Bone tumor degradation area resection range decision-making auxiliary system

Abstract

The invention relates to a decision-making auxiliary system for a resection range of a bone tumor degradation region. The decision-making auxiliary system is used for modeling biological characteristics of metabolic activity, blood perfusion and tissue permeability reflected in different image modalities including PET / MRI (Positron Emission Tomography / Magnetic Resonance Imaging); using a space coupling modeling method to generate a functional activity gradient map in the degradation area, and mapping three layers of areas including a central focus, a diffusion boundary and a latent erosion zone; constructing a dynamic interval label mechanism, and forming a variable removal range space; dividing a gradient map generated based on an image into a plurality of different areas, performing joint modeling on each area and corresponding functional positioning parameters and pathological risk parameters, layering tissue structures of cortical bones, cancellous bones, marrow cavities and soft tissue boundaries, and generating a local resection profile cluster according to an anatomical axis; each section cluster is hooked with a single function / risk factor weight, and the whole excision body is re-simulated by adjusting excision target parameters; the excision boundary has the form adaptive capacity in the space, and personalized operation strategy customization and dynamic adjustment are achieved.

Description

Technical Field

[0001] The present invention relates to a bone tumor decision support system, in particular to a bone tumor deterioration area resection range decision support system. Background Art

[0002] Although existing methods, devices, electronic devices and storage media for processing tumor image data such as Chinese patent CN115409827A provide a relatively systematic computational framework for predicting and guiding the range of tumor resection, and propose a strategy for obtaining tumor boundary information through tumor image data sets and generating corresponding tumor resection models, which can achieve visual assistance for the direction and position of tumor resection, thereby enhancing the objectivity and accuracy of the resection operation to a certain extent, such methods still have significant limitations and technical deficiencies when used for complex bone tumors, especially in the direction of resection range decision-making in degraded areas, and cannot fully meet the clinical needs of bone tumor surgery for the dynamic balance between functional protection and lesion clearance, which is specifically reflected in the following key aspects: First, the existing technology mainly relies on the extraction of tumor boundary information from static images. Although the boundary information is relatively clear in soft tissue tumors, in bone tumors, especially degraded osteosarcomas with strong erosion and blurred boundaries, the boundaries obtained solely based on image grayscale or segmentation algorithms are often difficult to accurately describe the actual pathological spread range, fail to consider the gradual process of latent tumor spread from the center to the outside, and lack the activity level division under the functional state, which makes the resection model It is easy to misjudge potential lesions as normal tissues, or vice versa, cause excessive resection of unnecessary areas, thereby reducing the proportion of postoperative functional retention. Secondly, the tumor resection model generated by this technical solution is expressed in the form of multiple resection sections. Although it provides spatial hints on direction and position, the section itself is a regular structure and does not have the ability of adaptive reconstruction. It cannot automatically adjust the boundary shape according to the functional value of different tissue levels or the risk level of lesions. Sectional modeling can easily cause rigidity, discontinuity, and non-fitting of the resection plan in complex bone structures, especially when the bone tissue has strong anatomical heterogeneity (such as The junction of cortical bone and cancellous bone, joint end and other areas), this method cannot accurately match the physiological structure characteristics. Thirdly, the patent does not consider the integration of multi-dimensional clinical parameters to participate in the modeling process. The system only generates the resection plan based on the image information, and lacks the interactive modeling mechanism with pathological information (such as Ki-67 proliferation index, necrosis ratio), preoperative functional assessment indicators (such as load-bearing path, joint protection needs) and other data. As a result, the resection boundary it generates lacks the function-risk dual-dimensional judgment basis, does not support the need for dynamic adjustment of resection priority, and cannot respond to the clinical goal setting of weighing the retention / clearance strategy.

[0003] Fourthly, the patent fails to introduce spatiotemporal continuity or graph structure methods. When judging the diffusion of complex deteriorated areas, it lacks a mechanism to jointly model the functional status and spatial adjacency relationship between each voxel. It is unable to construct an activity gradient map to identify the three layers of tissue, namely the central lesion, the diffusion boundary and the latent erosion zone, and thus cannot provide doctors with an intuitive understanding of the evolution trend of the lesion. It lacks temporal and hierarchical judgment. Fifthly, the patent only focuses on the generation of the resection model itself, and lacks subsequent functions such as three-dimensional boundary elastic reconstruction, profile cluster re-approximation, and target weight-driven adjustment. It cannot dynamically adjust the boundary range according to the multi-target strategies such as function retention priority or complete removal priority set before the operation. In the highly weighted bone tumor resection, the patent cannot be applied to the surgery. Sixth, in terms of data-driven and case library mapping, the existing technology does not demonstrate any case-control + risk similarity mapping capabilities, and cannot establish associations between the current patient's functional-pathological characteristics and previous postoperative effects. It lacks reasoning support and case comparison capabilities, which is not conducive to doctors optimizing current surgical strategies based on past experience. Seventh, the overall plan does not introduce an interactive decision-making feedback mechanism, that is, doctors cannot slide and adjust the weights of different resection strategy plans before surgery, nor can they obtain real-time feedback and navigation prompts on structural hierarchical risks during surgery. This one-way output technical path limits its application convenience in clinical multi-team collaborative decision-making. Summary of the Invention

[0004] The purpose of the present invention is to provide a decision-making support system for the resection range of a deteriorated area of ​​a bone tumor, thereby resolving some of the drawbacks and deficiencies pointed out in the background art.

[0005] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: a bone tumor deterioration area resection range decision support system, comprising: modeling metabolic activity, blood perfusion, and tissue permeability biological characteristics reflected in different imaging modalities, including PET / MRI; using a spatial coupling modeling method to generate a functional activity gradient map within the deterioration area, mapping the three layers of the central lesion, diffusion boundary, and latent erosion zone; and establishing a dynamic interval labeling mechanism to form a variable resection range space.

[0006] Divide the image-generated gradient map into multiple regions, and jointly model each region with corresponding functional localization parameters and pathological risk parameters, where the functional localization parameters include tissue load-bearing capacity, joint proximity, and vascular and neural location, and the pathological risk parameters include tumor cell proliferation index and tissue necrosis risk value;

[0007] After stratifying the tissue structures including cortical bone, cancellous bone, medullary cavity, and soft tissue boundary, local resection profile clusters are generated according to the anatomical axis. Each profile cluster is linked to a single function / risk factor weight, allowing the user to reshape the entire resection volume by adjusting the resection target parameters. The resection boundary is a single elastic decision-making body with morphological adaptability in space.

[0008] Furthermore, the parameters derived from the tumor cell proliferation index and tissue necrosis risk value are used to construct two continuous function curves: one is the functional relationship between resection depth and the probability of tissue function loss, and the other is the functional relationship between resection depth and the reduced risk of lesion recurrence. Dynamic weight balancing calculations are performed on the two function curves to derive the critical resection level, which is used to guide the determination of the resection range of the deteriorated area of ​​the bone tumor.

[0009] By calculating the tumor cell proliferation index and tissue necrosis risk value, two continuous function curves that change with the resection depth d are constructed:

[0010] The first function curve: the relationship between the resection depth and the probability of tissue function loss:

[0011] It is used to describe the increasing trend of loss of key bone function in the target area as the resection depth increases. It has nonlinear increasing characteristics and integrates regional functional redundancy and stress transfer capacity.

[0012] The function definition is as follows:

[0013]

[0014] in:

[0015] F(d) represents the probability score of tissue function loss at resection depth d, with a value range of [0,1]; α represents the functional sensitivity coefficient, which depends on the anatomical location of the tissue; γ represents the structural damage response coefficient, which reflects the speed at which depth affects structural integrity; λ represents the structural nonlinear parameter, which describes the shape of the response curve of different tissue structures to resection. When λ>1, it indicates accelerated damage; μ·sin(ωd) represents the introduction of a micro-local structural perturbation term, which integrates the functional jump loss caused by structural nodes at a specific depth level and has physiological and anatomical rationality.

[0016] The second function curve: the relationship between the resection depth and the reduction in the risk of lesion recurrence:

[0017] This function is used to describe the extent to which the risk of local tumor recurrence is reduced as the resection depth increases. It takes into account the uncertainty of the tumor edge and the microdiffusion state, adopts a progressive descent model, and introduces the tissue pathological heterogeneity factor.

[0018] The function definition is as follows:

[0019] R(d)=β·(1―tanh(ηdκ))·(1+δ·cos(φd))

[0020] in:

[0021] R(d) represents the reduction in the risk of lesion recurrence at resection depth d; β represents the influencing factor of lesion biological activity; η represents the tumor margin reduction response rate; κ represents the growth permeability index, reflecting whether the tumor exhibits deep-layer extension characteristics; δ·cos(φd) represents the pathological microheterogeneity fluctuation term, which simulates the instability of abnormal cell density at different levels.

[0022] By constructing the above two functions, the judgment criteria are as follows: Under the preset surgical target weight conditions, find the resection depth d that makes the function difference balanced or the comprehensive cost minimized * , as the critical level for resection.

[0023] Furthermore, the method for constructing the three-layer region of mapping the central lesion, the diffusion boundary, and the latent erosion zone includes:

[0024] Acquiring multimodal medical imaging data of the target bone tissue region, the imaging data including structural imaging and functional imaging; performing spatial registration and three-dimensional reconstruction on the imaging data to construct a joint structural and functional expression relationship of the target region, wherein each voxel contains a corresponding structural attribute and functional state vector;

[0025] Based on the spatial adjacency relationship and functional state change degree between voxels, a spatial coupling graph structure is established, and the coupling strength between nodes is defined in the graph. Active diffusion modeling is performed in the three-dimensional graph structure based on the coupling strength to generate a functional activity gradient map that spreads from the center of the lesion to the periphery.

[0026] Furthermore, the structural attributes include anatomical hierarchical information of bone tissue, including cortical bone, cancellous bone and medullary cavity; the functional state vector includes one or more of metabolic activity index, blood perfusion parameter, signal heterogeneity index or tissue permeability parameter.

[0027] Furthermore, the spatial coupling graph structure uses graph nodes to represent voxel units, and the coupling strength of the graph edges is calculated by jointly calculating the functional state difference and structural continuity between voxels; the functional activity gradient graph is generated by a graph diffusion algorithm, which sets the initial high-activity node according to the lesion center and attenuates and diffuses toward the low-coupling path to form a gradient continuous distribution.

[0028] Furthermore, the method for jointly modeling the functional positioning parameters and the pathological risk parameters includes:

[0029] Based on the functional activity gradient map, corresponding gradient values ​​are generated, and the lesion area is divided into multiple spatial region sub-blocks; corresponding functional localization parameters are extracted for each of the region sub-blocks, and the functional localization parameters include the spatial relationship between the region and the anatomical structure; and corresponding pathological risk parameters are simultaneously extracted for each of the region sub-blocks, and the pathological risk parameters include the metabolic activity of the region, signal heterogeneity, or the probability of biological diffusion of the lesion.

[0030] Furthermore, the functional activity gradient map is generated based on the functional state changes from the center to the edge of the tumor, reflecting the continuity trend of the tissue from highly abnormal to the edge latent state; the functional localization parameters are generated by the identified three-dimensional anatomical map and associated with the target area through spatial distance measurement.

[0031] Furthermore, the pathological risk parameters are derived from a comprehensive evaluation of imaging data and preoperative puncture biopsy results, and a control mapping can be established with a previous case database.

[0032] Furthermore, the method for reconstructing the entire resection volume includes:

[0033] Constructing multiple two-dimensional spatial cross-sectional clusters around the tumor deterioration area, each cross-sectional cluster representing the anatomical structure and lesion distribution characteristics of the target area in a specific direction; associating each cross-sectional cluster with a functional factor or risk factor, including but not limited to the load-bearing function of local tissue, joint proximity, or neurovascular proximity; and risk factors including the biological activity of the lesion or the probability of pathological spread;

[0034] Receive the resection target parameters input by the user, and the resection target parameters are used to set the relative weights of the function retention priority and the lesion clearance priority; based on the resection target parameters, dynamically adjust the weight scores of the factors associated with each section cluster, and update the resection priority of each section cluster.

[0035] Furthermore, the two-dimensional section cluster is generated with the tumor main axis or the anatomical structure axis as the reference direction, covering the entire lesion deterioration area and the edge area; the functional factors and risk factors are encoded in numerical form as section attribute vectors and used to calculate the resection feasibility score of each section cluster; the three-dimensional resection boundary volume is generated by interpolating, fusing and spatially fitting the boundary positions of all section clusters.

[0036] The beneficial effects of this invention include: By integrating multimodal imaging (such as PET, MRI, and CT), spatially coupled modeling, and pathological data, it enables refined identification and hierarchical modeling of degraded areas of bone tumors, thereby accurately demarcating the central lesion area, diffuse boundary area, and latent erosion zone, providing physicians with a more scientific basis for determining the resection range and avoiding over- or under-resection. The system allows for input of specific surgical target parameters (such as prioritizing function preservation or complete removal), automatically adjusts the weighting of functional and risk factors for each cross-sectional area, and dynamically reconstructs the individualized resection range, enabling data-driven and goal-oriented flexible surgical planning to adapt to the physiological structure and clinical needs of different patients.

[0037] By accurately assessing the functional sensitivity of tissues (such as weight-bearing capacity, joint proximity, and neurovascular relationships), the system can avoid inadvertent resection of key functional structures, thereby maximizing the patient's postoperative mobility and anatomical integrity. It is particularly suitable for adolescents and the elderly, who have high demands for functional recovery. It not only relies on imaging features but also integrates pathological data (such as proliferation index and necrosis ratio) with case database comparison analysis to accurately identify areas at high risk of recurrence. This allows for full consideration of hidden diffusion zones in the setting of resection boundaries, improving the thoroughness of tumor removal and reducing the probability of postoperative recurrence at the source. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is the main flow chart of the bone tumor deterioration area resection range decision support system of the present invention.

[0039] Figure 2 This is a functional relationship diagram of the multi-parameter comprehensive score of the bone tumor area and the resection priority of the present invention.

[0040] Figure 3 This is a flowchart of the elastic reconstruction of bone tumor resection volume based on anatomical stratification and parameter driving of the present invention.

[0041] Figure 4 This is a schematic diagram of the implementation flow of the bone tumor resection decision support system based on function-structure layering and spatial coupling according to Example 1 of the present invention.

[0042] Figure 5 This is a flow chart of an embodiment of tumor partitioning and personalized resection auxiliary decision-making based on joint modeling of functional and pathological parameters according to embodiment 2 of the present invention.

[0043] Figure 6 This is a flowchart for generating a personalized bone tumor resection volume based on profile cluster scoring and three-dimensional fitting according to Example 3 of the present invention.

[0044] Figure 7This is an intraoperative anatomical image of a proximal phalanx tumor in Example 4 of the present invention. The surgeon performs dissection between the bone and the flexor mechanism (7A). After the dissection is completed, the two tissues are separated (7B). The surgeon can then complete the dissection on the side opposite to the original surgical approach (7C) and finally complete the resection by severing the metacarpophalangeal joint (7D).

[0045] Figure 8 This is an intraoperative image of the proximal phalanx exposed after en bloc bone resection in Example 4 of the present invention (8A); the resected tumor-bearing phalanx is placed next to the allograft that will replace it (8B). The allograft appears yellow because it was soaked in rifampicin solution during the removal surgery. DETAILED DESCRIPTION

[0046] The following is a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings.

[0047] Combined with attachment Figure 1 The bone tumor deterioration area resection range decision support system of the present invention performs multi-source information fusion and spatial structure modeling based on medical imaging data. The core goal is to accurately identify the different pathological levels of bone tumor deterioration areas, and on this basis, to achieve dynamic and adjustable personalized resection range recommendations. First, the system acquires multimodal medical imaging data, including positron emission tomography (PET) and magnetic resonance imaging (MRI) sequences, where PET images are used to provide spatial distribution information of tumor metabolic activities, and MRI sequences provide reflection signals of tissue structure, blood perfusion status, and tissue permeability. The system jointly models the functional biological indicators in these images and extracts the functional state vector of each voxel unit. The vector contains key indicators such as metabolic level, local blood supply, and molecular diffusion behavior; then, a spatial coupling modeling method is adopted to establish a functional state coupling network between voxels based on the spatial adjacency relationship between each voxel. By performing active diffusion modeling in the coupling graph, a network composed of the central and central functional states is constructed. A functional activity gradient map that gradually decays from the center to the outside reflects the transition process of tumor tissue from the highly active core area to the latent edge area; the system sets multiple threshold intervals according to the active gradient value, and divides the lesion area mapping into three-layer spatial structures, namely the central lesion area (high metabolic activity, clear pathological characteristics), the diffusion boundary area (structure is not damaged but the signal is abnormally active), and the latent erosion zone (no significant abnormality but there is metabolic risk); on this basis, the system further constructs a dynamic interval labeling mechanism to parameterize the above three layers of areas, so that their boundaries are adjusted in real time as the surgical strategy changes (such as clearance priority or function preservation priority), and finally forms a resection range spatial model with variability, layering and spatial responsiveness.

[0048] Combined with attachment Figure 2, based on medical image analysis and multi-source clinical information fusion, accurate judgment of the resection range of tumor deterioration area and personalized auxiliary suggestions; by processing multimodal medical imaging data (such as MRI, PET or CT, etc.), a three-dimensional functional activity gradient map representing the trend of tissue functional state changes is generated. The gradient map is based on the metabolic activity, signal heterogeneity or blood perfusion characteristics of the lesion, and reflects the functional attenuation process from the center to the edge of the lesion; then, the system performs spatial stratification according to the continuous value of the gradient map, and divides the lesion area into multiple spatial sub-regions with different activity intervals. Each sub-region is assigned a unique spatial label for subsequent parameter binding; then, the system introduces functional positioning parameters and pathological risk parameters at each sub-region level for joint modeling, wherein the functional positioning parameters are used to describe the functional importance and retention value of the region in the overall bone tissue structure, including the load-bearing nature of the tissue (for example, whether it is in a load-bearing state). At the same time, the system further introduces pathological risk parameters to model the biological characteristics of each region. The pathological risk parameters include tumor cell proliferation index (PI, Proliferation Index) and tissue necrosis risk value (NRS, Necrosis Risk Score). The former indicates the active reproduction level of tumor cells in the region, and the latter is used to assess whether there is a risk of severe ischemia, exudation or degeneration in the region. After weighting and vector combining the above functional positioning parameters and pathological risk parameters, the system constructs a comprehensive risk-function scoring model for each region, and generates a regional resection priority mapping table through the scoring model. The mapping table can be adjusted in real time according to different surgical target strategies to achieve quantitative auxiliary judgment of individualized resection range.

[0049] Combined with attachment Figure 3, based on anatomical structure stratification, function / risk factor binding and parameter-driven geometric reconstruction mechanism, it is used to achieve dynamic adjustability and spatial adaptive expression of tumor resection range. The bone tissue structure of the target area is carefully anatomically stratified and divided into four major structural layers: cortical bone, cancellous bone, medullary cavity and soft tissue boundary according to tissue properties. Each layer is independently modeled in space through image segmentation and three-dimensional reconstruction technology, and the spatial topological continuity is maintained; then, the system uses the main anatomical axis of the bone tissue or the tumor growth axis as the reference direction, and generates a series of local two-dimensional resection section clusters along the axis in the target area. Each section cluster is a set of continuous cross sections, representing the anatomical structure of the bone tissue at different depths, the range of the lesion and its relative position to the surrounding functional structures; within each section cluster, the system introduces a function / risk factor weight binding mechanism to quantify the local clinical importance of the area traversed by the section. The functional factors include, for example, the load-bearing level of the local bone structure, whether it is close to the joint surface, the density of neurovascular distribution, etc. The risk factors include the tumor metabolic intensity, biological activity, infiltration probability or neighboring tumors in the area. Based on the risk of damage to vital tissues, each section cluster is assigned a comprehensive factor score weight, which serves as an important reference parameter for subsequent geometric reconstruction. The system designs an interactive resection target input interface, allowing users to set or adjust surgical target parameters, such as function preservation priority, complete resection priority, structure protection area restriction, and other strategic goals. The system adjusts the weight threshold of each section cluster based on these parameters, and recalculates the connection relationship, boundary position and morphological trend between sections. By fitting the boundary points of all section clusters, the system regenerates a complete three-dimensional resection volume. The resection volume is not a rigid structure, but an elastic decision-making body with the ability to adaptively change its morphology in space. Its boundary can be dynamically deformed according to the target strategy, regional structural complexity and multi-factor weights to adapt to the anatomical differences and surgical plan preferences of different patients, thereby providing the optimal personalized bone tumor resection range recommendation while ensuring surgical safety and efficacy.

[0050] Example 1:

[0051] Combined with attachment Figure 4In this example, a 52-year-old male patient presented with right distal femoral pain and intermittent claudication for three months. Combined MRI and PET examinations revealed a focal hypermetabolic lesion in the medial cortical and cancellous bone regions of the right distal femur. The lesion was approximately 26.5 mm × 19.2 mm in size with unclear boundaries. The FDG-PET metabolic value (SUVmax) was 9.2. MRI indicated localized bone marrow signal loss with mild cortical erosion. Biopsy pathology confirmed a malignant bone tumor with a Ki-67 proliferation index of 48% and a necrosis index of 32%. Considering the risk of early infiltration of the lesion into the medullary cavity, surgical assessment of the resection depth was required. However, as the lesion was close to the medial femoral weight-bearing axis, excessive expansion of the resection range would affect the overall femoral support and walking function. Therefore, a dual-function judgment model was used to derive the critical resection level.

[0052] First, based on the tumor cell proliferation index (PI = 0.48) and the tissue necrosis risk value (NRS = 0.32), the correlation coefficient between biological activity and structural damage response was derived, and the following ranges were selected for parameter modeling: (1) The functional sensitivity coefficient α was set to 0.75, reflecting that the region was a partial load-bearing site and had a greater impact on functional loss; (2) The structural damage response coefficient γ was set to 0.6, indicating that the bone structure in this region was moderately sensitive to the resection depth; (3) The structural nonlinear parameter λ was set to 1.8, indicating that the deeper the resection, the faster the growth rate of functional damage; (4) The perturbation amplitude μ was set to 0.12, and the perturbation frequency ω was set to 4, indicating that there were alternating structural support units inside the femur and the structure was complex. Substituting the above values ​​into the first function curve:

[0053] First function:

[0054]

[0055] Substitute:

[0056]

[0057] Similarly, considering that the Ki-67 proliferation rate of the tumor is 48%, the edge penetration is unclear, but PET shows heterogeneous high metabolism, we set the biological activity influence factor β = 0.65, the edge clearance response rate η = 0.5, the growth penetration index κ = 1.6, the microheterogeneity perturbation δ = 0.08, and the fluctuation frequency φ = 3.5 to construct the second function as follows:

[0058] Second function:

[0059] R(d)=β·(1―tanh(ηd κ ))·(1+δ·cos(φd))

[0060] Substitute:

[0061] R(d)=0.65·(1―tanh(0.5·d 1.6 ))·(1+0.08·cos(3.5d))

[0062] The system then numerically calculated the two functions within the interval d = 0 to d = 25 mm (i.e., the theoretical maximum resection depth) and searched for the point where the function difference minimized. Based on the surgical goal of prioritizing function preservation while requiring a minimum 85% reduction in the risk of tumor recurrence, the system balanced the two functions using the comprehensive cost function Ψ(d) = w1·F(d) - w2·R(d), where w1 = 0.6 and w2 = 0.4, indicating a slight bias toward function preservation. Through derivation and iterative search, the system found a local minimum for Ψ(d) at d = 13.4 mm, satisfying R(d) ≈ ​​0.87 and F(d) ≈ ​​0.41. Resection at this depth level is considered to effectively eliminate the active lesion while avoiding significant damage to the load-bearing function of the femoral structure. The system therefore recommends constructing the resection boundary surface at a depth level centered at 13.4 mm. This boundary is displayed as a red dashed line in the preoperative planning diagram, and its spatial extension to the second level of the medullary cavity is marked. This is provided in real-time within the intraoperative navigation system.

[0063] Through spatial coupling graph modeling and active diffusion analysis, the tumor lesion area is subdivided into three spatial structures: central lesion, diffusion boundary and latent erosion zone, to support the hierarchical decision of resection boundary. The system obtains multimodal medical imaging data of patients, among which structural imaging comes from high-resolution spiral CT with a resolution of 0.5mm×0.5mm×1.0mm, which can clearly restore the boundaries of cortical bone, cancellous bone and medullary cavity; functional imaging comes from combined PET-MRI scanning, and the PET part provides tumor metabolic distribution map with SUVmax of 9.2, and the boundary shows diffuse high metabolism; MRIT1 and DWI sequences are used to identify soft tissue boundaries and signal heterogeneity areas. After signal normalization, it was found that the lesion had strong diffusion restriction on DWI, and the ADC value dropped to 0.78×10^-3mm 2 / s. Subsequently, the system performs registration processing on the multimodal imaging data, using a non-rigid registration algorithm based on gradient information to accurately align the PET and MRI layers within the CT anatomical framework. Through voxel resampling and interpolation reconstruction, unified three-dimensional volume data is generated, and each voxel is assigned structural attributes (such as the bone tissue layer and density value) and functional state vectors (such as SUV, ADC, T2 value, blood perfusion parameters, etc.) to form a spatial information point cloud. Next, the system uses each voxel as a graph node to construct a spatial coupling graph structure, defining the connection weight between each node and its spatial neighborhood in the graph. The weight value is determined by two factors: one is the physical space distance, and the other is the similarity difference of the functional state vector. The greater the coupling strength, the more anatomically continuous the region is and the smoother the functional state changes. The smaller the coupling strength, the more significant the functional mutation or boundary switching. After the graph structure is completed, the system performs active diffusion modeling on the graph, selecting the highest point of lesion metabolism as the active source point. The system then controls the diffusion path and speed based on the coupling strength, propagating the activity value layer by layer to the surrounding nodes. Each voxel is assigned a functional activity gradient value, forming a complete three-dimensional gradient map. The system then divides the regions according to preset thresholds: regions with functional activity greater than 0.75 are classified as central lesion areas, indicating that the tissue is in a state of hypermetabolic abnormality or obvious structural damage, and requires complete resection first. Regions with functional activity between 0.35 and 0.75 are defined as diffusion boundary zones, indicating that although the tissue is not completely destroyed, there are signal abnormalities such as increased metabolism, blurred lesion boundaries, or decreased tissue elasticity, and appropriate resection is recommended according to the strategy. Regions with activity below 0.35 but still greater than twice the standard deviation of the background noise are identified as latent erosion zones. These areas have weak signals, but due to significant spatial adjacency and coupling, they have the potential to penetrate deep structures and are suitable for conservative resection or close postoperative follow-up. Taking this patient's data as an example, the central lesion area was identified as 6.1cm in the 3D reconstruction. 3 , the diffusion boundary area is 10.7cm 3 The potential erosion zone is 18.3 cm 3 The three layers are marked with different colors (red, orange, and light blue) and superimposed on the original CT anatomical map for preoperative planning reference. The system can further filter the resection level according to the surgical target parameters. For example, when the resection strategy is set to resect the central lesion + part of the boundary area and retain the latent zone, the system generates a 7.9 cm 3 The boundary model of the resection volume is provided to the intraoperative navigation system to assist in spatial positioning.

[0064] The system further uses the constructed three-dimensional imaging model to encode the structural attributes and functional state vectors of each voxel in the lesion area, and uses this for spatial modeling and resection judgment of tumor deterioration areas. In terms of structural attributes, the system uses high-resolution CT images to perform hierarchical identification of bone tissue. It first segments the three main anatomical layers of bone tissue: cortical bone, cancellous bone, and medullary cavity. The cortical bone is identified as a high-density continuous structure with a thickness of approximately 4.6 mm, distributed in the outer layer of the femur. The cancellous bone, as an intermediate transition layer, has a density lower than that of the cortical bone but higher than that of the medullary cavity tissue. It occupies the internal space of the bone shaft and is filled with spongy trabecular structures. The medullary cavity is located in the center of the femur, has a low signal intensity, and is mainly composed of fat tissue and bone marrow components. The system accurately segments and labels the three based on the CT value gradient and anatomical atlas, giving each voxel a clear structural label. In terms of constructing functional state vectors, the system integrates multimodal information from PET and MRI to generate a four-dimensional functional feature vector for each voxel, including metabolic activity indicators, blood perfusion parameters, signal heterogeneity index, and tissue permeability parameters. Among them, metabolic activity is derived from the standard uptake value (SUV) of PET. For this patient, the maximum SUV in the central lesion area was 9.2, and the average was 7.8, which was significantly higher than the normal bone tissue background value of 1.2. Blood perfusion parameters were obtained from dynamic contrast-enhanced MRI (DCE-MRI). The perfusion peak occurred at 37 seconds in the T2 sequence, indicating enhanced vascular permeability and short blood flow retention time in this area. The signal heterogeneity index was calculated based on the range of DWI image signal changes. The heterogeneity value in the central lesion area was as high as 0.76, reflecting significant differences in cell distribution and tissue density in this area. The tissue permeability parameter was derived from diffusion tensor imaging (DTI). The diffusion anisotropy value in this patient's lesion area was significantly decreased, indicating disordered cell arrangement and damaged barrier structure. Based on this information, the system records the anatomical level (e.g., cortical bone) and the corresponding functional state vector (e.g., SUV = 8.1, perfusion value = 0.83, heterogeneity index = 0.71, and permeability parameter = 0.58) for each voxel. This information is then used for subsequent spatial coupling map modeling and functional activity gradient calculation. The system incorporates a parameter weighting module that adjusts the importance of each functional parameter based on the voxel's level. For example, in cortical bone, signal heterogeneity and permeability are given higher weights to reflect the probability of structural integrity compromise; in cancellous bone, metabolic activity and blood perfusion are more critical, reflecting the tumor's biological activity; and in the medullary cavity, perfusion changes and diffusion permeability are prioritized to identify potential signs of metastasis. The system utilizes this encoded information to construct a comprehensive spatial-structural-functional three-dimensional atlas. The system allows for interactive viewing of the structural level, functional state, and overall activity score of any voxel at any level, enabling a multidimensional assessment of every micro-unit within the lesion.For example, in this patient, a group of voxels in the cancellous bone region of the medial right femur showed a metabolic value of 8.2, a perfusion value of 0.79, a heterogeneity index of 0.74, and a permeability of 0.65. These voxels were also labeled as cancellous bone layers. Based on this, the system determined that this region was a highly functionally active area, belonging to the transitional layer from the diffusion boundary to the central lesion, and should be prioritized for resection planning. Another group of voxels in the medullary cavity showed a metabolic value of 2.6, low perfusion, and mild heterogeneity. The system classified this as a latent erosion zone, recommending that resection be withheld for now but requiring close postoperative monitoring. Ultimately, the entire deteriorated area was structurally hierarchically labeled and functionally labeled with activity. This information allows for the selection of resection strategies (such as preserving the medullary cavity and removing cancellous and cortical bone lesions) in the preoperative assessment report. The system then reconstructs a three-dimensional resection boundary model and superimposes it on the surgical field view in real time using augmented reality during surgery, enabling precise navigation and boundary recognition.

[0065] The system's feasibility in complex anatomical environments was verified using parameters and lesion data to establish the spatial coupling graph structure and generate the functional activity gradient map. After completing multimodal image 3D registration and voxel-level data structuring, the system established a spatial coupling graph model for the patient's medial femoral tumor region. This model, consisting of approximately 1,120,000 nodes representing each individual 3D voxel, covers the entire region from cortical bone to cancellous bone and part of the medullary cavity. Each voxel is initialized and assigned a value based on its functional state vector and structural attributes. The coupling strength of graph edges is calculated using a joint scoring mechanism that integrates structural continuity and functional state diversity. Structural continuity is determined by the physical positional relationship between voxels and tissue type similarity. For example, if two adjacent voxels are located in the cancellous bone region and their relative position difference is less than 1.5 mm, the structural continuity score is high. Functional state diversity is calculated based on the distance between multidimensional vectors, such as metabolic activity values, perfusion indices, and heterogeneity indices. Highly consistent functional states between adjacent voxels result in a high score, while negative coupling strength is low. Taking this patient as an example, a pair of adjacent voxels located inside the cancellous bone, the functional state vector of voxel A is (SUV = 7.8, perfusion = 0.74, heterogeneity = 0.69), and that of voxel B is (SUV = 7.5, perfusion = 0.71, heterogeneity = 0.68). The Euclidean distance of the functional state between the two is 0.045, and the structural position distance is 1.2 mm. The system calculates its coupling strength score as 0.92, indicating that the two are highly consistent in structure and function, and are suitable as a high-pass path for the diffusion of lesion activity; while the other pair of voxels A' and B', located at the boundary between the cancellous bone and the medullary cavity, respectively, have a large difference in state, with a coupling score of only 0.41, and the diffusion path resistance is significantly increased. Next, the system executes the functional activity graph diffusion algorithm on the spatial coupling graph, selects the core voxel with the highest metabolic value in the PET image (SUVmax=9.2) as the high-activity source node, assigns its initial activity to 1.0, and then diffuses the activity value outward from the graph layer by layer according to the coupling strength. The diffusion follows an attenuation strategy of prioritizing high-coupling paths and restricting low-coupling paths. In each diffusion iteration, the system dynamically adjusts the activity value according to path resistance, propagation distance, and local gradient changes to ensure that information propagation conforms to structural anatomical logic and reflects the trend of biological function attenuation.After 20 rounds of diffusion, the system successfully generated a complete functional activity gradient map, in which the activity distribution value gradually decreased from 1.0 to 0.05, showing an obvious continuity of transition from high activity in the center to the outer layer; viewed in slice view, the patient's lesion was active in the 11th layer (close to the cortical bone surface) with an activity of 0.89, in the 18th layer (deep in the cancellous bone) with an activity of 0.61, in the 26th layer (marrow cavity transition zone) with an activity of 0.32, and the boundary activity of the 31st layer dropped to 0.14. Based on this, the system divided the lesion area into three layers: central lesion, diffusion boundary, and latent erosion zone, and displayed the structural distribution of different areas in red, orange, and blue on the preoperative three-dimensional visualization platform. Based on this gradient map, the spatial evolution trend of the current lesion and the risk position of different levels of tissue can be intuitively judged. After setting the surgical strategy goal to prioritize the removal of central lesions and high-risk diffusion zones and retain latent areas, the system locks all areas with an activity greater than 0.65, generates a three-dimensional resection boundary volume, overlaps it on the patient's individual skeletal model for preoperative navigation simulation, and simultaneously pushes it to the intraoperative AR navigation system to realize real-time visualization prompts of resection.

[0066] Example 2:

[0067] Combined with attachment Figure 5Based on Example 1, a joint modeling method of functional localization parameters and pathological risk parameters is used to refine the resection boundary. Previously, the patient had completed multimodal image acquisition, three-dimensional registration, and spatial coupling map modeling, and generated a functional activity gradient map based on PET-MRI fusion. Based on this map, the system first performed block segmentation according to the gradient value, dividing the entire lesion area into multiple spatial area sub-blocks with similar functional activity. Each sub-block represents a local lesion state unit, generating a total of 64 sub-blocks. Each sub-block contains approximately 300-800 voxel units, and the central activity value is distributed between 0.1 and 1.0. The system uses 0.15 as the activity progressive unit for stratification to form a hierarchical zoning. The system then extracts functional localization parameters for each sub-block. These parameters measure the spatial relationship between the sub-block and key anatomical structures, including distance from the articular surface, location within the load-bearing axis of the bone, and proximity to major neurovascular pathways. Using a three-dimensional vector localization method, the system calculates the Euclidean distance between the sub-block's center of mass and the boundaries of key structures and assigns a risk level based on an empirical function. For this patient, sub-block B13 is located in the middle layer of the medial femoral cancellous bone, 9.4 mm from the knee cartilage interface, on the primary longitudinal load-bearing path of the femoral shaft, and 3.2 mm from the femoral neural foramen. This area was assigned a functional localization risk value of 0.82 (high risk), indicating its criticality for maintaining lower limb motor function. Sub-block B22, located at the edge of the medullary cavity, away from the weight-bearing line and nerve course, received a score of only 0.21, indicating a functionally preserved area. At the same time, the system also extracts pathological risk parameters for each sub-block, including three core indicators: metabolic activity (calculated by taking the average of the mean and maximum SUV values ​​of the sub-block), signal heterogeneity (calculated by taking the DWI standard deviation and ADC coefficient of variation of the sub-block), and the biological diffusion probability of the lesion (prediction of spatial penetration probability based on the diffusion tensor model and empirical training data). Taking the B13 sub-block as an example, the mean SUV is 7.4, the heterogeneity index is 0.68, and the diffusion probability is 0.77. The system integrates the three parameters into a pathological risk score of 0.81 through a weighting function; the pathological risk value of the B22 sub-block is 0.38, which belongs to a low-activity and low-permeability area. Next, the system jointly models the functional positioning value and pathological risk value for each sub-block, constructs a two-dimensional joint scoring matrix, and calibrates the comprehensive risk position of each sub-block in the matrix. The resection priority of the sub-block is then determined by the set resection strategy weight. For example, if the current surgical goal is to prioritize the resection of areas with high pathological risk and functional redundancy, and retain key functional areas, the system will set sub-blocks with high pathological risk and low functional positioning value in the matrix (such as B29, pathology 0.79, function 0.34) as priority resection areas, while areas with high pathology and function (such as B13) are set as conservative treatment areas, and areas with low pathology and function are set as selective resection areas.Based on this information, the system generates multiple boundary resection recommendations in three-dimensional space, allowing real-time adjustments via a sliding strategy weight bar. For example, increasing the function preservation weight from 0.4 to 0.7 adjusts areas near tendon attachments, such as B17 and B19, from recommended resection to preserved areas. The three-dimensional resection boundary model is recalculated in real time, and the different resection levels are displayed with varying degrees of transparency on the preoperative planning platform. During surgery, a navigation device is used to view the marked sub-block structures. The system indicates high pathological risk or proximity to important anatomical structures at the edge of each resection level, assisting in the precise positioning and stratified treatment of tumor tissue.

[0068] The functional activity gradient map and functional localization parameters were further used to establish a spatial behavior prediction model, and the resection strategy was fine-tuned based on the functional state evolution trend. The system used the functional activity gradient map generated by fusion of PET and MRI information. The highest point of tumor metabolism was designated as the central active source, with activity at that location set to 1.0. This activity then diffused outward, depending on the coupling strength and functional state differences. This constructed an activity gradient distribution map reflecting the continuous trend from highly abnormal central areas to latent states at the edges. Based on the gradient variation trend, the system divided the lesion area into multiple functional zones. For example, in this case, six levels of activity zones were generated, extending from the central lesion to the medullary cavity. Each level represents a continuous interval with activity decreasing by approximately 0.15. For example, Level 1 (activity 0.85–1.0) represents a highly active core area, Level 2 represents a diffuse area of ​​moderate to high activity (0.7–0.85), and Levels 3 through 6 decrease in activity. Level 6 activity is approximately 0.1–0.25, primarily distributed in the distal medullary cavity and some adjacent soft tissue areas. Next, the system further extracts the functional positioning parameters of each functional band-level area. The anatomical data used comes from a pre-loaded standard three-dimensional structure atlas of the femur, which contains anatomical labels (such as cortical bone, cancellous bone, medullary cavity, articular surface, femoral neural foramen, etc.) and three-dimensional spatial coordinates. The system reconstructs the bone model of the target patient and performs spatial registration with the standard atlas, so that each structural label is accurately mapped to the patient's specific bone. On this basis, the system performs spatial association judgment by calculating the Euclidean distance between the center point of the functional sub-block and the boundary of the key structure, and calculates the functional sensitivity score of each area using the anatomical functional risk function. For example, in the Level 2 area, the patient identified a sub-block numbered F17, located below the cancellous bone, only 2.7 mm away from the medial femoral neural foramen, and less than 8 mm away from the knee joint surface. The system comprehensively determined that it was a high functional sensitivity area and assigned a functional risk score of 0.88; another Level 4 sub-block numbered F33 was located at the distal end of the medial medullary cavity, without overlapping with the main structure, and more than 20 mm away from all anatomical elements. Its functional risk score was only 0.22. Based on this, the system judged that its retention value was low, and if the pathological risk was high, it could be cleared first. The system jointly visualizes functional activity (gradient map) and functional sensitivity (spatial correlation map), and displays the spatial coordinate map of each area in two dimensions, namely activity level and structural criticality, in a three-dimensional planning interface. It supports interactive click-to-view of a certain area. After clicking the F17 area, a pop-up interface shows that its functional positioning is a high-sensitivity risk area, with an activity range of 0.71–0.82, a structural type of cancellous bone, and a critical depth of approximately 13 mm. It is recommended to adopt a cutting-edge and center-preserving method according to the surgical strategy, that is, to remove the peripheral high-pathological area and retain the central load-bearing path.When the system weights were set at 70% for function retention and 30% for thoroughness of removal, the system excluded all seven areas with a function sensitivity greater than 0.8 and an activity less than 0.65, retaining areas with strong functions but not yet significantly activated pathology. The system also focused on removing 12 areas with an activity greater than 0.75 and a function sensitivity less than 0.6, with a total resection volume of approximately 14.8 cm. 3 Compared to traditional en bloc resection, this approach reduces bone removal by 31%. During intraoperative navigation, these areas are highlighted, providing real-time reminders of areas near articular surfaces or neural structures. When the knife approaches the F17 edge, the system provides a voice prompt indicating that a critical structural zone is about to be entered, prompting confirmation of the path, thereby improving surgical safety and accuracy.

[0069] The acquisition and modeling method of pathological risk parameters improves the accuracy and predictive ability of risk assessment through comprehensive evaluation of imaging data and preoperative puncture biopsy results, and comparison and mapping with the previous case database. In the early stage of the patient's admission, PET / MRI examination revealed a metabolic abnormality in the medial area of ​​the distal right femur, with an SUVmax of 9.2, significant T2-weighted signal heterogeneity, DWI showing high signal, and ADC value decreased to 0.78×10-3 mm 2 / s, combined with imaging features, the system initially determined it to be a lesion with high metabolic activity. A subsequent image-guided biopsy revealed pathology suggestive of osteosarcoma, with prominent mitotic figures, a Ki-67 proliferation index of 48%, a high tumor cell density, and small foci of necrosis in some areas, with a necrosis rate of 32%. Based on this data, the system extracts three types of pathological risk parameters: first, metabolic activity from imaging. By integrating PET SUVmax / SUVmean, regional perfusion values, signal heterogeneity scores, and diffusion restriction, an imaging-level metabolic score is generated for each sub-block. For example, region C21 has an average SUV of 7.6, a heterogeneity score of 0.71, and a perfusion index of 0.74, resulting in a metabolic risk score of 0.82. Second, histopathological parameters include cell density, nuclear atypia grade, necrosis ratio, and Ki-67 index. The system read the Ki-67 as 48%, necrosis as 32%, and high density from the biopsy report, and assigned a histological risk score based on the existing parameter training model. The C21 pathology score of this area was 0.79. Third, the system weighted the two scores to form a comprehensive pathological risk value. The system then compared the regional characteristics with 256 bone tumor cases in a built-in case database for multidimensional vector similarity. The database includes typical imaging and pathological comparison data for benign, borderline, and malignant bone tumors. The system extracted the current region's seven-dimensional risk vector (SUV, ADC, T2 value, heterogeneity index, perfusion index, Ki-67, and necrosis ratio) and compared it with all samples in the database for cosine similarity. The system ultimately found that the current region C21 had a similarity greater than 0.91 with the regional characteristics of three patients in the database with pathologically positive margins and postoperative recurrence. The system classified it as a high-recurrence-risk, high-bioactivity region and marked it as a red warning zone. In contrast, region C31, although having an SUV of 4.1 and a heterogeneity index of 0.48, showed a Ki-67 of only 18% and no necrosis on pathology. Its similarity with multiple postoperative margin-negative cases in the database was only 0.51, resulting in the system being classified as a low-risk observation zone. In the preoperative planning system, weights are adjusted based on the pathological risk level layer, combined with the structural and functional layers. High-risk pathological areas are prioritized for resection. A prediction report is generated for postoperative follow-up data matching high-similarity areas in the database. The system shows that the recurrence rate is approximately 42% when the resection range of high-risk areas is less than 5 mm, decreasing to 12% when the resection range is greater than 8 mm. Based on this, the current resection strategy is to extend resection ≥10 mm in high-risk pathological areas and preserve function as much as possible in low-risk areas. The intraoperative navigation system uses the pathological risk images to mark the boundaries of the surgical area with red, yellow, and blue areas. The red area indicates where the resection margin requires a 10 mm extension. The navigation system dynamically prompts the user to extend the resection margin when entering high-risk pathological tissue. Postoperative pathology confirms the absence of residual active cells at the edge of the red area, scar tissue at the edge of the yellow area, and intact tissue and good functional recovery in the blue area.

[0070] Example 3:

[0071] Combined with attachment Figure 6Based on Example 2, the patient has completed three-dimensional reconstruction of the tumor area and generation of an active gradient map through multimodal imaging. Based on this, the system constructs a series of two-dimensional spatial profile clusters around the tumor deterioration area with a profile spacing of 1.5 mm. A total of 22 profiles are generated, and the distribution direction is expanded along the longitudinal axis of the femur. Each profile cluster covers the cortical bone, cancellous bone and part of the medullary cavity area, and at the same time superimposes the distribution outline of the lesion within its coverage area. Within each profile cluster, the system extracts and associates functional factors with risk factors. Functional factors include weight-bearing function scores, joint proximity scores, and neurovascular proximity scores. For example, profile 8 is located in the mid-medial femoral region, with its superior region 8.2 mm from the knee joint surface, within the long-bearing bearing weight line, and only 3.7 mm from the femoral neural foramen. Therefore, the total functional factor score for this profile is 0.87 (with a maximum score of 1, with higher scores indicating greater criticality). Regarding risk factors, the mean metabolic activity of the lesion area covered by this profile is 7.4, the pathological Ki-67 index is 45%, the DWI signal heterogeneity is 0.69, and the pathological diffusion model predicts a diffusion probability of 0.81. The system assigns this profile a risk factor score of 0.84. The system establishes a matrix combining the functional and risk factor scores of all profile clusters to facilitate dynamic weight adjustments based on subsequent surgical strategies. The system then set resection target parameters: the initial strategy was a function preservation priority of 0.6 and a lesion clearance priority of 0.4. The system then calculated the comprehensive resection priority for each profile cluster using the following formula: weight multiplication followed by linear normalization: final resection score = 0.6 × (1 – function score) + 0.4 × risk score. For profile 8, with a function score of 0.87 and a risk score of 0.84, the calculated final resection score was 0.6 × (1 – 0.87) + 0.4 × 0.84 = 0.6 × 0.13 + 0.4 × 0.84 = 0.078 + 0.336 = 0.414, placing it in a medium-to-low priority category. However, profile 15, with a function score of only 0.34 and a risk score of 0.91, was assigned a score of 0.6 × (1 – 0.34) + 0.4 × 0.91 = 0.396 + 0.364 = 0.76, placing it in a high-priority resection category. The system reclassified all sections based on their scores and annotated their boundaries with different colors in three-dimensional space. Scores above 0.7 indicated a strong recommendation for resection, 0.4–0.7 indicated conditional resection, and below 0.4 indicated a priority preservation zone. The system then refitted the section edge point cloud based on this score hierarchy and generated a continuous and smooth three-dimensional resection boundary volume using a slice interpolation method. Boundary changes were observed in real time on the planning platform, and the strategy was adjusted to prioritize lesion removal (with a weighting of 0.7:0.3). After recalculation, the score for section 8 increased to 0.3 × (1–0.87) + 0.7 × 0.84 = 0.039 + 0.588 = 0.627, indicating a higher priority for resection. The margin area previously marked as a preservation zone was dynamically converted to a recommended resection zone.In this way, the system dynamically redraws the entire resection volume and supports multiple rounds of continuous weight adjustment and visual feedback of results. After the resection strategy is finally determined, it can export an individualized resection plan file, including the three-dimensional coordinates of the resection volume, cross-sectional structure and resection sequence recommendations, which can be directly loaded into the intraoperative navigation system to assist in completing precise layered resection operations.

[0072] The method for constructing resection volumes based on two-dimensional section clusters involves generating sections using the tumor's principal axis as a reference, converting functional factors and risk factors into attribute vectors for section scoring, and fitting the final three-dimensional resection boundary volume from all section clusters. After MRI and CT registration, the system identifies the lesion's spatial principal axis, which is slightly inclined 5.3 degrees along the longitudinal direction of the femur and has a principal axis length of approximately 31.8 mm, covering the entire process of the tumor penetrating from the medial cortical bone to the cancellous bone and part of the medullary cavity. The system then uses this principal axis as a baseline and generates a total of 22 two-dimensional section clusters at 1.5 mm intervals. Each section is 1.5 mm thick, and its width and depth are tailored to the bone cross-section, forming a spatial segment that covers the tumor area and extends outward from the edge by at least 5 mm. Within each section, the system identifies boundary structures, lesion location, adjacent functional tissues, and pathological status, and generates a set of numerical attribute vectors for each section. The functional factor encoding includes three dimensions: first, load-bearing value, which is calculated based on whether the section landing point is in the load-bearing path of the backbone, combined with CT three-dimensional load simulation to calculate the score. In this case, the 9th section is located on the cortical bone pressure point, and the load-bearing value is 0.91; second, joint proximity, which calculates the shortest distance between the section and the knee joint surface. When it is less than 10 mm, it is high risk, and the system assigns 0.84 to the 3rd section; third, neurovascular proximity, which is based on the distance assessment from the center point of the section to the identified femoral neural foramen or main vascular path. For example, the 14th section is 7.2 mm away from the neural path and has a score of 0.79. Risk factors are also encoded as section attribute vectors, including metabolic activity (such as mean SUV), pathological activity indicators (such as Ki-67 index and necrosis percentage), and diffusion probability (the probability of lesion invasion depth determined by the graph diffusion algorithm). In section 11, the mean SUV was 7.8, Ki-67 was locally measured at 52% on needle biopsy, and the predicted diffusion probability was 0.83. The system assigned this composite risk score to 0.86. All factors were normalized to a score vector ranging from 0 to 1 and input into the resection scoring module. This module calculates a resection feasibility score for each section cluster based on the configured surgical strategy weights (e.g., currently set to 60% function preservation and 40% clearance priority). Higher scores indicate a higher likelihood of inclusion in the final resection volume, while lower scores recommend preservation. Based on this score, the system generates local resection margins for each section and marks boundary points within each 2D section, forming 22 sets of planar margins. Subsequently, the system enters the 3D fitting module and performs spatial interpolation and fusion on all section boundaries. First, a spline-based method is used to continuously connect the key points of each section boundary line to solve the boundary jump problem caused by the irregular structure between sections. Then, a complete 3D resection boundary volume is constructed through volume fitting and boundary smoothing. During the fitting process, the structural continuity is retained and the formation of morphological depressions or arc-shaped edges in the highly sensitive areas of the structure is avoided to avoid entering the inappropriate operation area during the operation. In this patient, the final generated 3D resection volume was 15.6cm 3Compared to the initial rectangular bounding volume, this volume reduced resection redundancy by approximately 38%. This volume was displayed as a transparent blue solid in the preoperative 3D model, and a suggested resection sequence was provided. During surgery, the navigation system loaded this bounding volume data and calibrated the surgical blade position and section index in real time. Following the instructions, the lesion was removed in layers in sections 6–16, avoiding the high-function areas in sections 3 and 9. Postoperative pathology confirmed negative margins and intact functional areas.

[0073] Example 4:

[0074] The precise resection system of bone tumor deterioration area based on functional gradient map and anatomical functional risk factor is implemented in this embodiment. Figure 7 and Figure 8 The surgical example shown demonstrates the actual application process of the bone tumor deterioration area resection system based on functional active gradient modeling and coupling with anatomical structure function / risk factors, especially for bone tumor lesions in the area adjacent to the joint.

[0075] This example selects a patient's right distal radius tumor lesion area, collects multimodal imaging data, including T1 contrast-enhanced MRI, DWI, and PET-CT, and completes spatial registration and three-dimensional reconstruction of the images. The following biological parameters are obtained:

[0076] PET metabolic uptake rate (SUV): reflects the degree of cell metabolic activity;

[0077] DCE-MRI perfusion parameter (Ktrans): indicates vascular permeability;

[0078] ADC value (apparent diffusion coefficient): reflects the changes in tissue cell density;

[0079] T2 signal heterogeneity index (Heterogeneity Index, HI): reflects the degree of blurring of the lesion boundary.

[0080] The above multidimensional indicators are normalized to construct the functional state vector Where each f ij is the j-th functional index value on voxel i.

[0081] Based on the spatial adjacency matrix A, a coupling graph G = (V, E) is constructed, and the coupling strength between nodes in the graph is defined as:

[0082]

[0083] in:

[0084] α, β are weight coefficients; D ijis the Euclidean distance between voxels or the structural continuity penalty factor. Set the central lesion as the high activity source point, and use the graph diffusion algorithm to generate the functional activity gradient map G that spreads from the lesion to the edge. grad .

[0085] like Figure 7 As shown in A to D, three layers of structural areas were identified during surgery:

[0086] Central lesion area: highly metabolically active (SUV>6.0), high signal heterogeneity, and clear morphology;

[0087] Diffusion border zone: decreased PET activity (SUV 2.5-5.0), moderate perfusion parameters, and irregular borders;

[0088] Latent erosion zone: fuzzy boundaries, high heterogeneity index but low metabolic value.

[0089] For each region voxel, define the joint risk function:

[0090] R i =λ1·PI i +λ2·HI i +λ3·DF i

[0091] in:

[0092] PI i : Tumor cell proliferation index (via Ki-67 labeling or imaging replacement); HI i : Heterogeneity index; DF i : probability of tumor spread (from the prediction of functional status change trend); λ1, λ2, λ3: artificially set risk weights; by combining Figure 8 The intraoperative processing flow shown in AB embeds anatomical functional factors (joint proximity, neurovascular course, bone cortical thickness, etc.) into functional area division. Define each two-dimensional section cluster P k Function retention importance factor FI k RI k :

[0093] S k =w f ·(1―FI k )+w r ·RI k

[0094] in:

[0095] S k : Resection priority score of the kth section cluster

[0096] w f ,wr : User input function - risk weight parameter

[0097] By adjusting the parameter w f :w r proportions, achieving reconfigurable individualized resection volume design.

[0098] Construct the relationship between resection depth d and the probability of functional loss L(d) and the reduction value of recurrence risk R(d):

[0099] L(d)=1―exp(―γd), R(d)=1―exp(―δd)

[0100] The two functions are fitted in the patient data training set through the Bayesian model, and the final objective function is:

[0101] C(d)=ωL(d)+(1―ω)(1―R(d))

[0102] Solve the minimum value of C(d), that is, obtain the optimal resection critical layer depth d.

[0103] Figure 7 and Figure 8 For a visual example of the resection process during surgery: Figure 7 A to D show lesion visualization achieved through multi-angle exposure of soft tissue and cortical bone; Figure 8 In A, a curette and needle holder are used to expose the intraosseous interface; Figure 8 B indicates the intact bone tissue sample after resection, which is highly consistent with the 3D image reconstruction volume (error <1mm). Throughout the entire process, the system provides real-time feedback on resection depth, section priority, and the risk layer to which the current area belongs, enabling precise intraoperative resection navigation.

[0104] Postoperative tissue biopsy confirmed the absence of residual margins; the predicted risk of recurrence decreased to 28% of the original model's average value; wrist range of motion was retained to over 90% of preoperative values, and bone stability was well maintained. Thus, this example effectively validates the engineering feasibility and clinical value of the "Bone Tumor Degraded Area Resection Range Decision Support System," which boasts excellent personalized adaptability and high spatial accuracy, making it suitable for the design and implementation of resection paths near joints for various types of long bone tumors.

[0105] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A decision-making support system for the resection range of degraded bone tumor areas, characterized by: Modeling of metabolic activity, blood perfusion, and tissue permeability biomarkers captured by various imaging modalities including PET / MRI; A spatial coupling modeling method is used to generate a functional activity gradient map within the degraded area, mapping the three layers of the central lesion, diffusion boundary, and latent erosion zone. A dynamic interval labeling mechanism is constructed to form a variable resection range space. Divide the image-generated gradient map into multiple regions, and jointly model each region with corresponding functional localization parameters and pathological risk parameters, where the functional localization parameters include tissue load-bearing capacity, joint proximity, and vascular and neural location, and the pathological risk parameters include tumor cell proliferation index and tissue necrosis risk value; After stratifying the tissue structures including cortical bone, cancellous bone, medullary cavity, and soft tissue boundary, local resection profile clusters are generated according to the anatomical axis. Each profile cluster is linked to a single function / risk factor weight, allowing the user to reshape the entire resection volume by adjusting the resection target parameters. The resection boundary is a single elastic decision-making body with morphological adaptability in space.

2. The bone tumor deterioration area resection range decision support system according to claim 1 is characterized in that The parameters obtained from the tumor cell proliferation index and tissue necrosis risk value are used to construct two continuous function curves: one is the functional relationship between the resection depth and the probability of tissue function loss, and the other is the functional relationship between the resection depth and the reduced risk of lesion recurrence. The two function curves are dynamically weighted and balanced to derive the critical resection level, which is used to guide the determination of the resection range of the deteriorated area of ​​the bone tumor.

3. The bone tumor deterioration area resection range decision support system according to claim 1, characterized in that The method for constructing the three-layer region of mapping the central lesion, the diffusion boundary, and the latent erosion zone includes: Acquiring multimodal medical imaging data of the target bone tissue region, the imaging data including structural imaging and functional imaging; performing spatial registration and three-dimensional reconstruction on the imaging data to construct a joint structural and functional expression relationship of the target region, wherein each voxel contains a corresponding structural attribute and functional state vector; Based on the spatial adjacency relationship and functional state change degree between voxels, a spatial coupling graph structure is established, and the coupling strength between nodes is defined in the graph. Active diffusion modeling is performed in the three-dimensional graph structure based on the coupling strength to generate a functional activity gradient map that spreads from the center of the lesion to the periphery.

4. The bone tumor deterioration area resection range decision support system according to claim 3, characterized in that The structural attributes include anatomical hierarchical information of bone tissue, including cortical bone, cancellous bone, and medullary cavity; the functional state vector includes one or more of a metabolic activity index, a blood perfusion parameter, a signal heterogeneity index, or a tissue permeability parameter.

5. The bone tumor deterioration area resection range decision support system according to claim 4 is characterized in that The spatial coupling graph structure uses graph nodes to represent voxel units, and the coupling strength of the graph edges is calculated by jointly calculating the functional state differences and structural continuity between voxels; the functional activity gradient map is generated by a graph diffusion algorithm, which sets the initial high-activity node according to the lesion center and attenuates and diffuses toward the low-coupling path to form a gradient continuous distribution.

6. The bone tumor deterioration area resection range decision support system according to claim 1, characterized in that The method for jointly modeling the functional positioning parameters and the pathological risk parameters includes: Based on the functional activity gradient map, corresponding gradient values ​​are generated, and the lesion area is divided into multiple spatial region sub-blocks; corresponding functional localization parameters are extracted for each of the region sub-blocks, and the functional localization parameters include the spatial relationship between the region and the anatomical structure; and corresponding pathological risk parameters are simultaneously extracted for each of the region sub-blocks, and the pathological risk parameters include the metabolic activity of the region, signal heterogeneity, or the probability of biological diffusion of the lesion.

7. The bone tumor deterioration area resection range decision support system according to claim 6, characterized in that The functional activity gradient map is generated based on the functional status change from the center to the edge of the tumor, reflecting the continuous trend of the tissue from highly abnormal to marginal latent state; The functional localization parameters are generated from the identified three-dimensional anatomical atlas and associated with the target area through a spatial distance metric.

8. The bone tumor deterioration area resection range decision support system according to claim 7, characterized in that The pathological risk parameters are derived from a comprehensive evaluation of imaging data and preoperative puncture biopsy results, and a control mapping can be established with a previous case database.

9. The bone tumor deterioration area resection range decision support system according to claim 1, characterized in that The method for reconstructing the entire resection volume comprises: Constructing multiple two-dimensional spatial cross-sectional clusters around the tumor deterioration area, each cross-sectional cluster representing the anatomical structure and lesion distribution characteristics of the target area in a specific direction; associating each cross-sectional cluster with a functional factor or risk factor, including but not limited to the load-bearing function of local tissue, joint proximity, or neurovascular proximity; and risk factors including the biological activity of the lesion or the probability of pathological spread; Receive the resection target parameters input by the user, and the resection target parameters are used to set the relative weights of the function retention priority and the lesion clearance priority; based on the resection target parameters, dynamically adjust the weight scores of the factors associated with each section cluster, and update the resection priority of each section cluster.

10. The bone tumor deterioration area resection range decision support system according to claim 9, characterized in that The two-dimensional section cluster is generated with the tumor main axis or anatomical structure axis as the reference direction, covering the entire lesion deterioration area and the edge area; the functional factors and risk factors are encoded in numerical form as section attribute vectors and used to calculate the resection feasibility score of each section cluster; the three-dimensional resection boundary volume is generated by interpolating, fusing and spatially fitting the boundary positions of all section clusters.

Images