A method and system for constructing trabecular bone features

Abstract

This invention relates to the field of medical imaging analysis and discloses a method and system for constructing trabecular bone features. The method includes: firstly extracting vulnerable regions from acquired vertebral body image sequences to generate an initial feature set characterizing fine-grained structural stability; performing a coherent comparison of the initial feature set across slice dimensions to obtain a composite feature vector reflecting the structural vulnerability level; based on the composite feature vector, using multi-layer progressive clustering to extract the connectivity change curves of trabecular bone under loading conditions, marking candidate trabecular bone units with potential micro-damage trends; classifying the candidate trabecular bone units according to risk gradients to reconstruct a local trabecular bone topology network with weighted hierarchies; and identifying high-risk nodes exhibiting imbalance tendencies within a specific mechanical simulation cycle based on the local topology network, generating a final feature construction path for microfracture risk prediction. This invention has the advantage of improving the sensitivity of microfracture risk prediction.

Description

Technical Field

[0001] This invention relates to the field of medical imaging analysis, specifically to a method and system for constructing trabecular bone features. Background Technology

[0002] In current medical imaging analysis, the acquisition of trabecular bone structural features mainly relies on bone mineral density (BMD) measurement or threshold-based 3D reconstruction methods. The former quantitatively assesses bone quality using DXA or CBCT, while the latter utilizes image segmentation and reconstruction algorithms to model the overall morphology of trabecular bone. However, these methods generally focus on macroscopic parameters or overall statistical characteristics, failing to adequately characterize the fine-grained structural features of trabecular bone in local areas. This leads to discrepancies between predicted results and actual conditions in certain clinical scenarios. For example, in the prediction of osteoporotic vertebral microfractures, current risk assessments primarily rely on BMD measured by DXA. However, numerous studies have shown that some patients have normal or only slightly decreased BMD values, but the trabecular bone in the vertebral body has already shown sparsity, directional disorder, or weakened connectivity, still resulting in occult microfractures under minor external forces. Because existing analytical methods often employ global statistics or mean-based processing, neglecting the fine-grained features of trabecular bone in areas of vertebral stress concentration, the predictive sensitivity is insufficient, making it difficult to promptly identify high-risk individuals. Therefore, it is necessary to design a method and system for constructing trabecular bone features to improve the sensitivity of microfracture risk prediction. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for constructing trabecular bone features, which has the advantage of improving the sensitivity of microfracture risk prediction and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving the sensitivity of microfracture risk prediction, this invention provides the following technical solution: a method for constructing trabecular bone features, comprising the following steps:

[0005] In the acquired vertebral body image sequence, vulnerable areas are extracted first, and the microscopic orientation, porosity and local linkage of the trabecular arrangement are captured by the directional consistency filter, thereby generating an initial feature set characterizing the stability of fine-grained structures.

[0006] The initial feature set is coherently compared across slice dimensions, and morphological evolution clues of historical fracture sites, branching and degeneration patterns of neighboring trabeculae, and local artifact perturbation probabilities are superimposed to obtain a composite feature vector that can reflect the level of structural vulnerability.

[0007] Based on composite feature vectors, multi-layer progressive clustering is used to extract the connectivity change curves of trabecular bone under loading conditions. By analyzing the degree of offset between the curves and the reference structure, candidate trabecular bone units with potential micro-damage trends are marked.

[0008] Candidate trabecular units are classified according to risk gradients, and a local topological network of trabecular units with weighted hierarchy is reconstructed by combining key support points, minor branch nodes and critical fracture nodes in high load-bearing areas.

[0009] Based on the local topology network, high-risk nodes that exhibit an imbalance tendency within a specific mechanical simulation cycle are identified. Combined with the dynamic fluctuations of trabecular connectivity under real-time imaging, a feature-based construction path is generated for the final prediction of microfracture risk.

[0010] Preferably, the process of generating an initial feature set characterizing the stability of fine-grained structures is as follows:

[0011] In the vertebral body imaging sequence, the target bone tissue and surrounding soft tissue are segmented based on the region growing algorithm to extract the target trabecular bone region with potential microfracture risk;

[0012] A directional consistency filter is applied to the target trabecular bone region to capture the main direction of trabecular bone arrangement, local fiber continuity, and pore spacing distribution.

[0013] Based on the captured direction and density parameters, the nodal connectivity and local interlocking rate of the trabeculae are further calculated to form a multidimensional structural parameter set;

[0014] By standardizing and normalizing the parameters, the multidimensional structural parameter set is mapped to an initial feature set characterizing the stability of fine-grained structures.

[0015] Preferably, the process of performing a coherent comparison of the initial feature set across slice dimensions is as follows:

[0016] In continuous image slices, the orientation of trabecular bone and the distribution of pores in the initial feature set are traced slice by slice to identify structural continuity and orientation shift.

[0017] The feature parameters between each slice are dynamically registered, and local artifact errors between slices are eliminated by morphological methods.

[0018] By combining morphological evolution clues of historical fracture sites with the branching and degeneration patterns of adjacent trabecular bone, a time-series control matrix was constructed.

[0019] Based on the time series comparison matrix, the feature consistency index across slices is calculated to form a comparison result that can characterize structural integrity and stability.

[0020] Preferably, the process of obtaining a composite feature vector that reflects the structural vulnerability level is as follows:

[0021] The feature consistency index obtained from cross-slice alignment is weighted and fused with the local perturbation probability to obtain an intermediate feature set containing multiple parameters;

[0022] Principal component analysis and feature selection algorithms were applied to the intermediate feature set to extract the core factors that contribute most to the stability of trabecular bone.

[0023] By combining the morphological feature evolution trajectory in the historical fracture patient database, core factors are matched and compared to establish risk markers and form a composite feature vector that can quantitatively reflect the vulnerability level of the trabecular bone structure.

[0024] Preferably, the process of extracting the connectivity change curves of trabeculae under loading environment using a multi-level progressive clustering method is as follows:

[0025] The composite feature vector is mapped to a multidimensional feature space, and a trabecular connectivity index is constructed based on different loading levels.

[0026] A hierarchical clustering method was used to progressively separate the connectivity patterns of trabeculae under different stress conditions;

[0027] The clustering results at each level are tracked over time to generate dynamic connectivity change curves of trabeculae under loading conditions.

[0028] Preferably, the process of identifying candidate trabecular bone units with a potential tendency for micro-damage is as follows:

[0029] Based on the difference between the connectivity change curve and the baseline structure curve, node regions with offset exceeding a set threshold are extracted;

[0030] A clustering anomaly detection algorithm is used to identify key points of local structural shifts, and the location of these shifts is determined by combining temporal features.

[0031] A set of candidate trabecular bone units is established for the identified offset regions, and the coordinates and connections of the candidate trabecular bone unit set in the overall trabecular bone network are marked to form candidate trabecular bone units with potential micro-damage tendencies.

[0032] Preferably, the process of classifying candidate trabecular bone units according to risk gradient is as follows:

[0033] The connectivity decay rate, porosity growth rate, and orientation shift magnitude of candidate trabecular bone units are quantified.

[0034] The quantitative indicators are mapped to the risk scoring function, and the risk levels are divided into low-risk, medium-risk, and high-risk gradients based on the set risk level ranges.

[0035] Based on the overall structural position of the candidate trabecular bone unit, higher weights are assigned to candidate units in high-load-bearing areas, resulting in a classification of candidate trabecular bone units according to risk gradients.

[0036] Preferably, the process of reconstructing a local topological network of trabecular bone with weighted hierarchy is as follows:

[0037] Candidate trabecular bone units under different risk gradients are used as nodes to establish a multi-attribute node set;

[0038] Based on the spatial adjacency, connectivity, and supporting role between nodes, construct a set of edges with directionality and weight;

[0039] A graph modeling method is used to form a topological network that reflects the local structural mechanical relationships of the trabecular bone.

[0040] By superimposing node risk weights with structural positions, a local topology network with weighted hierarchy is constructed.

[0041] The process of generating the final feature construction path for microfracture risk prediction is as follows:

[0042] In a local topology network, mechanical simulation methods are applied to simulate the response of trabecular bone under different loads;

[0043] Extract the displacement response, stress concentration, and connectivity fluctuation of nodes during the simulation period;

[0044] By combining the extracted mechanical response parameters with the risk gradient, high-risk nodes exhibiting a tendency to become unbalanced are identified, and these high-risk nodes are marked in the simulation results.

[0045] Based on the identified high-risk nodes, the propagation path and neighborhood dependency of the high-risk nodes in the local topology network are traced.

[0046] By combining the dynamic fluctuations of trabecular connectivity under real-time imaging, the risk evolution trajectory of high-risk nodes in the time dimension is formed;

[0047] Sequence modeling of risk evolution trajectory generates a complete feature construction path.

[0048] A trabecular bone feature construction system, comprising:

[0049] Vulnerability extraction module: Locates potentially vulnerable areas in vertebral body imaging sequences and extracts the microscopic arrangement and porosity features of trabecular bone;

[0050] The coherent comparison module integrates initial features across slice dimensions and combines historical fracture and artifact factors to generate a composite feature vector;

[0051] Clustering Curve Module: Extracts the connectivity change curves of trabeculae under loading conditions and identifies the offset trend through a progressive clustering method;

[0052] Risk classification module: performs risk gradient classification on candidate trabecular units and reconstructs the local topology network;

[0053] Dynamic prediction module: Detects imbalances in high-risk nodes during the simulation period and generates the final risk prediction path by combining real-time connectivity.

[0054] Compared with the prior art, the present invention provides a method and system for constructing trabecular bone features, which has the following beneficial effects:

[0055] This invention prioritizes the identification of vulnerable areas in vertebral imaging sequences and combines directional consistency filters to accurately extract the microscopic orientation, porosity, and local linkage of trabecular bone. This comprehensively captures the fine-grained structural information of trabecular bone, significantly improving the sensitivity to signs of local micro-injury. By comparing the coherence across slice dimensions and superimposing historical fracture area evolution clues, a composite feature vector is constructed to achieve a quantitative assessment of the vulnerability level of trabecular bone structure, compensating for the neglect of local anomalies by traditional bone density measurement or threshold segmentation methods. Using multi-layer progressive clustering and connectivity change analysis under loading conditions, the stability and potential damage trends of trabecular bone under different mechanical actions can be dynamically characterized, accurately marking candidate micro-injury units. Through risk gradient classification and weighted hierarchical local topology network reconstruction, the risk distribution patterns of trabecular bone in high-load areas and key support nodes can be revealed, clarifying the potential location of microfractures. Finally, by combining mechanical simulation and dynamic fluctuations in trabecular bone connectivity under real-time imaging to generate feature-constructed paths, a spatiotemporally continuous, structured, and quantifiable data foundation is provided for microfracture risk prediction. This invention can significantly improve the sensitivity and accuracy of microfracture risk prediction, providing a scientific basis for early clinical intervention and personalized osteoporosis prevention, while also having scalability and application potential to adapt to different image resolutions. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the method of the present invention;

[0057] Figure 2 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, a method for constructing trabecular bone features includes the following steps:

[0060] S1: In the acquired vertebral body image sequence, vulnerable areas are extracted first, and the microscopic orientation, porosity and local linkage of the trabecular arrangement are captured by the directional consistency filter, thereby generating an initial feature set characterizing the stability of fine-grained structures.

[0061] The process of generating the initial feature set characterizing the stability of fine-grained structures in S1 is as follows:

[0062] In vertebral body image sequences, the target bone tissue and surrounding soft tissue are segmented based on a region growing algorithm to extract target trabecular bone regions with potential microfracture risks. By setting grayscale thresholds and connectivity conditions, pixels in the vertebral body images that are close to the trabecular bone are used as initial seed points, and the region growing algorithm is used to expand layer by layer to automatically segment bone tissue regions with significant grayscale differences from the surrounding soft tissue. Artifacts and isolated points are removed through morphological operations (such as erosion and dilation) to finally obtain candidate trabecular bone regions containing potential microfracture risks.

[0063] An orientation consistency filter is applied to the target trabecular bone region to capture the principal orientation of the trabecular bone arrangement, local fiber continuity, and pore spacing distribution. An orientation consistency filter based on structural tensor is used to perform convolution operation on the pixel gradient field in the target region to calculate the principal orientation vector and orientation variance of each point. Based on this, the principal orientation distribution of trabecular bone fibers is extracted, and the orientation consistency in the neighborhood is statistically analyzed through a sliding window to obtain local fiber continuity. At the same time, the pore spacing is estimated based on the periodic distribution of pixel grayscale gaps.

[0064] Based on the captured orientation and density parameters, the node connectivity and local interlocking rate of the trabecular meshwork are further calculated to form a multidimensional structural parameter set. The trabecular meshwork structure is abstracted as a graph structure model of nodes and edges, where nodes represent intersections or branch points and edges represent fiber paths. The degree of each node in this graph model is counted to reflect the connectivity. At the same time, the local interlocking rate is obtained by calculating the ratio of the actual number of edges in the subgraph to the theoretical number of possible edges, thus forming a multidimensional structural parameter set including orientation, continuity, porosity, node connectivity, and interlocking rate.

[0065] Through parameter standardization and normalization, the multidimensional structural parameter set is mapped into an initial feature set characterizing the stability of fine-grained structures. Structural parameters of different dimensions (such as directional consistency, pore spacing, and node connectivity) are standardized (zero mean and variance normalization) and their numerical range is linearly normalized to the [0,1] interval to eliminate the dimensional differences between different parameters. Finally, all types of parameters are uniformly mapped to the feature space to form an initial feature set that can comprehensively characterize the stability of fine-grained trabecular structures.

[0066] S2: The initial feature set is coherently compared across slice dimensions, and morphological evolution clues of historical fracture sites, branching and degeneration patterns of neighboring trabeculae, and local artifact perturbation probabilities are superimposed to obtain a composite feature vector that can reflect the structural vulnerability level.

[0067] The process of performing a coherent comparison of the initial feature set across slice dimensions in S2 is as follows:

[0068] In continuous image slices, the orientation and pore distribution of trabecular bone in the initial feature set are tracked slice by slice to identify structural continuity and orientation shift. Continuous vertebral body image slices are input into a 3D reconstruction module to establish spatial index relationships between slices. Subsequently, principal component analysis or structural tensor methods are used to calculate the principal orientation of trabecular fibers in each slice, and the cosine similarity of the orientation vector fields in adjacent slices is compared to achieve orientation tracking slice by slice. Simultaneously, Fourier spectrum analysis is performed on the pore distribution among trabecular bone in each slice to calculate the center-to-center spacing and distribution density of pores. Combined with the difference changes between adjacent slices, the continuity, breakpoints, and orientation shift of the trabecular bone structure in continuous space are identified to further locate potential structurally weak areas.

[0069] The feature parameters between slices are dynamically registered, and local artifact errors between slices are eliminated through morphological methods. Key points of the trabecular bone (such as node intersections, branch origins, and boundary points) of each slice are extracted, and feature points-based affine transformation algorithms or elastic registration algorithms are used to map the feature points of adjacent slices to a unified reference coordinate system. A dynamic weight adjustment mechanism is introduced during the registration process to achieve higher fitting accuracy in high-variability regions and maintain low computational overhead in stable regions. After registration, morphological filtering (such as opening and closing operations and connected component analysis) is used to remove isolated pixels and unstructured edges caused by motion artifacts and scanning noise, thereby improving the stability and reliability of feature alignment between slices and ensuring the accuracy of subsequent comparison results.

[0070] By combining morphological evolution clues of historical fracture sites with branching and degeneration patterns of adjacent trabecular bone, a time-series comparison matrix was constructed. Morphological evolution trajectories of confirmed fracture areas at different time points were extracted from a database of past fracture cases, including typical patterns such as reduced trabecular bone number, sparse branching, and directional disorder. These evolutionary features were then encoded into prior pattern curves and stored in a feature library. Subsequently, cross-slice trabecular bone feature parameters (including branch number, branch length, thickness, and porosity) of the current target patient were input into the comparison system slice by slice for dynamic time-normalization comparison with the prior pattern, and the degree of deviation was calculated. Finally, these deviation parameters were stored in matrix form to construct a time-series comparison matrix, which reflects whether the target trabecular bone exhibits a degenerative trend similar to fracture evolution in both temporal and spatial dimensions.

[0071] Based on the time-series comparison matrix, a cross-slice feature consistency index is calculated to generate comparison results that characterize structural integrity and stability. A comprehensive evaluation function, including directional offset, porosity distribution difference, and branching degradation rate, is established based on the parameters in the time-series comparison matrix. These indicators are then weighted and summarized to obtain the cross-slice consistency index. The index ranges from 0 to 1, where a value close to 1 indicates high consistency of the trabecular structure in consecutive slices, suggesting overall structural stability; while a value close to 0 indicates significant differences in cross-slice features, suggesting a risk of local structural fracture or abnormal degradation. Finally, the system compares the consistency index with a clinical threshold. If the index is below the set value, a high-risk warning is output, serving as input for subsequent risk prediction and intervention recommendations.

[0072] The process of obtaining the composite feature vector that reflects the structural vulnerability level in S2 is as follows:

[0073] The feature consistency index obtained from cross-slice alignment is weighted and fused with local perturbation probabilities to obtain an intermediate feature set containing multiple parameters. The obtained cross-slice feature consistency index, reflecting the overall stability of trabeculae in continuous slices, is used as the main input parameter. Simultaneously, perturbation probabilities are extracted from local region images, including abnormal fluctuation probabilities caused by noise, artifacts, or trabeculae fracture points. Then, fusion weight coefficients are set, and the consistency index and perturbation probabilities are fused using linear weighting or multi-objective optimization methods. A dynamic adjustment mechanism is introduced during the weighting process, assigning higher weights to perturbation probabilities in high-risk areas and maintaining the dominance of the consistency index in low-risk areas. The final intermediate feature set not only contains parameters reflecting global stability but also incorporates factors characterizing local anomaly sensitivity.

[0074] Principal component analysis (PCA) and feature selection algorithms were applied to the intermediate feature set to extract the core factors that contribute most to trabecular bone stability. The intermediate feature set was then input into the PCA module, where covariance matrix decomposition was performed on the multidimensional parameters to obtain the principal component directions and their corresponding variance contribution rates. By selecting the top principal components with a cumulative contribution rate greater than 85%, key variables with high explanatory power were retained. Simultaneously, LASSO regression and recursive feature elimination algorithms were used to sparsify the parameters, removing features with weak impact on trabecular bone stability or those that were redundant. The final retained core factors typically include trabecular bone directional stability indices, porosity distribution variability, node connectivity, and local perturbation intensity. These factors have been clinically proven to be highly correlated with fracture risk and can represent the core stability characteristics of trabecular bone structure.

[0075] By combining the morphological evolution trajectories of historical fracture patients in a database, core factors are matched and compared to establish risk markers, forming a composite feature vector that quantitatively reflects the vulnerability level of trabecular bone structure. The historical fracture patient database stores a large amount of trabecular bone structure evolution information for confirmed cases, including dynamic change curves such as directional shifts, increased porosity, and decreased connectivity at different disease stages. The extracted core factors are matched and compared one by one with the evolution trajectories in the database, and a similarity score is calculated using the cosine similarity method. When a core factor highly matches the historical evolution pattern, it is labeled with a corresponding risk tag, such as "high vulnerability," "medium vulnerability," or "low vulnerability." Based on this, multiple risk-labeled core factors are recombined and encoded to form a multidimensional vector. This vector not only retains the quantitative characteristics of the parameters but also integrates clinical risk information, thus constituting a composite feature vector that quantitatively characterizes the vulnerability level of trabecular bone, providing input for subsequent microfracture risk prediction.

[0076] S3: Based on composite feature vectors, multi-layer progressive clustering is used to extract the connectivity change curves of trabecular bone under loading conditions. By analyzing the degree of offset between the curves and the reference structure, candidate trabecular bone units with potential micro-damage trends are marked.

[0077] The process of extracting the connectivity change curves of trabeculae under loading environment using multi-layer progressive clustering in S3 is as follows:

[0078] Composite feature vectors are mapped onto a multidimensional feature space, and trabecular connectivity indices are constructed based on different loading levels. The composite feature vectors obtained in the previous step are input into the feature mapping module, and projected onto the multidimensional feature space through kernel function mapping or dimensionality reduction methods (such as t-SNE or PCA) to more clearly reveal the nonlinear relationships between different parameters. Subsequently, multiple loading levels are set in this space, including four stages: static load, light load, medium load, and high load. The stress field distribution of each stage is generated through finite element mechanical simulation or experimental data calibration. Based on the stress field and trabecular geometric parameters, connectivity indices such as average nodal density, number of trabecular support paths, and stability of critical connection points are calculated. These connectivity indices exhibit different variation patterns under different loading levels.

[0079] A hierarchical clustering method was employed to progressively separate the connectivity patterns of trabecular bone under different stress conditions. In a multidimensional feature space, using the loading level as a reference, the similarity measure between connectivity indices was calculated using the Euclidean distance method. Then, a bottom-up hierarchical clustering algorithm (such as agglomerative clustering) was used to gradually group highly similar connectivity patterns into clusters. During each layer of clustering, a threshold criterion was set to ensure that highly similar patterns were merged preferentially, while low-similarity patterns were retained at higher levels. The results of this progressive clustering can reveal the evolution of trabecular bone structure from steady state to instability under different stress conditions, distinguishing multiple levels of connectivity patterns, including "stable connection patterns," "potential weakening patterns," and "precursor fracture patterns."

[0080] The clustering results at each level are tracked over time to form a dynamic connectivity change curve of trabecular bone under loading conditions. The corresponding cluster center point is extracted as a representative connectivity state at each loading stage, and its position in the time series is recorded. The cluster center points at each stage are connected in the loading order to form a dynamic curve, which can intuitively reflect the changing trend of connectivity index of trabecular bone under gradually increasing stress, such as gradually decreasing from high stability to the critical fracture point. The critical turning point of structural vulnerability is identified by the curve slope and inflection point detection algorithm, thus providing a basis for early warning of microfracture risk.

[0081] The process of identifying candidate trabecular bone units with potential micro-damage tendencies in S3 is as follows:

[0082] Based on the difference between the connectivity change curve and the baseline structure curve, node regions with offset exceeding a set threshold are extracted. From the obtained dynamic connectivity change curve, connectivity indicators of each trabecular node at different loading stages are extracted, such as node degree, number of paths, and stability of key connection points. These curves are compared node by node with the established baseline structure curve, which is generated by the average connectivity evolution of healthy vertebrae or low-risk samples. By calculating the degree of node offset, including absolute difference, percentage change, or standardized residual, node regions with offset exceeding a pre-set threshold (such as twice the standard deviation of the average offset) are identified. The selection of the threshold takes into account image noise, measurement error, and clinically acceptable micro-damage sensitivity to ensure that the identified node regions truly reflect potential structural weakening.

[0083] A clustering anomaly detection algorithm is used to identify key points of local structural displacement, and the location of their occurrence is located by combining temporal features. Based on the selected displacement nodes, anomaly detection algorithms are used to perform cluster analysis on the displacement patterns of local nodes. By analyzing the spatial proximity of nodes and the degree of anomaly in the changes of connectivity indicators, the key points with the most significant displacement are identified. Furthermore, by combining time series features, such as the loading stage of node displacement, continuous displacement trend and peak point, the location and stage where micro-damage may occur are accurately located. This process not only considers spatial concentration, but also uses temporal information to distinguish between temporary fluctuations and real micro-damage trends, thereby enhancing the accuracy of candidate node identification.

[0084] A candidate trabecular bone unit set is established for the identified offset regions, and the coordinates and connectivity of the candidate trabecular bone unit set in the overall trabecular bone network are marked to form candidate trabecular bone units with potential micro-damage trends. The identified key nodes and their neighboring nodes are spatially expanded to form a complete candidate trabecular bone unit set. Each unit contains node location, connectivity attributes, and connection path information. Using the overall trabecular bone network topology, the candidate units are mapped in coordinates in the three-dimensional skeleton, and the connectivity between nodes is recorded, including the number of directly connected trabecular bone lines, support paths, and key bridging points. Each candidate unit is uniquely identified in the network, and combined with its offset degree, key weight, and neighborhood influence, a systematic and traceable candidate trabecular bone unit is formed.

[0085] S4: Classify candidate trabecular units according to risk gradient, and reconstruct a local trabecular topology network with weighted hierarchy by combining key support points, minor branch nodes and critical fracture nodes in high load-bearing areas.

[0086] The process of classifying candidate trabecular bone units according to risk gradient in S4 is as follows:

[0087] The connectivity decay rate, porosity growth rate, and orientation offset of candidate trabecular meshwork units are quantified. Node and connectivity information for each unit is extracted from the generated candidate trabecular meshwork unit set, including nodal degree, number of support paths, and stability of key bridging points. The connectivity decay rate of nodes is calculated under different loading stages. The structural weakening trend of the units is quantified by differential analysis of the rate of change of connectivity indices with time or stress level. The porosity growth rate of the units is calculated using the porosity distribution obtained from image segmentation, quantifying the expansion of micropores and the increase in the proportion of voids. The offset of the trabecular meshwork's main orientation is also measured, represented by the directional consistency angle difference or vector offset. These three key indices are normalized to form multidimensional quantitative parameters for candidate units.

[0088] Quantitative indicators are mapped to a risk scoring function, and risk levels are divided into low, medium, and high risk gradients based on a set risk level range. A multi-parameter risk scoring function is designed to sum the connectivity decay rate, porosity growth rate, and orientation shift magnitude according to preset weights or to map them to a unified risk score range using multi-objective optimization. For example, high decay rate, high porosity growth, and large orientation shift will collectively lead to a higher risk score. Based on statistical analysis and clinical experience, a risk score range threshold is set, and candidate units whose scores fall within different ranges are marked as low, medium, and high risk, respectively. During the risk classification process, confidence intervals or fuzzy logic processing can be introduced to address image noise and measurement errors, thereby improving the reliability and robustness of risk level classification.

[0089] By combining the overall structural location of the candidate unit within the trabecular bone, higher weights are assigned to candidate units in high-load-bearing areas, resulting in a classification of candidate trabecular bone units according to risk gradients. The spatial coordinates and topological location of each candidate unit are located within the overall trabecular bone network, identifying its high-load-bearing areas, such as vertebral load-bearing surfaces, stress concentration points, or critical support zones. For candidate units located in high-load-bearing areas, weighting factors are added to the original risk score to appropriately increase the risk level, reflecting its potentially greater impact in structural instability. Through the weighted and corrected scores, all candidate units are ultimately classified into low, medium, and high-risk gradients, generating a structured output table or network diagram. This approach retains local micro-damage indicators while also reflecting the mechanical importance of the overall structure, thus providing clear and traceable classification results for microfracture risk prediction.

[0090] The process of reconstructing a local trabecular topology network with weighted hierarchy in S4 is as follows:

[0091] Candidate trabecular units under different risk gradients are used as nodes to establish a multi-attribute node set. The spatial coordinates, connectivity parameters, node degree and micro-damage index of each unit are extracted from the candidate trabecular units divided by risk gradient obtained in the previous step. This information is encoded as node attributes. Each node not only contains physical location, but also records its risk level, porosity, connectivity decay rate and orientation offset, etc. In this way, a multi-attribute node set is formed, so that each node can reflect its microstructure state and potential mechanical influence in the topology network.

[0092] Based on the spatial adjacency, connectivity, and supporting role between nodes, a set of edges with directionality and weight is constructed. The spatial proximity of candidate units in the overall trabecular network is analyzed, including the Euclidean distance between nodes, connecting paths, and cross-support relationships. According to the mechanical connection characteristics of the trabecular, directed or undirected edges are established between nodes. The direction of the edge reflects the main stress transmission direction, and the weight of the edge is assigned according to the strength of node connectivity, the importance of local support, and the risk level. For example, connecting edges located on high-bearing paths and with high risk levels are given larger weights, indicating their key role in the stability of the local structure. In this way, a set of edges containing directionality and weight is generated.

[0093] A graph modeling approach is employed to form a topological network reflecting the local structural mechanical relationships of trabeculae. Multi-attribute node sets and weighted edge sets are input into a graph modeling framework, such as an adjacency matrix or graph database, to form a complete local trabeculae topological network. Each node in the network and its connected edges collectively describe the spatial distribution and mechanical transmission relationships of the local trabeculae. By introducing graph indices, such as node centrality, edge weight distribution, and local clustering coefficients, the structural stability, critical support paths, and potentially vulnerable areas of the network are analyzed, ensuring that the topological network accurately reflects the microscopic mechanical properties and potential damage risks of the trabeculae.

[0094] By superimposing node risk weights with structural locations, a local topological network with weight hierarchy is constructed. Based on the generated topological network, the risk level of each node and its spatial importance in the overall trabecular structure are superimposed and calculated to form a comprehensive weight index. Nodes with high risk and located in critical support areas are given higher weights in the network, while low-risk or non-critical nodes have relatively lower weights. By normalizing and classifying the node weights hierarchically, the weight hierarchy of the network is realized, so that the network not only reflects structural connectivity but also local mechanical importance and the distribution of potential micro-damage risks.

[0095] S5: Based on the local topology network, identify high-risk nodes that show a tendency to become unbalanced within a specific mechanical simulation cycle, and combine the dynamic fluctuations of trabecular connectivity under real-time imaging to generate the final feature construction path for microfracture risk prediction.

[0096] The process of generating the final feature construction path for microfracture risk prediction in S5 is as follows:

[0097] In a local topology network, mechanical simulation methods are applied to simulate the response of the trabecular mesh under different loads. The constructed local topology network with weighted hierarchy is input into a finite element analysis or multibody dynamics simulation platform, and different types of external load conditions are set, including static load, periodic load and sudden impact load. By parametrically modeling the material properties of each node and its edges, the connection strength between nodes and the weights, the deformation, stress distribution and strain response of the trabecular mesh under different mechanical environments are simulated, thereby obtaining the mechanical behavior of each trabecular mesh element under real load conditions.

[0098] The displacement response, stress concentration, and connectivity fluctuation of nodes are extracted during the simulation period. During the simulation, data on the displacement vector, stress tensor, and local connectivity of each node are collected over time. The displacement response reflects the deformation amplitude of the node under load, the stress concentration indicates the risk area where the node and its adjacent elements may suffer micro-damage, and the connectivity fluctuation reveals the potential trend of local fracture or bridging loss of the trabecular meshwork. By performing time-series sampling and quantification on these indicators, the dynamic response characteristics of the nodes throughout the entire simulation period are formed into an analyzable dataset.

[0099] By combining the extracted mechanical response parameters with the risk gradient, high-risk nodes exhibiting a tendency to become unbalanced are identified and marked in the simulation results. The displacement, stress, and connectivity fluctuation parameters of the nodes are comprehensively evaluated with the previously calculated risk levels. Nodes exhibiting abnormal shifts, stress concentrations, or significant connectivity loss trends during the simulation are identified using weighted scoring or multidimensional discriminant functions. These nodes are marked as high-risk nodes, and their spatial location, weight level, and association with neighboring nodes in the local topology network are recorded to highlight them in the simulation visualization results, providing clear high-risk areas for microfracture prediction.

[0100] Based on the identified high-risk nodes, the propagation paths and neighborhood dependencies of these nodes in the local topology network are traced. Starting from the high-risk nodes, their connections with adjacent nodes are traced along the topology network to analyze the possible propagation paths of mechanical stress or deformation along the network. The strength of the dependencies between nodes is recorded, including the stability contribution of key support nodes to the high-risk nodes and the propagation effect of bridging nodes. Through this path tracing, a complete mapping of the potential influence range of high-risk nodes in the network is formed, revealing the possible propagation paths of micro-damage and the mechanical coupling characteristics with surrounding trabecular units.

[0101] By combining the dynamic fluctuations of trabecular connectivity under real-time imaging, a risk evolution trajectory of high-risk nodes in the time dimension is formed. The response data of high-risk nodes obtained by simulation are compared and fused with the actual changes of trabecular connectivity over time in continuous image sequences. Through time alignment and interpolation methods, the structural connectivity, stress state and displacement response of high-risk nodes at different time points are mapped onto the time axis to form the risk evolution curve of the node. The curve not only reflects the risk change trend of a single node, but also shows the possible chain effects it may have in the local network, providing time dimension data support for dynamic monitoring and prediction of microfractures.

[0102] Sequence modeling of risk evolution trajectories generates complete feature construction paths. Time series data of each high-risk node are input into sequence modeling algorithms, such as long short-term memory networks, hidden Markov models, or autoregressive models, to model and predict risk evolution trajectories. By analyzing the dependencies, response patterns, and temporal correlations between nodes, complete feature construction paths are generated, comprehensively reflecting the spatial and temporal distribution of micro-injury risk in trabecular bone. This provides systematic and quantifiable analytical results for microfracture risk prediction and provides a basis for clinical intervention decisions.

[0103] Example 2: As Figure 2 As shown, a trabecular bone feature construction system includes:

[0104] Vulnerability extraction module: Locates potentially vulnerable areas in vertebral body imaging sequences and extracts the microscopic arrangement and porosity features of trabecular bone;

[0105] The coherent comparison module integrates initial features across slice dimensions and combines historical fracture and artifact factors to generate a composite feature vector;

[0106] Clustering Curve Module: Extracts the connectivity change curves of trabeculae under loading conditions and identifies the offset trend through a progressive clustering method;

[0107] Risk classification module: performs risk gradient classification on candidate trabecular units and reconstructs the local topology network;

[0108] Dynamic prediction module: Detects imbalances in high-risk nodes during the simulation period and generates the final risk prediction path by combining real-time connectivity.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing trabecular bone features, characterized in that, Includes the following steps: In the acquired vertebral body image sequence, vulnerable areas are extracted first, and the microscopic orientation, porosity and local linkage of the trabecular arrangement are captured by the directional consistency filter, thereby generating an initial feature set characterizing the stability of fine-grained structures. The initial feature set is coherently compared across slice dimensions, and morphological evolution clues of historical fracture sites, branching and degeneration patterns of neighboring trabeculae, and local artifact perturbation probabilities are superimposed to obtain a composite feature vector that can reflect the level of structural vulnerability. Based on composite feature vectors, multi-layer progressive clustering is used to extract the connectivity change curves of trabecular bone under loading conditions. By analyzing the degree of offset between the curves and the reference structure, candidate trabecular bone units with potential micro-damage trends are marked. Candidate trabecular units are classified according to risk gradients, and a local topological network of trabecular units with weighted hierarchy is reconstructed by combining key support points, minor branch nodes and critical fracture nodes in high load-bearing areas. Based on the local topology network, high-risk nodes that exhibit an imbalance tendency within a specific mechanical simulation cycle are identified. Combined with the dynamic fluctuations of trabecular connectivity under real-time imaging, a feature-based construction path is generated for the final prediction of microfracture risk.

2. The method for constructing trabecular bone features according to claim 1, characterized in that, The process of generating an initial feature set characterizing the stability of fine-grained structures is as follows: In the vertebral body imaging sequence, the target bone tissue and surrounding soft tissue are segmented based on the region growing algorithm to extract the target trabecular bone region with potential microfracture risk; A directional consistency filter is applied to the target trabecular bone region to capture the main direction of trabecular bone arrangement, local fiber continuity, and pore spacing distribution. Based on the captured direction and density parameters, the nodal connectivity and local interlocking rate of the trabeculae are further calculated to form a multidimensional structural parameter set; By standardizing and normalizing the parameters, the multidimensional structural parameter set is mapped to an initial feature set characterizing the stability of fine-grained structures.

3. The method for constructing trabecular bone features according to claim 2, characterized in that, The process of performing a coherent comparison of the initial feature set across slice dimensions is as follows: In continuous image slices, the orientation of trabecular bone and the distribution of pores in the initial feature set are traced slice by slice to identify structural continuity and orientation shift. The feature parameters between each slice are dynamically registered, and local artifact errors between slices are eliminated by morphological methods. By combining morphological evolution clues of historical fracture sites with the branching and degeneration patterns of adjacent trabecular bone, a time-series control matrix was constructed. Based on the time series comparison matrix, the feature consistency index across slices is calculated to form a comparison result that can characterize structural integrity and stability.

4. The method for constructing trabecular bone features according to claim 3, characterized in that, The process of obtaining a composite eigenvector that reflects the structural vulnerability level is as follows: The feature consistency index obtained from cross-slice alignment is weighted and fused with the local perturbation probability to obtain an intermediate feature set containing multiple parameters; Principal component analysis and feature selection algorithms were applied to the intermediate feature set to extract the core factors that contribute most to the stability of trabecular bone. By combining the morphological feature evolution trajectory in the historical fracture patient database, core factors are matched and compared to establish risk markers and form a composite feature vector that can quantitatively reflect the vulnerability level of the trabecular bone structure.

5. The method for constructing trabecular bone features according to claim 4, characterized in that, The process of extracting the connectivity change curves of trabeculae under loading conditions using a multi-level progressive clustering approach is as follows: The composite feature vector is mapped to a multidimensional feature space, and a trabecular connectivity index is constructed based on different loading levels. A hierarchical clustering method was used to progressively separate the connectivity patterns of trabeculae under different stress conditions; The clustering results at each level are tracked over time to generate dynamic connectivity change curves of trabeculae under loading conditions.

6. The method for constructing trabecular bone features according to claim 5, characterized in that, The process of identifying candidate trabecular bone units with a potential tendency for micro-damage is as follows: Based on the difference between the connectivity change curve and the baseline structure curve, node regions with offset exceeding a set threshold are extracted; A clustering anomaly detection algorithm is used to identify key points of local structural shifts, and the location of these shifts is determined by combining temporal features. A set of candidate trabecular bone units is established for the identified offset regions, and the coordinates and connections of the candidate trabecular bone unit set in the overall trabecular bone network are marked to form candidate trabecular bone units with potential micro-damage tendencies.

7. The method for constructing trabecular bone features according to claim 6, characterized in that, The process of classifying candidate trabecular bone units according to risk gradient is as follows: The connectivity decay rate, porosity growth rate, and orientation shift magnitude of candidate trabecular bone units are quantified. The quantitative indicators are mapped to the risk scoring function, and the risk levels are divided into low-risk, medium-risk, and high-risk gradients based on the set risk level ranges. Based on the overall structural position of the candidate trabecular bone unit, higher weights are assigned to candidate units in high-load-bearing areas, resulting in a classification of candidate trabecular bone units according to risk gradients.

8. The method for constructing trabecular bone features according to claim 7, characterized in that, The process of reconstructing a local topological network of trabecular bone with weighted hierarchies is as follows: Candidate trabecular bone units under different risk gradients are used as nodes to establish a multi-attribute node set; Based on the spatial adjacency, connectivity, and supporting role between nodes, construct a set of edges with directionality and weight; A graph modeling method is used to form a topological network that reflects the local structural mechanical relationships of the trabecular bone. By superimposing node risk weights with structural positions, a local topology network with weighted hierarchy is constructed.

9. A method for constructing trabecular bone features according to claim 8, characterized in that, The process of generating the final feature construction path for microfracture risk prediction is as follows: In a local topology network, mechanical simulation methods are applied to simulate the response of trabecular bone under different loads; Extract the displacement response, stress concentration, and connectivity fluctuation of nodes during the simulation period; By combining the extracted mechanical response parameters with the risk gradient, high-risk nodes exhibiting a tendency to become unbalanced are identified, and these high-risk nodes are marked in the simulation results. Based on the identified high-risk nodes, the propagation path and neighborhood dependency of the high-risk nodes in the local topology network are traced. By combining the dynamic fluctuations of trabecular connectivity under real-time imaging, the risk evolution trajectory of high-risk nodes in the time dimension is formed; Sequence modeling of risk evolution trajectory generates a complete feature construction path.

10. A trabecular bone feature construction system, applied to the method as described in any one of claims 1-9, characterized in that, include: Vulnerability extraction module: Locates potentially vulnerable areas in vertebral body imaging sequences and extracts the microscopic arrangement and porosity features of trabecular bone; The coherent comparison module integrates initial features across slice dimensions and combines historical fracture and artifact factors to generate a composite feature vector; Clustering Curve Module: Extracts the connectivity change curves of trabeculae under loading conditions and identifies the offset trend through a progressive clustering method; Risk classification module: performs risk gradient classification on candidate trabecular units and reconstructs the local topology network; Dynamic prediction module: Detects imbalances in high-risk nodes during the simulation period and generates the final risk prediction path by combining real-time connectivity.

Images