional consistency parameter as an evaluation indicator of the topological connection quality. The intersection density, birefringence anomaly and directional consistency parameter are weighted and fused to generate the initial node weight. The node attribute channel extracts biochemical features reflecting the degree of collagen fiber breakage by weighted aggregation of spatial coordinates, gradient direction and birefringence. The geometric relationship channel constructs the relationship strength based on the inverse ratio of the Euclidean distance of the connecting edge and the initial weight to capture the extension pattern of the crack branch. The dual-channel graph attention network performs attention weighting on the two types of features respectively, and generates topological features including the crack bifurcation pattern and osteophyte density distribution through feature splicing and fusion with the fully connected layer.

[0078] Figure 2 FIG. 1 is a flow chart showing a method for extracting lesion topological features of a target lesion area provided by an embodiment of the present application. Figure 2 As shown, steps S210 to S250 are included.

[0079] In one feasible implementation, based on the connection relationship between nodes, the node weight of the node is determined according to the node intersection density, birefringence and gradient direction consistency, and a dual-channel graph attention network is used to aggregate node attributes and geometric relationship features respectively. After fusion, the lesion topology features of the target lesion area are extracted, including:

[0080] S210: Calculate the gradient direction consistency parameter of the node according to the angle between the gradient direction of the node and the gradient direction of the adjacent node.

[0081] The gradient direction consistency parameter quantifies the continuity of the collagen fiber alignment between a node and its neighbors. In bone and joint lesions, collagen fiber breakage or disorganization can lead to sudden changes in the local gradient direction. This parameter reflects the local directional consistency of the fiber breakage region by calculating the cosine of the angle between the gradient direction of a node and its neighbors.

[0082] First, traverse each node in the topological network and obtain its gradient direction vector and the gradient direction vectors of all adjacent nodes. Calculate the cosine value of the angle between the current node and the gradient direction vector of each adjacent node. The calculation formula is shown in Formula (1).

[0083] (1)

[0084] in, It is the cosine value of the angle between the gradient direction of the node and the adjacent node, and its value range is [-1,1]. The closer the value is to 1, the higher the direction consistency is. is the gradient direction vector of the current node, which represents the arrangement direction of collagen fibers in three-dimensional space. is the gradient direction vector of the adjacent nodes. By comparing the direction vectors of adjacent nodes, the local direction continuity can be quantified.

[0085] Afterwards, the cosine values of all adjacent nodes are averaged to generate the gradient direction consistency parameter of the node, which ranges from [-1, 1]. The closer the value is to 1, the higher the direction consistency.

[0086] S220: Based on the intersection density, birefringence and gradient direction consistency parameter of the node, an initial node weight of the node is generated by weighted summation.

[0087] The birefringence difference refers to the absolute deviation between the birefringence value of a node and the neighborhood mean, reflecting the degree of abnormality of local biochemical characteristics.

[0088] First, the actual number of connected edges of the node is counted and divided by the preset maximum possible number of connected edges to calculate the node intersection density D. Then, the node birefringence value is compared with the normal tissue baseline value to calculate the birefringence abnormality B. Finally, the intersection density, birefringence abnormality, and gradient direction consistency parameters are linearly weighted and summed to obtain the initial node weight. Specifically, the process of calculating the initial node weight is shown in formula (2).

[0089] (2)

[0090] in, is the initial node weight, is the preset weight coefficient, C is the gradient direction consistency parameter; D is the intersection density, , E is the actual number of edges connecting the node, is the preset maximum possible number of connected edges; B is the birefringence anomaly, , b is the node birefringence value, It is the reference value of birefringence of normal tissue.

[0091] S230: Perform weighted aggregation based on the spatial coordinates, birefringence, and gradient direction of the initial node weights to generate node attribute features that characterize the biochemical characteristics of the lesion.

[0092] Node attribute features are multidimensional feature vectors that integrate the node's spatial position, biochemical properties, and orientation. Spatial coordinates reflect the three-dimensional distribution of the lesion, birefringence correlates with the orderliness of collagen fiber alignment, and gradient direction characterizes fiber fracture patterns.

[0093] First, the node spatial coordinates are normalized to eliminate dimensional differences. Next, a node attribute matrix is constructed, with each row containing the normalized coordinates, birefringence values, and gradient direction vectors. Finally, the attribute matrix is weighted and aggregated based on the initial node weights to obtain the node attribute features. The calculation formula is shown in Equation (3).

[0094] (3)

[0095] in, It is a multi-dimensional feature vector that integrates the node attribute characteristics, spatial coordinates, birefringence and gradient direction information; is the initial node weight of the i-th node, which is used to adjust the contribution of node attributes in the aggregation process; It is a node attribute vector, including normalized space coordinates, birefringence value and gradient direction vector.

[0096] S240: According to the ratio of the Euclidean distance of the connection relationship between the nodes to the initial node weight, the geometric relationship strength of the nodes is constructed to generate a geometric relationship feature that characterizes the morphological distribution of the lesion.

[0097] The geometric relationship strength refers to the tightness of the spatial connection between quantitative nodes, and reflects the spatial constraint relationship of the lesion morphology through the ratio of the Euclidean distance to the initial weight.

[0098] First, the three-dimensional geometric distance is calculated based on the standardized spatial coordinates of the nodes to obtain the Euclidean distance between nodes. Then, the geometric relationship strength matrix is constructed based on the initial node weights and the Euclidean distance between nodes. Finally, the geometric relationship features are generated by normalizing the geometric relationship strength matrix. Specifically, the process of calculating the geometric relationship strength is shown in Formula (4).

[0099] (4)

[0100] in, is the geometric relationship strength between the i-th node and the j-th node, is the initial node weight of the i-th node, is the Euclidean distance between the i-th node and the j-th node, , is the node space coordinate.

[0101] S250: Input the node attribute features and the geometric relationship features into the dual-channel graph attention network, perform weighted fusion on the node attribute features and the geometric relationship features according to the initial node weights, and generate lesion topology features including the fissure branch topology structure and osteophyte morphology.

[0102] First, a dual-channel graph attention network is initialized, with independent attention heads set for both channels. The node attribute feature matrix is then input into the attribute channel, and the geometric relationship feature matrix is input into the geometry channel. Within each channel, feature correlation is calculated using the self-attention mechanism, as shown in Equation (5).

[0103] (5)

[0104] in, is the attention weight, and the feature correlation between nodes is calculated through the self-attention mechanism. and are the query vector and key vector, respectively, which are mapped from the node attribute features and geometric relationship features; d is the feature dimension, which is used to scale the attention score to avoid gradient disappearance.

[0105] Afterwards, the two-channel outputs are weightedly fused based on the initial node weights to obtain the lesion topology features. The specific process is shown in formula (6).

[0106] (6)

[0107] in, is the fusion weight coefficient, is the weighted fusion feature, is the node attribute feature, It is a geometric relationship feature.

[0108] In the dual-channel graph attention network processing, node attribute features and geometric relationship features are first input into independent channels for feature learning. The node attribute channel focuses on the biochemical characteristics and spatial distribution of the node itself. The self-attention mechanism calculates the correlation between node attributes and dynamically adjusts feature propagation weights to strengthen the influence of key biochemical indicators on the lesion area. The geometric relationship channel focuses on the spatial connectivity patterns between nodes, using the self-attention mechanism to capture the extension patterns of fissure branches and the reticular distribution characteristics of osteophytes. The geometric relationship strength matrix quantifies the spatial constraints of lesion morphology. Within each channel, multiple layers of attention heads are used to concurrently extract feature representations at different levels of abstraction, forming a multidimensional feature vector.

[0109] Subsequently, the dual-channel outputs are adaptively weighted and fused based on the initial node weights. The initial weights integrate the node's intersection density, birefringence anomaly, and gradient direction consistency, reflecting the importance of the node in the lesion structure. During the fusion process, attribute channels are given higher weights to highlight differences in biochemical characteristics. At the same time, the spatial connection pattern of the geometric channels is combined to generate a composite feature that fuses the fissure branching topology and the osteophyte's three-dimensional morphology. The final output lesion topology features include both the fiber fracture pattern and bifurcation angle information of the microcracks, as well as the density gradient and spatial growth direction of the osteophyte region, forming a multi-dimensional representation of the complex lesion area.

[0110] For example, taking the knee cartilage fissure as an example, the process of extracting the lesion topological features of the target lesion area is as follows. First, the topological network nodes are traversed to calculate the gradient direction consistency parameter. For each node and its adjacent nodes, the cosine value of the gradient direction vector angle is calculated using formula (1). The average is taken to generate the direction continuity index. Assuming that the cosine value of the gradient direction angle with the three adjacent nodes is (0.45, 0.38, 0.51), the gradient direction consistency parameter 0.45 is obtained after averaging. Then, based on the intersection density, i.e., the ratio of the actual number of connected edges to the preset maximum number of edges, the birefringence abnormality, i.e., the absolute deviation between the node value and the normal tissue baseline value, and the gradient direction consistency parameter, the initial node weights were generated by linear weighting according to formula (2). The intersection density weight α was set to 0.4, the birefringence abnormality weight β was set to 0.5, and the direction consistency weight γ was set to 0.1. Combined with the intersection density of 0.8 and the birefringence abnormality of 0.3, the initial weight was generated as 0.4×0.8+0.5×0.3+0.1×(1-0.45)=0.61.

[0111] Next, we construct a node attribute feature matrix, which includes standardized spatial coordinates, birefringence values, and normalized gradient direction vectors, and then aggregate the features using the initial weights of formula (3). Based on this weight, we aggregate the node's three-dimensional coordinates (12.5, 7.2, 3.8), birefringence 1.0, and gradient direction vector (0.7, -0.1, 0.7), generating an attribute feature vector [0.32, 0.18, 0.24, 0.61]. Simultaneously, we calculate the inter-node geometric relationship strength matrix according to formula (4), and construct spatial constraint relationships using the inverse of the Euclidean distance and the initial weight. We calculate the Euclidean distance between this node and its adjacent nodes (2.1 μm, 1.8 μm, 2.3 μm), and construct a geometric relationship strength matrix [0.61 / 2.1, 0.61 / 1.8, 0.61 / 2.3]. Finally, the attribute features and geometric relationship features are input into the dual-channel graph attention network. The attribute channel captures birefringence anomalies, the geometric channel strengthens short-distance connections, and the fusion features identify the bifurcation point as the key node of crack extension. After extracting the feature correlations respectively through the self-attention mechanism of formula (5), the weighted splicing of formula (6) is performed with a fusion weight of λ=0.6 to generate high-order topological features that integrate biochemical characteristics and morphological distribution.

[0112] In one feasible embodiment, step S110: transmitting a light signal to the bone joint region using a polarization-tunable laser light source, receiving the reflected signal using polarization-sensitive optical coherence tomography, and calculating the birefringence distribution using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information, including:

[0113] By adjusting the wavelength and polarization direction parameters of the polarization-tunable laser light source, multi-angle polarized light signals are emitted to the surface and deep layers of the cartilage in the bone joint area. The multi-angle polarized light signals cover the collagen fiber arrangement direction of the surface and deep layers of the cartilage.

[0114] Multi-angle polarized light signals are generated by adjusting the polarization direction of the light source output to form a sequence of light beams with different polarization states. In bone and joint inspection scenarios, collagen fibers in the surface cartilage are arranged horizontally, while fibers in deeper layers gradually transition to a vertical arrangement. Multi-angle polarized light can cover fiber orientations at different depths. The polarization direction parameter refers to the direction of the electric field vibration of the light beam. This is continuously adjusted by rotating a wave plate or liquid crystal modulator to create a specific angle between the polarization direction of the incident light and the orientation of the fibers, thereby enhancing the difference in reflected signals from fibers with different orientations.

[0115] First, a polarization-tunable laser light source is configured, selecting the near-infrared band of 840nm to balance tissue penetration depth and resolution. A polarization controller generates three sets of orthogonal polarization directions of incident light at 0°, 45°, and 90°. Each set of polarized light scans in a fan-shaped pattern, covering the cartilage surface to the deep layer. During the scanning process, the polarization direction is dynamically adjusted according to the preset depth stratification strategy: surface scans use horizontal polarization light to match the arrangement of surface fibers, and deep scans switch to vertical polarization light to enhance the reflection signal of deep fibers. The light source output power is dynamically adjusted according to tissue transmittance to ensure balanced signal intensity at each depth layer.

[0116] The interference signals reflected by multi-angle polarized light signals at different depths in the bone joint area are received, and the local birefringence value of the cartilage is analyzed based on the phase delay and polarization state changes of the interference signals. The phase delay and polarization state changes are caused by the differences in the arrangement direction of the collagen fibers in the cartilage.

[0117] Phase retardation refers to the phase difference between the two orthogonal polarization components of polarized light propagating through tissue due to birefringence. This difference is directly related to the density and orientation of local collagen fibers. Polarization state change refers to the change in the polarization direction of the reflected light relative to the incident light. The Jones matrix can be used to quantitatively describe the polarization modulation effect caused by birefringence.

[0118] The interference signal is generated by the superposition of the reflected light from the reference arm and the sample arm. After being split by the spectrometer, the time-domain interference fringes are obtained. A polarization-resolution algorithm based on the Jones matrix is used to decompose the interference signal corresponding to each set of incident light polarization directions: first, the interference intensity of the orthogonal polarization channels is extracted, and the phase difference of each sampling point is calculated; then, the local birefringence value is solved through matrix inversion operation. The calculation process is shown in Equation (7).

[0119] Δn=λΔφ / (4πd)(7)

[0120] Where Δn is the local birefringence value, Δφ is the phase delay, d is the optical path difference, and λ is the wavelength of the light source.

[0121] For multi-angle polarization data, the maximum likelihood estimation method is used to fuse the measurement results of different polarization directions to eliminate the measurement deviation caused by anisotropy and finally generate a depth-resolved birefringence distribution curve.

[0122] The local birefringence value is mapped to the three-dimensional spatial coordinate system of the bone joint area to generate a birefringence gradient distribution map associated with the collagen fiber orientation to obtain a three-dimensional birefringence distribution map, in which the gradient direction is orthogonal to the collagen fiber arrangement direction.

[0123] The birefringence gradient distribution map is a vector field generated by calculating the rate of change of birefringence between adjacent voxels in three-dimensional space. The gradient direction is orthogonal to the orientation of the collagen fibers. This map can visually reveal areas of disrupted fiber alignment, such as cracks where the gradient direction suddenly changes due to fiber breakage.

[0124] The calculated birefringence values are mapped to a three-dimensional Cartesian coordinate system, and a three-dimensional gradient operation is performed on each voxel: the birefringence differences in the X, Y, and Z directions are calculated to synthesize the gradient vector. A Gaussian kernel function is used to smooth the gradient field to suppress noise. A pseudo-color-coded image is generated based on the modulus and direction of the gradient vector. The modulus indicates the order of fiber arrangement, with high modulus corresponding to highly ordered regions. The direction is represented in the HSV color space, with hue corresponding to azimuth and saturation corresponding to elevation. The final three-dimensional distribution map shows normal cartilage areas exhibiting a continuous gradient of color, while diseased areas appear as discrete color blocks or sharp color bands.

[0125] For example, taking the lesion area of the knee joint as an example, first, an 840nm superluminescent diode light source is used, and the polarization-maintaining fiber is coupled to a handheld PS-OCT probe, equipped with a high-speed galvanometer to achieve three-dimensional scanning, with a lateral resolution of 15μm and an axial resolution of 10μm. After data acquisition, the sample is layered and scanned, with each layer 50μm thick, and a total of 200 layers are collected. Each layer is irradiated in sequence with 0°, 45°, and 90° polarization directions, and 1024×1024 pixels of interference signal are collected in each direction. Then, the signal is analyzed, the Jones matrix elements are calculated for each pixel, and the birefringence value is extracted by singular value decomposition. After fusing the three layers of polarization data, the birefringence of the cartilage surface layer is 0.002±0.0003, the deep layer is 0.0015±0.0002, and the crack area has an abnormal value below 0.0008. Gradient calculations were performed, using the Sobel operator to calculate the three-dimensional gradient. A peak gradient modulus (>0.0015 / μm) was detected at the edge of the fissure, with the gradient direction differing by 90° from the surrounding normal area. Finally, the generated three-dimensional birefringence gradient distribution map showed that normal cartilage exhibited a blue gradient, indicating a consistent gradient direction, while the fissure area suddenly transformed into a red patch, indicating an orthogonal gradient direction. Osteophytes exhibited a circular gradient distribution. This example achieved high-precision three-dimensional characterization of collagen fiber orientation, providing reliable joint feature data for subsequent lesion annotation.

[0126] In one feasible embodiment, step S120: performing non-local mean filtering on the birefringence distribution map, extracting the edge contour of the target lesion area based on the gradient amplitude range, and generating binary mask data that retains microcrack texture features, includes:

[0127] According to the consistency of collagen fiber arrangement direction in adjacent areas of the birefringence distribution map, the local structural uniformity parameter of the birefringence distribution map is calculated. The local structural uniformity parameter is used to characterize the fiber continuity in the bone joint area.

[0128] The local structural uniformity parameter quantifies the directional consistency of collagen fiber alignment across adjacent regions in the birefringence distribution map. In bone and joint inspection scenarios, collagen fibers in normal cartilage exhibit directional continuity, while pathological areas (such as microcracks) can cause localized fiber disorganization. This parameter measures the degree of tissue structural continuity by measuring the similarity in the gradient directions of adjacent pixels. Higher values indicate more ordered fiber alignment and a lower risk of lesions.

[0129] First, a local neighborhood window (e.g., 5×5 pixels) is defined within the birefringence distribution map. The gradient direction vector is calculated for each pixel within the window. Next, cosine similarity is used to measure the consistency of the gradient directions between the central pixel and its neighbors. The mean cosine of the angles between all neighboring pixels and the central pixel is calculated as the uniformity parameter for the window. For three-dimensional data, the same operation is performed along the X, Y, and Z axes, ultimately generating a three-dimensional uniformity parameter matrix. The parameter ranges from -1 to 1, with values close to 1 indicating high uniformity and values close to -1 indicating severe disorder.

[0130] The gradient amplitude range is adjusted based on the local structural uniformity parameter. The gradient amplitude range is expanded in areas where the local structural uniformity parameter is lower than the set threshold, and the gradient amplitude range is narrowed in areas where the local structural uniformity parameter is higher than the set threshold.

[0131] The gradient amplitude range refers to the effective range of gradient intensity values at pixel points in an image, used to distinguish edges from background noise. Its dynamic adjustment strategy is based on the local structural uniformity parameter. In areas with uniform fiber arrangement, the threshold is reduced to reduce false detections due to noise, while in areas with disordered structure, the threshold is increased to capture the true edge of the lesion.

[0132] Regions are divided according to the uniformity parameter matrix, and a uniformity threshold is set, such as 0.7. Areas above the threshold are considered normal tissue, using a fixed narrow threshold, such as the top 10% of gradient amplitude. Areas below the threshold are considered potential lesions, using an adaptive wide threshold, such as the top 30%. Specifically, each pixel is determined to be an edge point based on the threshold range of its region. If the pixel gradient amplitude falls within the threshold range of the region, it is marked as a candidate edge point. This process requires traversing the entire 3D distribution map, adjusting the threshold layer by layer.

[0133] The birefringence distribution map is traversed according to the adjusted gradient amplitude range, and the pixel points within the gradient amplitude range and matching the gradient direction are marked as edge contours to generate binary mask data including the topological connection relationship of microcracks.

[0134] The topological connectivity of microcracks refers to the branching structure of the cracks reconstructed through the spatial connectivity of edge pixels. The binary mask data is a three-dimensional binary image that marks edge pixels, where foreground pixels (value 1) represent the edge of the lesion and background pixels (value 0) represent normal tissue.

[0135] After dynamic thresholding, candidate edge points are verified for directional consistency. The angle between the gradient direction of each edge point and its adjacent edge point is calculated. If the angle is less than a preset angle tolerance (e.g., 30°), the edge is considered to be a continuous edge of the same fissure branch. A 3D region growing algorithm is used to connect edge points that meet directional consistency, removing isolated noise points. In the resulting 3D mask data, each connected region represents a fissure or osteophyte boundary, and its topological structure is fully preserved through pixel adjacency.

[0136] For example, taking the detection of microcracks in the knee cartilage as an example, the three-dimensional birefringence distribution map generated by polarization-sensitive optical coherence tomography is first preprocessed: a local neighborhood window of 5×5×3 (X / Y / Z) is defined, and the gradient direction vector of the pixels in each window is calculated, such as using the Sobel operator and the cosine similarity. For example, the gradient direction vector of the central pixel is [0.8, 0.6, 0], and the cosine similarity is 0.98 when the neighborhood pixels are [0.7, 0.7, 0]. The directional consistency of adjacent pixels is counted to generate a three-dimensional uniformity parameter matrix. A uniformity threshold is set to 0.7. When the parameters of a certain region are lower than the threshold, such as the parameters around microcracks are 0.3-0.5, the upper limit of the gradient amplitude range in this region is expanded from the default 10% to 30%. After traversal adjustment, candidate edge points with gradient amplitudes between [15, 85] (normalized values) and directional angles less than 30° are selected. For example, the gradient directions of crack edge points are concentrated at 45°±10°. Adjacent points are then connected using a three-dimensional region growing algorithm. For example, crack branches must meet directional consistency for five consecutive pixels. Finally, binary mask data containing 12 independent crack branches is generated. The topological structure of the mask has a topological error of less than 5μm compared to the actual microscopically observed cartilage microcrack network. This embodiment achieves high-precision extraction of lesion edges and provides reliable input for subsequent topological network construction and feature enhancement.

[0137] In one feasible embodiment, step S130: converting pixels in the binary mask data into nodes, and constructing a topological network representing the fissure branches and osteophyte morphology through the connection relationship between the nodes, including:

[0138] Traverse the pixel points in the binary mask data and mark the pixel points that meet the preset gradient direction continuity as the initial node. The preset gradient direction continuity is determined by the angle between the gradient direction of the pixel point and the gradient direction of the adjacent pixel point and the preset angle threshold.

[0139] Preset gradient direction continuity refers to filtering pixels with continuous fiber alignment by setting a threshold for the angle between adjacent pixel gradient directions. In bone and joint lesions, sudden changes in gradient direction in microcracks or osteophytes can lead to increased differences in adjacent pixel orientations, while fiber alignment in normal areas exhibits consistent orientation.

[0140] Traverse each edge pixel in the binary mask data and calculate the angle between the gradient direction vector and the adjacent pixels in the neighborhood. The two-dimensional neighborhood is 8, and the three-dimensional neighborhood is 26. If the angle of all neighborhood pixels is less than the preset threshold, the pixel is marked as the initial node. Specifically, the vector dot product is used to calculate the cosine value, and the inverse cosine is used to obtain the angle. For example, when the gradient direction vector is 0.8, 0.6, 00.8, 0.6, 0 and 0.7, 0.7, 00.7, 0.7, 0, the cosine value is 0.98, corresponding to an angle of about 12°, which meets the continuity requirements. The final generated initial node set contains edge pixels that are continuous in direction and spatially coherent.

[0141] Based on the spatial coordinates, gradient direction and birefringence of the initial nodes, the spatial distribution density and birefringence difference of the initial nodes are calculated, and the initial nodes with birefringence difference greater than a preset difference threshold are screened out as target nodes.

[0142] Spatial distribution density measures the degree of clustering of initial nodes in three-dimensional space and is calculated using kernel density estimation or neighborhood counting. Birefringence variability refers to the absolute deviation between the birefringence value of a node and the mean of its neighborhood, reflecting the degree of local biochemical abnormality. When selecting target nodes, nodes with high birefringence variability should be retained to focus on the lesion area.

[0143] First, the spatial distribution density is calculated. Counting the number of nodes within a 50-μm radius centered on each initial node indicates greater clustering. The statistical radius can be adjusted based on the resolution. Next, the birefringence difference is calculated. For each node, the absolute difference between its birefringence value and the mean value within a 10 × 10 × 5 voxel neighborhood is calculated. Finally, a screening process is performed, setting a birefringence difference threshold (e.g., 0.001) to retain only nodes with a difference above the threshold. For example, the node difference in normal areas is 0.0002, the node difference at the edge of a fissure can reach 0.0015, and the difference in osteophyte areas is approximately 0.002.

[0144] With the target node as the center, connecting edges are generated according to the Euclidean distance of adjacent pixels in the binary mask data to construct a topological network representing the fissure branches and osteophyte morphology. The maximum connection distance of the connecting edges is determined by the difference in birefringence.

[0145] The maximum distance between connected edges is the maximum Euclidean distance allowed between nodes, dynamically adjusted based on the birefringence difference. Regions with high birefringence difference, such as crack tips, require a smaller distance to accurately model them, while regions with low birefringence difference can be relaxed to capture the overall morphology.

[0146] First, for each target node, the maximum connection distance D_max is calculated based on its birefringence difference (B). The formula is D_max = D_base × (1 + k × B), where D_base is the base distance (e.g., 20 μm), and k is the adjustment coefficient. Next, with the target node as the center, adjacent nodes within the D_max range are searched. If the angle between the two gradient directions is less than a threshold and the birefringence difference is significant (e.g., difference > 0.0005), a connecting edge is established. Finally, the network structure is optimized, using the Delaunay triangulation algorithm to supplement long-distance connections and ensure the continuity of fracture branches. Mesh densification is performed in osteophyte regions to increase node connection density.

[0147] For example, taking the detection of microcracks in the knee cartilage as an example, the initial node marking is first performed, the binary mask data is traversed, and the gradient direction angle threshold is set to 30°. The gradient direction of the pixel at the edge of a crack is 0.7, 0.7, 00.7, 0.7, 0, and its neighborhood pixel direction is 0.68, 0.72, 0.10.68, 0.72, 0.1, with an angle of 25°. It meets the conditions and is marked as an initial node. Afterwards, the target node is screened, and when the spatial distribution density is calculated, the node neighborhood density at a crack bifurcation is 8 nodes / 100μm. When the neighborhood mean is 0.001 and the node value is 0.0023, the birefringence difference is 0.0013, which exceeds the threshold of 0.001 and is retained. Finally, a topological network was constructed. The node birefringence difference was 0.0013. Setting D_base = 20 μm and k = 10, D_max = 20 × (1 + 10 × 0.0013) = 22.6 μm. Three adjacent nodes were found within the search range, and connecting edges were established to form the fracture branching topology. The final network contained 120 nodes and 350 edges, accurately reconstructing the fracture's three-dimensional morphology with an error of less than 8 μm compared to the microscopic imaging results.

[0148] Figure 3 FIG. 1 shows a flow chart of a method for generating interactive annotation results provided by an embodiment of the present application. Figure 3 As shown, it includes steps S310 to S330.

[0149] In one feasible embodiment, step S150: mapping the lesion topological features to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field, and generating an interactive annotation result that supports three-dimensional spatial rotation and multi-plane slice display by performing isosurface extraction and transparency mixed rendering based on the lesion probability distribution field, including:

[0150] S310: Establishing a probability mapping relationship between target node weights and spatial coordinates according to target node weights in the lesion topology feature and the spatial coordinates of the three-dimensional birefringence distribution map, and generating a three-dimensional probability distribution field including a lesion probability gradient.

[0151] Target node weights are high-order feature weights obtained by optimizing and adjusting initial weights in a dual-channel graph attention network. These weights incorporate contextual information about node attributes and geometric relationships, dynamically assigned through a self-attention mechanism, and are used to identify the spatial distribution weights of fissure bifurcations and osteophyte cores. The formation of lesion topological features directly depends on target node weights, which determine the node's contribution to feature aggregation. High-weighted nodes dominate the feature representation of fissure bifurcations and osteophyte cores, thereby affecting the accuracy of the topological network's modeling of lesion structures.

[0152] First, the node weights in the lesion topology features are bound to their corresponding three-dimensional spatial coordinates (X / Y / Z) to form weighted discrete point cloud data. Then, spatial interpolation is performed using the inverse distance weighted interpolation algorithm to calculate the probability value for each voxel in the three-dimensional space. The calculation structure is shown in Formula (8).

[0153] (8)

[0154] in, is the node weight, is the Euclidean distance from the voxel to the i-th node, and k is the attenuation coefficient.

[0155] Next, a gradient calculation is performed on the interpolated probability field, generating a probability gradient vector field. The gradient points to the region of fastest probability growth, typically corresponding to the direction of crack extension or the edge of osteophyte growth.

[0156] S320: Determine the isosurface boundary of the lesion probability distribution field based on the lesion probability gradient direction of the three-dimensional probability distribution field and the collagen fiber orientation information of the birefringence distribution map. The extraction range of the isosurface boundary is determined by the product of the lesion probability gradient and the birefringence difference.

[0157] The isosurface boundary is an equiprobable surface extracted from the three-dimensional probability distribution field and is used to define the spatial extent of the lesion area. The extraction range is dynamically adjusted by the product of the lesion probability gradient and the birefringence difference: the boundary range is expanded in areas with high gradient and high birefringence difference, while the range is narrowed in areas with low difference to accurately match the pathological characteristics.

[0158] First, dynamic threshold setting is performed to calculate the adjustment factor for each voxel ,in is the probability gradient modulus, and B is the birefringence difference. Then, the MarchingCubes algorithm is used to traverse the probability field and extract the values that meet the threshold conditions. The isosurface of the extracted image is obtained by smoothing the extracted image, such as Laplace smoothing, to remove isolated patches and retain the continuous surface.

[0159] S330: Aligning the isosurface boundary with the spatial coordinates of the three-dimensional birefringence distribution map, assigning transparency parameters according to the birefringence difference, performing transparency blending rendering on the area covered by the isosurface boundary, and generating an interactive annotation result that supports multi-plane slice display.

[0160] Transparency is a rendering attribute assigned based on the degree of birefringence difference, controlling the overlay display of annotation results with the original image. High-difference areas (lesion cores) have low transparency to highlight them, while low-difference areas (transition zones) have high transparency to blur their boundaries. Interactive annotation results support 3D rotation, multi-planar slices (MPR), and zooming, facilitating multi-angle clinical observation.

[0161] First, transparency mapping is used to build the birefringence difference B to transparency The linear mapping relationship: ,in, and is the ratio of the minimum to the maximum difference. Then the data is superimposed, the isosurface grid is aligned with the original OCT volume data space, and the alpha blending technique is used for rendering. ): (low transparency); transition area ( ): ; Normal area ( ): Finally, through graphics engines such as VTK and Three.js, the rotation, slicing, and transparency adjustment of 3D models are supported, resulting in interactive annotation results that support multi-plane slice display.

[0162] For example, the interactive annotation results include a multi-plane fusion display of a 3D rendered lesion model and the original OCT image. The annotation results are superimposed on the grayscale OCT volume data as a semi-transparent colored mesh (red for fissure tips and blue for osteophyte edges), supporting free 3D rotation, scaling, and arbitrary plane slicing.

[0163] For example, in the detection of microcracks in knee cartilage, the node weights in the lesion topology are first spatially bound to the spatial coordinates of the three-dimensional birefringence distribution map. An inverse distance weighted interpolation algorithm is then used to calculate a probability value for each voxel, with a weight of 0.85 for the fissure bifurcation nodes and 0.92 for the osteophyte core nodes. A three-dimensional field containing probability gradients is generated by setting the attenuation coefficient k = 2. A dynamic adjustment factor R = |P| × B is then calculated based on the orthogonal relationship between the probability gradient direction and collagen fiber orientation. When the probability gradient modulus |P| in the fissure tip region is 0.035 / μm and the birefringence difference B = 0.0023, the isosurface extraction threshold is automatically expanded to T × R = 0.7 × 0.0000805. Finally, the alpha blending technique was used to superimpose the isosurface mesh with the original OCT volume data. The core region of the crack with a birefringence difference of B>0.001 was assigned a transparency of α=0.2, the transition region α=0.5, and the normal region α=0.8. This generated an interactive annotation model that supports three-dimensional rotation and multi-plane slicing. This model is used to display the interactive annotation results.

[0164] Based on the same concept, the embodiment of the present application provides an automatic labeling system for bone and joint lesion areas based on image processing. Figure 4 The automatic labeling system for bone and joint lesion areas based on image processing provided in an embodiment of the present application is described in detail.

[0165] Figure 4 This is a structural block diagram of an automatic labeling system for bone and joint lesion areas based on image processing shown in an embodiment of the present application.

[0166] like Figure 4 As shown in FIG, the automatic labeling system for bone and joint lesion areas based on image processing is applicable to bone and joint lesion areas with osteophytes or cartilage cracks. The system includes:

[0167] Transmitter module 410 is configured to transmit a light signal to the bone joint region using a polarization-tunable laser light source, receive the reflected signal using polarization-sensitive optical coherence tomography, and calculate the birefringence distribution using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information;

[0168] Extraction module 420, configured to perform non-local mean filtering on the birefringence distribution map and extract the edge contour of the target lesion region based on a gradient amplitude range, thereby generating binary mask data that retains microcrack texture features. The gradient amplitude range is dynamically adjusted based on the local structural characteristics of the birefringence distribution map.

[0169] A construction module 430 is configured to convert pixels in the binary mask data into nodes and construct a topological network representing the fissure branching and osteophyte morphology based on the inter-node connectivity. Node attributes include spatial coordinates, gradient direction, and birefringence. The inter-node connectivity is determined by the Euclidean distance between adjacent pixels in the binary mask data.

[0170] The extraction module 420 is further configured to determine the node weight of the node based on the connection relationship between the nodes, the intersection density and the birefringence of the nodes, and perform multi-layer feature aggregation using a graph attention network to extract the lesion topology features of the target lesion area;

[0171] Generation module 440 is used to map the lesion topological features to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field, and based on the lesion probability distribution field, generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display through isosurface extraction and transparency mixed rendering.

[0172] In one embodiment, the extraction module 420 is specifically used to determine the node weight of the node based on the connection relationship between the nodes, according to the node's intersection density, birefringence and gradient direction consistency, and use a dual-channel graph attention network to aggregate the node attributes and geometric relationship features respectively, and extract the lesion topology features of the target lesion area after fusion. The intersection density is the ratio of the number of node adjacent edges to the preset maximum number of connection edges.

[0173] In one embodiment, the extraction module 420 is specifically used to calculate the gradient direction consistency parameter of the node based on the angle between the gradient direction of the node and the gradient direction of the adjacent node; based on the intersection density, birefringence and gradient direction consistency parameter of the node, the initial node weight of the node is generated by weighted summation; based on the initial node weight, the spatial coordinates, birefringence and gradient direction of the node are weightedly aggregated to generate node attribute features that characterize the biochemical characteristics of the lesion; based on the ratio of the Euclidean distance of the connection relationship between the nodes and the initial node weight, the geometric relationship strength of the node is constructed to generate geometric relationship features that characterize the morphological distribution of the lesion; the node attribute features and the geometric relationship features are input into the dual-channel graph attention network, and the node attribute features and the geometric relationship features are weightedly fused according to the initial node weight to generate lesion topology features including the fissure branch topology structure and the osteophyte morphology.

[0174] In one embodiment, the transmitting module 410 is specifically used to transmit multi-angle polarized light signals to the surface and deep layers of the cartilage in the bone joint area by adjusting the wavelength and polarization direction parameters of the polarization-adjustable laser light source, and the multi-angle polarized light signals cover the collagen fiber arrangement directions of the surface and deep layers of the cartilage; receive interference signals reflected by the multi-angle polarized light signals at different depths in the bone joint area, and analyze the local birefringence value of the cartilage according to the phase delay and polarization state changes of the interference signals, and the phase delay and polarization state changes are generated by the differences in the collagen fiber arrangement directions of the cartilage; map the local birefringence value to the three-dimensional spatial coordinate system of the bone joint area, generate a birefringence gradient distribution map associated with the collagen fiber orientation, and obtain a three-dimensional birefringence distribution map, in which the gradient direction is orthogonal to the collagen fiber arrangement direction.

[0175] In one embodiment, the extraction module 420 is specifically used to calculate the local structural uniformity parameter of the birefringence distribution map based on the consistency of the collagen fiber arrangement direction in adjacent areas in the birefringence distribution map, and the local structural uniformity parameter is used to characterize the fiber continuity in the bone joint area; adjust the gradient amplitude range based on the local structural uniformity parameter, expand the gradient amplitude range in areas where the local structural uniformity parameter is lower than the set threshold, and narrow the gradient amplitude range in areas where the local structural uniformity parameter is higher than the set threshold; traverse the birefringence distribution map according to the adjusted gradient amplitude range, mark the pixel points within the gradient amplitude range and matching the gradient direction as edge contours, and generate binary mask data including the topological connection relationship of microcracks.

[0176] In one embodiment, the construction module 430 is specifically used to traverse the pixel points in the binary mask data, mark the pixel points that meet the preset gradient direction continuity as initial nodes, and the preset gradient direction continuity is determined by the angle between the gradient direction of the pixel point and the gradient direction of the adjacent pixel point and the preset angle threshold; based on the spatial coordinates, gradient direction and birefringence of the initial node, calculate the spatial distribution density and birefringence difference of the initial node, and screen out the initial node whose birefringence difference is greater than the preset difference threshold as the target node; with the target node as the center, generate connecting edges according to the Euclidean distance of adjacent pixel points in the binary mask data, and construct a topological network characterizing the fissure branches and osteophyte morphology, and the maximum connection distance of the connecting edges is determined according to the birefringence difference.

[0177] In one embodiment, the generation module 440 is specifically used to establish a probability mapping relationship between the target node weight and the spatial coordinates according to the target node weight in the lesion topological feature and the spatial coordinates of the three-dimensional birefringence distribution map, and generate a three-dimensional probability distribution field including a lesion probability gradient; based on the lesion probability gradient direction of the three-dimensional probability distribution field and the collagen fiber orientation information of the birefringence distribution map, determine the isosurface boundary of the lesion probability distribution field, and the extraction range of the isosurface boundary is determined by the product of the lesion probability gradient and the birefringence difference; align the isosurface boundary with the spatial coordinates of the three-dimensional birefringence distribution map, assign a transparency parameter according to the birefringence difference, perform transparency mixed rendering on the area covered by the isosurface boundary, and generate an interactive annotation result that supports multi-plane slice display.

[0178] Figure 4 Each module in the system shown has the function of implementing Figures 1 to 3 The functions of each step in the embodiment can achieve the corresponding technical effects, which will not be described in detail here for the sake of brevity.

[0179] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in one embodiment of the present application is shown.

[0180] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0181] Specifically, the processor 510 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0182] The memory 520 may include a large-capacity memory for data or instructions. By way of example and not limitation, the memory 520 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 520 may include removable or non-removable (or fixed) media. Where appropriate, the memory 520 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 520 is a non-volatile solid-state memory.

[0183] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of the present disclosure.

[0184] The processor 510 reads and executes the computer program instructions stored in the memory 520 to implement any of the methods for automatically labeling bone and joint lesion areas based on image processing in the above embodiments.

[0185] In one example, the electronic device may further include a communication interface 530 and a bus 540. Figure 5 As shown, the processor 510 , the memory 520 , and the communication interface 530 are connected via a bus 540 and communicate with each other.

[0186] The communication interface 530 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0187] Bus 540 includes hardware, software, or both, and couples the components of the online data traffic metering device to each other. By way of example, and not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industrial Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Area Network (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 540 may include one or more buses. Although the embodiments of the present application describe and illustrate specific buses, the present application contemplates any suitable bus or interconnect.

[0188] The electronic device can execute the automatic labeling method of bone and joint lesion area based on image processing in the embodiment of the present application, thereby realizing the combination of Figures 1 to 3 The described method is an automatic labeling method for bone and joint lesion areas based on image processing.

[0189] In addition, in conjunction with the above-mentioned methods for automatically labeling bone and joint lesion areas based on image processing, embodiments of the present application may provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when executed by a processor, the computer program instructions implement any of the above-mentioned methods for automatically labeling bone and joint lesion areas based on image processing.

[0190] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0191] The functional blocks shown in the block diagrams described above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they may be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, and the like. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments may be stored in a machine-readable medium or transmitted via a data signal carried in a carrier wave over a transmission medium or communication link. "Machine-readable medium" may include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memory, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. Code segments may be downloaded via a computer network such as the Internet or an intranet.

[0192] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0193] Aspects of the present application have been described above with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each block in the flowcharts and / or block diagrams, as well as combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine such that execution of these instructions by the processor of the computer or other programmable data processing device enables the implementation of the functions / actions specified in one or more blocks in the flowcharts and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It should also be understood that each block in the block diagrams and / or flowcharts, as well as combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware that performs the specified functions or actions, or by a combination of dedicated hardware and computer instructions.

[0194] The above description is only a specific embodiment of the present application. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the scope of protection of the present application.

Claims

1. An automatic labeling method for bone and joint lesion areas based on image processing, suitable for bone and joint lesion areas with osteophytes or cartilage fissures, characterized by: include: A polarization-tunable laser light source is used to emit light signals into the bone joint area, and polarization-sensitive optical coherence tomography is used to receive the reflected signals. The birefringence distribution is calculated using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information. performing non-local mean filtering on the birefringence distribution map and extracting edge contours of the target lesion region based on a gradient amplitude range that is dynamically adjusted based on local structural characteristics of the birefringence distribution map to generate binary mask data that retains microcrack texture features; Converting pixels in the binary mask data into nodes, and constructing a topological network characterizing fissure branches and osteophyte morphology through inter-node connectivity, wherein the node attributes of the nodes include spatial coordinates, gradient directions, and birefringence, and the inter-node connectivity is determined by the Euclidean distance between adjacent pixels in the binary mask data; Based on the connection relationship between the nodes, the node weight of the node is determined according to the intersection density and birefringence of the node, and a graph attention network is used to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area; The lesion topological features are mapped to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field. Based on the lesion probability distribution field, isosurface extraction and transparency mixed rendering are performed to generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display.

2. The method according to claim 1, characterized in that The node weight of the node is determined based on the connection relationship between the nodes and the intersection density and birefringence of the nodes, and a graph attention network is used to perform multi-layer feature aggregation to extract the lesion topology features of the target lesion area, including: Based on the connection relationship between the nodes, the node weight of the node is determined according to the intersection density, birefringence and gradient direction consistency of the node, and a dual-channel graph attention network is used to aggregate the node attributes and geometric relationship features respectively. After fusion, the lesion topology features of the target lesion area are extracted. The intersection density is the ratio of the number of adjacent edges of the node to the preset maximum number of connection edges.

3. The method according to claim 2, characterized in that Based on the connection relationship between the nodes, the node weight of the node is determined according to the intersection density, birefringence and gradient direction consistency of the node, and a dual-channel graph attention network is used to respectively aggregate node attributes and geometric relationship features, and after fusion, the lesion topology features of the target lesion area are extracted, including: Calculating a gradient direction consistency parameter of the node based on the angle between the gradient direction of the node and the gradient direction of the adjacent node; Based on the intersection density, the birefringence and the gradient direction consistency parameter of the node, performing weighted summation to generate an initial node weight of the node; Performing weighted aggregation on the spatial coordinates, birefringence, and gradient direction of the nodes based on the initial node weights to generate node attribute features that characterize the biochemical characteristics of the lesion; According to the ratio of the Euclidean distance of the connection relationship between the nodes to the initial node weight, the geometric relationship strength of the nodes is constructed to generate a geometric relationship feature representing the morphological distribution of the lesion; The node attribute features and geometric relationship features are input into a dual-channel graph attention network, and the node attribute features and geometric relationship features are weightedly fused according to the initial node weights to generate lesion topology features including fissure branch topology structure and osteophyte morphology.

4. The method according to claim 1, wherein The method includes emitting a light signal to the bone joint region through a polarization-adjustable laser light source, receiving the reflected signal through polarization-sensitive optical coherence tomography, and calculating the birefringence distribution using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information, including: By adjusting the wavelength and polarization direction parameters of the polarization-adjustable laser light source, multi-angle polarized light signals are emitted to the surface layer and deep layer of the cartilage in the bone joint area, and the multi-angle polarized light signals cover the collagen fiber arrangement direction of the surface layer and deep layer of the cartilage; receiving interference signals reflected by the multi-angle polarized light signal at different depths in the bone joint region, and analyzing the local birefringence value of the cartilage based on phase delay and polarization state changes of the interference signals, wherein the phase delay and polarization state changes are generated by differences in the arrangement direction of collagen fibers in the cartilage; The local birefringence value is mapped to the three-dimensional spatial coordinate system of the bone joint area to generate a birefringence gradient distribution map associated with the collagen fiber orientation to obtain the three-dimensional birefringence distribution map, wherein the gradient direction in the birefringence gradient distribution map is orthogonal to the arrangement direction of the collagen fibers.

5. The method according to claim 4, characterized in that The non-local mean filtering is performed on the birefringence distribution map, and the edge contour of the target lesion area is extracted based on the gradient amplitude range to generate binary mask data that retains microcrack texture features, including: Calculating a local structural uniformity parameter of the birefringence distribution map based on the consistency of collagen fiber arrangement directions in adjacent regions in the birefringence distribution map, wherein the local structural uniformity parameter is used to characterize the fiber continuity of the bone joint region; adjusting the gradient amplitude range based on the local structural uniformity parameter, expanding the gradient amplitude range in an area where the local structural uniformity parameter is lower than a set threshold, and narrowing the gradient amplitude range in an area where the local structural uniformity parameter is higher than the set threshold; The birefringence distribution map is traversed according to the adjusted gradient amplitude range, pixel points within the gradient amplitude range and matching the gradient direction are marked as edge contours, and binary mask data including microcrack topological connection relationships are generated.

6. The method according to claim 1, characterized in that The step of converting the pixels in the binary mask data into nodes and constructing a topological network representing the fissure branches and osteophyte morphology through the connection relationship between the nodes includes: Traversing the pixel points in the binary mask data, marking the pixel points that meet the preset gradient direction continuity as initial nodes, wherein the preset gradient direction continuity is determined by the angle between the gradient direction of the pixel point and the gradient direction of the adjacent pixel point and a preset angle threshold; Calculating the spatial distribution density and birefringence difference of the initial nodes based on the spatial coordinates, gradient directions, and birefringence of the initial nodes, and selecting the initial nodes whose birefringence difference is greater than a preset difference threshold as target nodes; With the target node as the center, connecting edges are generated according to the Euclidean distance of adjacent pixels in the binary mask data to construct a topological network characterizing the fissure branches and osteophyte morphology, and the maximum connection distance of the connecting edges is determined according to the birefringence difference.

7. The method according to claim 1, characterized in that The mapping of the lesion topological features to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field, and generating an interactive annotation result supporting three-dimensional space rotation and multi-plane slice display by isosurface extraction and transparency mixed rendering based on the lesion probability distribution field, including: Establishing a probability mapping relationship between the target node weight and the spatial coordinates according to the target node weight in the lesion topological feature and the spatial coordinates of the three-dimensional birefringence distribution map, and generating a three-dimensional probability distribution field including a lesion probability gradient; Determining an isosurface boundary of the lesion probability distribution field based on the lesion probability gradient direction of the three-dimensional probability distribution field and the collagen fiber orientation information of the birefringence distribution map, wherein the extraction range of the isosurface boundary is determined by multiplying the lesion probability gradient by the birefringence difference; The isosurface boundary is aligned with the spatial coordinates of the three-dimensional birefringence distribution map, a transparency parameter is allocated according to the birefringence difference, and transparency mixed rendering is performed on the area covered by the isosurface boundary to generate an interactive annotation result that supports multi-plane slice display.

8. An automatic labeling system for bone and joint lesion areas based on image processing, suitable for bone and joint lesion areas with osteophytes or cartilage fissures, characterized by: The system comprises: The transmitting module is used to transmit light signals to the bone joint area through a polarization-tunable laser light source, receive the reflected signals through polarization-sensitive optical coherence tomography, and calculate the birefringence distribution using matrix analysis to obtain a three-dimensional birefringence distribution map including collagen fiber orientation information; an extraction module for performing non-local mean filtering on the birefringence distribution map and extracting the edge contour of the target lesion area based on a gradient amplitude range that is dynamically adjusted based on the local structural characteristics of the birefringence distribution map, thereby generating binary mask data that retains microcrack texture features; A construction module is used to convert the pixels in the binary mask data into nodes, and construct a topological network representing the fracture branching and osteophyte morphology through the connection relationship between the nodes, wherein the node attributes of the nodes include spatial coordinates, gradient direction and birefringence, and the connection relationship between the nodes is determined by the Euclidean distance between adjacent pixels in the binary mask data; The extraction module is further configured to determine the node weight of the node based on the connection relationship between the nodes and the intersection density and birefringence of the nodes, and to perform multi-layer feature aggregation using a graph attention network to extract the lesion topology features of the target lesion area; A generation module is used to map the lesion topological features to the spatial coordinate system of the three-dimensional birefringence distribution map to generate a lesion probability distribution field, and based on the lesion probability distribution field, generate interactive annotation results that support three-dimensional space rotation and multi-plane slice display through isosurface extraction and transparency mixed rendering.

9. An electronic device, characterized in that: The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for automatic labeling of bone and joint lesion areas based on image processing as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the method for automatically labeling bone and joint lesion areas based on image processing as described in any one of claims 1 to 7.

Images