A method, system and storage medium for predicting the risk of subchondral bone filling material overflow in knee arthroplasty by fusing MRI lesion localization and intraoperative fluoroscopy sequences

Abstract

This invention discloses a method, system, and storage medium for predicting the risk of subchondral bone filler spillage in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences, belonging to the field of medical image processing technology. The method acquires MRI images of the knee joint, constructs an anisotropic permeation tensor field, and extracts the main permeation axis vector and the extreme value band of the permeation gradient; acquires intraoperative fluoroscopic sequences, performs dimensionality reduction mapping using the bone anatomical contour as the registration anchor point, and generates a fluoroscopic background baseline image; identifies the rheological front contour of the filler material in the fluoroscopic sequence, calculates the local velocity vector and morphological curvature to locate the jet intrusion node; maps this node to the baseline image, calculates the included angle and spatial approximation distance, and constructs a temporal risk feature matrix, which is input into a prediction network to predict the probability of material spillage and arrival time. The system and medium are used to implement the above method.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method, system, and storage medium for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences. Background Technology

[0002] Subchondral bone remodeling of the knee is a minimally invasive surgical procedure for treating knee osteoarthritis and subchondral bone marrow lesions. This surgery primarily involves injecting bone cement or other filler materials into the lesion area under pressure to enhance the mechanical support of the subchondral bone and alleviate joint pain. However, due to the complex microscopic anatomy of the knee subchondral bone, containing numerous interwoven trabeculae and venous plexuses, the filler material is prone to leakage into the joint cavity, surrounding soft tissues, and blood vessels during pressurized injection, potentially leading to severe postoperative pain, joint stiffness, and even requiring secondary debridement. Therefore, precise monitoring and proactive mitigation of the risk of filler material leakage during minimally invasive surgery are crucial for improving the safety of this procedure.

[0003] Existing extravasation monitoring methods typically rely on preoperative static MRI image planning or intraoperative C-arm fluoroscopy. Due to the spatial anatomical separation between preoperative images and intraoperative patient positioning and operative perspective, a single two-dimensional fluoroscopic image can only present the planar projection outline of high-density materials, masking the spatial information of three-dimensional depth within the lesion and local bone density differences. In this operational mode, surgeons can only visually detect abnormalities when substantial leakage of the filling material has occurred and become visible beyond the boundaries. This post-operative intervention method results in a significant time delay in the system's safety response. Furthermore, existing methods, relying on human experience, cannot quantitatively assess or predict trends regarding the abrupt changes in flow rate and spike-like spread of non-Newtonian fluid filling materials in the complex intraosseous environment. This easily leads to significant differences in surgical safety limits among different operators, failing to meet the clinical need for proactive interception of filling material leakage risks. Therefore, it is still necessary to provide a method for predicting the risk of extravasation of subchondral bone filling materials in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences to address the problems of dimensionality reduction in intraoperative spatial perception and the lag in rheological risk warning. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method, system, and storage medium for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences, comprising the following steps: S1, acquiring multi-sequence MRI images of the knee joint, extracting the subchondral bone lesion area and the cortical bone perimeter; analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion area, constructing an anisotropic permeation tensor field, and extracting the principal permeation axis vector and permeation gradient extreme value band within the anisotropic permeation tensor field; S2, acquiring continuous intraoperative fluoroscopic sequences of the knee joint, using the bone anatomical contour within the field of view as spatial registration anchor points, dimensionally reducing and mapping the subchondral bone lesion area, cortical bone perimeter, and anisotropic permeation tensor field to a two-dimensional fluoroscopic plane, generating a fluoroscopic background reference map superimposed with the principal permeation axis vector and permeation gradient extreme value band; S3, identifying... In a continuous fluoroscopic sequence, the rheological front profile of the subchondral bone filling material is analyzed. The local propagation velocity vector and front morphological curvature along the normal direction of the rheological front profile are calculated. Non-Newtonian rheological characterization parameters are constructed by combining the local propagation velocity vector and the front morphological curvature to locate the jet intrusion node on the rheological front profile. In step S4, the jet intrusion node is mapped onto the fluoroscopic background reference image, and the angle between the local propagation velocity vector of the jet intrusion node and the main permeation axis vector is calculated. The spatial approximation distance from the jet intrusion node to the periphery of the bone cortex and the extreme value zone of the permeation gradient is measured simultaneously. A temporal risk feature matrix is ​​constructed by combining the angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance. The temporal risk feature matrix is ​​input into a temporal prediction network to predict the probability of subchondral bone filling material overflowing the periphery of the bone cortex and the overflow arrival time.

[0006] In a preferred embodiment, the specific process of acquiring multi-sequence MRI images of the knee joint and extracting the subchondral bone lesion region and the cortical bone periphery is as follows: acquiring proton density-weighted images and fat-suppressed sequence images of the knee joint, performing spatial registration and fusion to generate a multimodal contrast voxel matrix; extracting the edema signal attenuation edge in the multimodal contrast voxel matrix, stripping the normal bone marrow fat signal region, and closing and defining the subchondral bone lesion region; tracing the signal-free black band region outside the multimodal contrast voxel matrix, extracting the continuity features of the inner and outer physical boundaries of the signal-free black band region, and smoothly fitting to generate a three-dimensional closed cortical bone periphery.

[0007] In a preferred embodiment, the specific process of analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion region, constructing an anisotropic permeability tensor field, and extracting the principal permeability axis vector and permeability gradient extremum band within the anisotropic permeability tensor field is as follows: traversing the three-dimensional voxel blocks within the subchondral bone lesion region, extracting the three-dimensional gray-level co-occurrence matrix of the three-dimensional voxel blocks, and constructing a spatial structure tensor based on the local directional gradient of the three-dimensional gray-level co-occurrence matrix; performing eigenvalue decomposition on the spatial structure tensor, extracting the eigenvector corresponding to the largest eigenvalue as the spatial permeability dominance direction, and mapping the spatial permeability dominance direction back to the subchondral bone lesion region to generate an anisotropic permeability tensor field; calculating the spatial derivative of the anisotropic index within the anisotropic permeability tensor field, extracting the three-dimensional connected regions where the spatial derivative reaches a local maximum and marking them as permeability gradient extremum bands; extracting the unitized direction vector along the spatial permeability dominance direction within the permeability gradient extremum band, and defining it as the principal permeability axis vector.

[0008] In a preferred embodiment, the specific process of acquiring a continuous intraoperative fluoroscopic sequence of the knee joint, using the bone anatomical contour within the field of view as the spatial registration anchor point, and dimensionally reducing and mapping the subchondral bone lesion region, the cortical bone perimeter, and the anisotropic permeation tensor field to a two-dimensional fluoroscopic plane to generate a fluoroscopic background reference map superimposed with the principal permeation axis vector and the permeation gradient extreme value band is as follows: Anatomical edge contours of the femoral medial and lateral condyles and the tibial plateau within the field of view of the continuous fluoroscopic sequence are extracted using an edge detection operator to construct the bone anatomical contour in a two-dimensional coordinate system; a fluoroscopic projection matrix of a virtual camera is established based on the geometric parameters of intraoperative fluoroscopic imaging, and the cortical bone perimeter is projected onto the projection plane using the fluoroscopic projection matrix to generate a two-dimensional virtual projection contour; iterative nearest point spatial alignment of the bone anatomical contour and the two-dimensional virtual projection contour is performed, and the optimal pose transformation matrix is ​​output; based on the optimal pose transformation matrix, the subchondral bone lesion region and the anisotropic permeation tensor field are simultaneously dimensionally reduced and mapped to a two-dimensional fluoroscopic plane, and two-dimensional mapping weights are assigned to the principal permeation axis vector and the permeation gradient extreme value band according to the fluoroscopic projection depth attenuation function, and the fluoroscopic background reference map is fused and output.

[0009] In a preferred embodiment, the specific process of identifying the rheological front profile of the subchondral bone filling material in a continuous perspective sequence and calculating the local propagation velocity vector and front morphological curvature of the rheological front profile along the normal direction is as follows: High-density differential pixel sets of adjacent perspective frames in the continuous perspective sequence are extracted, and the active contour model is driven to perform closed convergence evolution along the outer edge of the high-density differential pixel sets to output the rheological front profile; boundary nodes are discretized on the rheological front profile, and the outward-pointing normal direction field of the boundary nodes is extracted; the dense optical flow field of adjacent perspective frames in the continuous perspective sequence is calculated, and the dense optical flow field is projected onto the normal direction field to generate the local propagation velocity vector of each boundary node on the rheological front profile; the discrete spatial coordinates of the rheological front profile are extracted, and a second spatial differential is performed on the discrete spatial coordinates to extract the spatial boundary bending features on the normal direction field to generate the front morphological curvature.

[0010] In a preferred embodiment, the specific process of constructing non-Newtonian rheological characterization parameters by combining the local spread velocity vector and the front morphology curvature, and locating the jet intrusion node on the rheological front profile is as follows: extract the modulus parameter of the local spread velocity vector, and extract the peak parameter of the front morphology curvature; fuse the modulus parameter and the peak parameter using tensor product features to generate non-Newtonian rheological characterization parameters that characterize shear thinning and viscous fingering rheological properties; calculate the spatial distribution gradient of the non-Newtonian rheological characterization parameters along the rheological front profile; extract the spatial extreme coordinates where the spatial distribution gradient undergoes step distortion, and lock the spatial extreme coordinates as the jet intrusion node.

[0011] In a preferred embodiment, the specific process of mapping the jet intrusion node onto the fluoroscopic background reference image, calculating the angle between the local propagation velocity vector and the main permeation axis vector of the jet intrusion node, and simultaneously measuring the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient is as follows: extract the two-dimensional spatial pixel coordinates of the jet intrusion node, and register and map the two-dimensional spatial pixel coordinates into the two-dimensional coordinate system of the fluoroscopic background reference image; extract the main permeation axis vector that coincides with the two-dimensional spatial pixel coordinates in the two-dimensional coordinate system, perform the inverse cosine operation of the vector inner product of the local propagation velocity vector and the main permeation axis vector, and generate the angle analytically; extract the shortest Euclidean distance vector from the two-dimensional spatial pixel coordinates to the cortical periphery, and extract the shortest Euclidean distance vector from the two-dimensional spatial pixel coordinates to the two-dimensional mapped boundary of the extreme value zone of the permeation gradient; extract the magnitude of the shortest Euclidean distance vector respectively, and generate the spatial approximation distance.

[0012] In a preferred embodiment, the process of constructing a temporal risk feature matrix by combining the included angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance, and inputting the temporal risk feature matrix into a temporal prediction network to predict the probability of spillover of subchondral bone filling material across the cortical periphery and the spillover arrival time is as follows: Following the timestamp dimension of the continuous perspective sequence, the included angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance extracted in each iteration are sequentially spliced ​​to construct a temporal risk feature matrix containing latent temporal dimension features; the temporal risk feature matrix is ​​input into a temporal prediction network equipped with temporal convolutional kernels and gated memory units; the temporal convolutional kernels are called to extract the dynamic jump features of the temporal risk feature matrix at adjacent timestamps, and the gated memory units are called to establish a rheological evolution trend latent state across timestamps; the dynamic jump features and the rheological evolution trend latent state are fused and input into a nonlinear regression decoding module to output the spillover probability, and the spillover arrival time is extrapolated along the gradient descent direction of the rheological evolution trend latent state.

[0013] A system for predicting the risk of subchondral bone filler spillage in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences is used to execute the aforementioned method for predicting the risk of subchondral bone filler spillage in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences. The system includes: a tensor construction module for acquiring multi-sequence MRI images of the knee joint, extracting the subchondral bone lesion region and the cortical bone perimeter; analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion region, constructing an anisotropic permeability tensor field, and extracting the principal permeability axis vector and permeability gradient extreme value band within the anisotropic permeability tensor field; and a dimension reduction mapping module for acquiring continuous intraoperative fluoroscopic sequences of the knee joint, using the bone anatomical contours within the field of view as spatial registration anchor points, and dimensionally mapping the subchondral bone lesion region, cortical bone perimeter, and anisotropic permeability tensor field to a two-dimensional fluoroscopic plane, generating a fluoroscopic back view superimposed with the principal permeability axis vector and permeability gradient extreme value band. The system consists of a baseline image and a rheological identification module. The first module identifies the rheological front profile of subchondral bone filler in a continuous fluoroscopic sequence, calculating the local propagation velocity vector and front curvature along the normal direction. The second module combines the local propagation velocity vector and front curvature to construct non-Newtonian rheological parameters, locating jet intrusion nodes on the rheological front profile. The third module maps the jet intrusion nodes onto the fluoroscopic background baseline image, calculates the angle between the local propagation velocity vector and the main permeation axis vector, and simultaneously measures the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient. The fourth module combines the angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance to construct a temporal risk feature matrix. This matrix is ​​then input into a temporal prediction network to predict the probability of subchondral bone filler overflowing the cortical periphery and the overflow arrival time.

[0014] A computer-readable storage medium storing a computer program, the computer program being executed by a processor, is described above as a method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by fusing MRI lesion localization and intraoperative fluoroscopic sequences.

[0015] The technical effects and advantages of this invention are as follows:

[0016] (1) A method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequence. In the preoperative stage, the anatomical boundary of the subchondral bone lesion in the MRI image is extracted, and the spatial texture distribution of the trabecular bone inside the lesion is analyzed to construct an anisotropic permeation tensor field, and the main permeation axis vector and permeation gradient extreme value zone inside the bone are located. Then, in the intraoperative stage, the lesion area containing the three-dimensional resistance distribution characteristics and the tensor field are dimensionally reduced and mapped to the two-dimensional fluoroscopic plane using the bone anatomical contour as the spatial registration anchor point, generating a fluoroscopic background reference map. Thus, this invention realizes the coordinate system of preoperative three-dimensional boundary prior and intraoperative two-dimensional fluoroscopic operation space. The dimensionality reduction mapping supplements the missing anatomical depth information and bone density gradient in the two-dimensional X-ray fluoroscopic image, providing a spatial reference map with real physical constraint boundaries for subsequent fluid tracking.

[0017] (2) A knee subchondral bone filling material spillage risk prediction system integrating MRI lesion localization and intraoperative fluoroscopic sequence dynamically identifies the rheological front contour of the filling material in the fluoroscopic sequence during injection. It locates the jet intrusion node on the rheological front contour by combining the local spread velocity vector and the front morphology curvature. Furthermore, it maps this node to the fluoroscopic background reference image and simultaneously calculates the angle between the spread velocity vector and the main penetration axis vector and the spatial approximation distance. Based on this, a temporal risk feature matrix is ​​constructed and input into a temporal prediction network to calculate the spillage probability and spillage arrival time. Thus, this invention quantitatively characterizes the non-Newtonian shear thinning properties of the filling material in the micropores of the trabecular bone, extracts the flow velocity mutation and spread trend, and outputs the quantitative threshold of the spillage probability and remaining time before the material substantially breaks through the physical boundary of the bone cortex. This provides objective data support for controlling the injection device to perform hard interruption.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method for predicting the risk of extravasation of subchondral bone filling material in the knee joint, which integrates MRI lesion localization and intraoperative fluoroscopic sequence according to the present invention.

[0020] Figure 2 This is a schematic diagram of the three-dimensional spatial distribution of subchondral bone lesions and permeability tensor field in an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the time-series prediction evolution of the risk of overflow of filling material in an embodiment of the present invention.

[0022] Figure 4 This is a flowchart of the knee joint subchondral bone filling material spillage risk prediction system that integrates MRI lesion localization and intraoperative fluoroscopic sequence according to the present invention. Detailed Implementation

[0023] This application's embodiments address the problems of reference loss due to dimensional reduction of three-dimensional anatomical space perception and delayed overflow warnings caused by relying on manual observation of two-dimensional fluoroscopic images in existing monitoring schemes by integrating MRI lesion localization and intraoperative fluoroscopic sequences for predicting the risk of overflow of subchondral bone filling material in the knee joint.

[0024] The overall approach of the scheme in this application embodiment is as follows: Acquire multi-sequence MRI images of the knee joint; extract the subchondral bone lesion region and the cortical bone perimeter; analyze the spatial texture distribution of trabeculae within the subchondral bone lesion region to construct an anisotropic permeation tensor field; extract the principal permeation axis vector and the permeation gradient extreme value band within the field; acquire continuous fluoroscopic sequences of the knee joint during surgery; using the bone anatomical contour within the field of view as spatial registration anchor points, reduce the dimension of the subchondral bone lesion region, the cortical bone perimeter, and the anisotropic permeation tensor field to a two-dimensional fluoroscopic plane, generating a fluoroscopic background reference map superimposed with the principal permeation axis vector and the permeation gradient extreme value band; identify the subchondral bone filling material in the continuous fluoroscopic sequence. The rheological front profile is obtained, and the local propagation velocity vector and front morphological curvature along the normal direction are calculated. The local propagation velocity vector and front morphological curvature are combined to construct non-Newtonian rheological characterization parameters to locate the jet intrusion node. The jet intrusion node is mapped onto the perspective background reference map, and the angle between the local propagation velocity vector and the main permeation axis vector is calculated. The spatial approximation distance to the bone cortex perimeter and the permeation gradient extreme zone is measured. The temporal risk feature matrix is ​​constructed by combining the angle, the magnitude of the local propagation velocity vector and the spatial approximation distance. The temporal risk feature matrix is ​​input into the temporal prediction network to predict the probability of subchondral bone filling material overflowing the bone cortex perimeter and the overflow arrival time.

[0025] Example 1; please refer to Figure 1This invention provides a technical solution: a method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences, comprising the following steps: S1, acquiring multi-sequence MRI images of the knee joint, extracting the subchondral bone lesion area and the cortical bone perimeter; analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion area, constructing an anisotropic permeation tensor field, and extracting the principal permeation axis vector and permeation gradient extreme value band within the anisotropic permeation tensor field; S2, acquiring continuous intraoperative fluoroscopic sequences of the knee joint, using the bone anatomical contour within the field of view as spatial registration anchor points, dimensionally reducing and mapping the subchondral bone lesion area, cortical bone perimeter, and anisotropic permeation tensor field to a two-dimensional fluoroscopic plane, generating a fluoroscopic background reference map superimposed with the principal permeation axis vector and permeation gradient extreme value band; S3, identifying continuous... In the fluoroscopic sequence, the rheological front profile of the subchondral bone filling material is analyzed, and the local propagation velocity vector and front morphological curvature along the normal direction of the rheological front profile are calculated. Non-Newtonian rheological characterization parameters are constructed by combining the local propagation velocity vector and the front morphological curvature to locate the jet intrusion node on the rheological front profile. S4. The jet intrusion node is mapped onto the fluoroscopic background reference image, and the angle between the local propagation velocity vector of the jet intrusion node and the main permeation axis vector is calculated. Simultaneously, the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient is measured. A temporal risk feature matrix is ​​constructed by combining the angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance. This temporal risk feature matrix is ​​input into a temporal prediction network to predict the probability of subchondral bone filling material overflowing the cortical periphery and the overflow arrival time.

[0026] In this implementation scheme, step S1 is used to establish a physical resistance distribution model within the subchondral bone lesion using MRI images during the preoperative stage. By extracting the spatial texture of the trabeculae, the system can quantify the porosity of different regions within the bone. The anisotropic permeability tensor field is a spatial mathematical expression model of the differences in fluid permeability resistance in different directions due to the irregular arrangement of the trabeculae. The principal permeability axis vector represents the specific direction in which the fluid experiences the least resistance and is most likely to diffuse within this tensor field. The permeability gradient extreme zone refers to the anatomical zone where the resistance changes drastically at the boundary between the osteoporotic and dense regions; it is usually an anatomically vulnerable area where the filling material is most prone to rupture and leakage under pressure.

[0027] Step S2 establishes a spatial reference relationship between the preoperatively acquired three-dimensional resistance distribution model and the intraoperative real-time two-dimensional images. After acquiring the intraoperative fluoroscopic images, the system uses clearly visible bony landmarks in the image as spatial registration anchor points, that is, using the edges of fixed anatomical structures such as the femoral condyle or tibial plateau as the reference positions for coordinate system alignment. Dimensionality reduction mapping refers to the calculation process of transforming the three-dimensional lesion area and tensor field parameters into two-dimensional planar data by matrix projection transformation based on the projection angle of the intraoperative X-ray emission source. The fluoroscopic background reference map is a comprehensive two-dimensional reference base map after dimensionality reduction mapping, superimposed with invisible internal bone weakness channels and real-time bone contours, providing spatial environmental constraints for subsequent fluid tracking.

[0028] Step S3 is used to capture the microscopic physical diffusion morphology of the fluid during the injection of filler materials such as bone cement. The system identifies the material edges in consecutive frame images and calculates their local propagation velocity vectors, i.e., the instantaneous diffusion velocity and direction of each discrete point on the edge in the normal direction. Non-Newtonian rheological characterization parameters refer to the numerical values ​​that combine the diffusion velocity of the fluid with the curvature of the edge morphology to quantify the shear thinning physical properties of non-Newtonian fluids such as bone cement when they pass through narrow bone microfissures under pressure. Jet intrusion nodes refer to local points at the fluid front that are spike-shaped and exhibit abnormally high flow velocities. The appearance of such nodes indicates that the material has entered a local low-resistance channel and has a strong tendency to penetrate and destroy.

[0029] Step S4 is used to calculate the probability and timing of spillover events by combining the anatomical weak space with the dynamic evolution trend of the fluid. After placing the jet intrusion node into the fluoroscopic background reference image, the system calculates the angle between the fluid intrusion direction and the main permeation axis vector, and calculates the spatial approximation distance, i.e., the remaining physical space length for the current fluid front to reach the nearest physical boundary of the cortical bone or the region of drastic change in permeability resistance. The temporal risk feature matrix is ​​a high-dimensional data set composed of parameters such as the angle, velocity modulus, and remaining distance at consecutive time nodes according to the time axis sequence of the fluoroscopic image. The temporal prediction network extracts the jump and evolution patterns of each parameter in this matrix over time to predict the probability and remaining seconds of the filling material eventually piercing the cortical bone and causing substantial leakage under the current spread trend.

[0030] Specifically, the process of acquiring multi-sequence MRI images of the knee joint and extracting the subchondral bone lesion region and the cortical bone perimeter is as follows: acquire proton density-weighted images and fat-suppressed sequence images of the knee joint, perform spatial registration and fusion to generate a multimodal contrast voxel matrix; extract the edema signal attenuation edge in the multimodal contrast voxel matrix, strip the normal bone marrow fat signal region, and close and define the subchondral bone lesion region; track the signal-free black band region outside the multimodal contrast voxel matrix, extract the continuity features of the inner and outer physical boundaries of the signal-free black band region, and smoothly fit to generate a three-dimensional closed cortical bone perimeter.

[0031] In this implementation scheme, for the registration and tissue edge resolution of multi-sequence images of the knee joint, the proton density-weighted image and the fat-suppressed sequence image acquired in the same spatial coordinate system are first aligned at the voxel level. By extracting the variance of the local structural information of the two, the fusion ratio coefficient of the multimodal contrast voxel matrix is ​​dynamically calculated. Its calculation formula is ;in, This represents the variance of voxel gray levels in a proton density-weighted image within a local sliding window; This represents the variance of voxel gray levels in fat-suppressed sequence images within the same scale window; This represents a minimal positive real constant to prevent the denominator from being zero. Based on the weighted fusion of the multimodal contrast voxel matrix, the three-dimensional gray-level gradient magnitude of each voxel node is calculated. When the gradient magnitude of a continuous voxel is greater than the sum of the average gradient magnitude and the standard deviation of the gradient in its local neighborhood, it is extracted as an edge of edema signal attenuation. For the stripping of normal bone marrow fat, the system adaptively establishes a signal suppression threshold based on the non-tissue background region of the image. , ;in, This represents the average gray value of voxels in the non-detection area of ​​the image background; This represents the grayscale standard deviation of voxels in the non-detection area of ​​the background. Highlighted fat accumulation areas in the voxel matrix are removed using this suppression threshold, and a three-dimensional morphological closing operation is performed along the edge of edema signal attenuation to define the subchondral bone lesion region of the solid. Further, voxels with grayscale values ​​lower than [a certain value] in the multimodal contrast voxel matrix are extracted. The outermost spatially connected domain is treated as a signal-free black band. Discrete spatial point clouds of the physical boundaries inside and outside this connected domain are captured, and the energy functional is minimized. Iterative fitting of the cortical bone perimeter; where, This represents the total surface energy during the cortical bone fitting process; This represents the three-dimensional cortical perimeter surface function to be solved; Represents the first-order spatial gradient vector of the surface function; This represents the local mean curvature parameter of the surface; The tensile force weighting constant represents the force that controls the surface area shrinkage characteristics of the curved surface; Represents the rigid bending weight constant that controls the curvature smoothness of the curved surface. The integral infinitesimal element representing the surface area is used to construct a three-dimensional closed cortical periphery without breaks.

[0032] Specifically, the process of analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion region, constructing an anisotropic permeability tensor field, and extracting the principal permeability axis vector and permeability gradient extremum band within the anisotropic permeability tensor field is as follows: Traversing the three-dimensional voxel blocks within the subchondral bone lesion region, extracting the three-dimensional gray-level co-occurrence matrix of the three-dimensional voxel blocks, and constructing a spatial structure tensor based on the local directional gradient of the three-dimensional gray-level co-occurrence matrix; performing eigenvalue decomposition on the spatial structure tensor, extracting the eigenvector corresponding to the largest eigenvalue as the spatial permeability dominance direction, and mapping the spatial permeability dominance direction back to the subchondral bone lesion region to generate the anisotropic permeability tensor field; calculating the spatial derivative of the anisotropic index within the anisotropic permeability tensor field, extracting the three-dimensional connected regions where the spatial derivative reaches a local maximum and marking them as permeability gradient extremum bands; extracting the unitized direction vector along the spatial permeability dominance direction within the permeability gradient extremum band, defining it as the principal permeability axis vector.

[0033] In this implementation scheme, during the process of mapping trabecular bone texture to a physical resistance distribution model, all voxel blocks within the subchondral bone lesion area are traversed using a fixed-size cubic window. The three-dimensional gray-level co-occurrence matrix under each directional bias is extracted, and the contrast parameter characterizing the coarseness of the trabecular bone texture is calculated. .in accordance with Establish local directional gradient vectors along the spatial rate of change in three orthogonal dimensions. Where x represents the horizontal spatial coordinate axis; y represents the vertical spatial coordinate axis; and z represents the depth spatial coordinate axis; subsequently, the spatial autocorrelation matrix of the gradient vector is integrated and smoothed within a local Gaussian window to generate the spatial structure tensor. ;in, A third-order symmetric matrix representing the anisotropy of the trabecular texture direction; Let represent the Gaussian smoothing kernel function in 3D space; u represents the horizontal integral infinitesimal coordinates within the voxel block; v represents the vertical integral infinitesimal coordinates; and w represents the depth integral infinitesimal coordinates. For the spatial structure tensor... Solving the characteristic polynomial involves performing eigenvalue decomposition, and the three non-negative eigenvalues ​​obtained are denoted as follows: And the numerical arrangement satisfies , the largest eigenvalue The corresponding normalized feature vector The direction of spatial permeability is recorded. To quantify the directional cohesion strength of fluid permeation in micropores, the anisotropy index is calculated. ;in, Represents the scalarized texture heterogeneity index; This represents the arithmetic mean of three eigenvalues. Represents the minimum stability constant for preventing computational overflow; for Solving for the derivative in three-dimensional space When a certain node The transition exceeds the mean within its 3D bounding box, and the node is in When the second partial derivative along the ray direction is zero, it is identified as a local maximum point. The shell region formed by the topological connectivity of all local maxima points in three-dimensional space is marked as the permeability gradient extremum zone. Finally, extract the portion located at... The internal spatial penetration advantage direction is divided by its own Euclidean norm to eliminate the difference in modulus dimensions, generating a normalized direction vector. This is defined as the main permeation axis vector. Please refer to the relevant documentation. Figure 2 , Figure 2 The figure illustrates the three-dimensional spatial distribution characteristics of the anisotropic permeability tensor field and anatomical boundary constructed in step S1. In the figure, the outer gray discrete dot matrix maps to the cortical bone perimeter generated by fitting a multimodal contrast voxel matrix, defining a safe external physical injection boundary. The dark blue high-density dot cloud clustered within the dot matrix represents the solid subchondral bone lesion region extracted after stripping normal bone marrow fat signals. The red clustered dot cloud located outside the lesion region corresponds to the permeability gradient extremum band formed by local maxima of the spatial derivative of the anisotropic index, indicating a physically prone leakage area where local micropore resistance changes drastically. The orange directed line segment connecting the blue lesion region and the red extremum band strictly maps to the principal permeability axis vector extracted by performing eigenvalue decomposition on the spatial structure tensor. This vector trajectory reveals the dominant spatial permeation direction in which non-Newtonian fluids are most likely to finger-spread within the trabecular bone network under pressure.

[0034] Specifically, the process of acquiring a continuous intraoperative fluoroscopic sequence of the knee joint, using the bone anatomical contours within the field of view as spatial registration anchor points, and dimensionally reducing and mapping the subchondral bone lesion region, the cortical bone perimeter, and the anisotropic permeation tensor field to a two-dimensional fluoroscopic plane to generate a fluoroscopic background reference map superimposed with the principal permeation axis vector and the permeation gradient extremum band is as follows: Anatomical edge contours of the femoral medial and lateral condyles and the tibial plateau within the field of view of the continuous fluoroscopic sequence are extracted using an edge detection operator to construct the bone anatomical contours in a two-dimensional coordinate system; a fluoroscopic projection matrix of the virtual camera is established based on the intraoperative fluoroscopic imaging geometric parameters, and the cortical bone perimeter is projected onto the projection plane using the fluoroscopic projection matrix to generate a two-dimensional virtual projection contour; iterative nearest-point spatial alignment of the bone anatomical contours and the two-dimensional virtual projection contours is performed, and the optimal pose transformation matrix is ​​output; based on the optimal pose transformation matrix, the subchondral bone lesion region and the anisotropic permeation tensor field are simultaneously dimensionally reduced and mapped to a two-dimensional fluoroscopic plane, and two-dimensional mapping weights are assigned to the principal permeation axis vector and the permeation gradient extremum band according to the fluoroscopic projection depth attenuation function, and the fluoroscopic background reference map is fused and output.

[0035] In this implementation scheme, for the spatial alignment and dimensionality reduction mapping process of intraoperative cross-modal images, firstly, adaptive Gaussian-Laplacian operator filtering is performed on the acquired two-dimensional continuous fluoroscopic sequence to extract edge pixels of the femoral medial and lateral condyles and the tibial plateau with high contrast gradients. Then, spline curve interpolation is used to fit discrete edge pixels to construct a continuous and smooth two-dimensional coordinate system bone anatomical contour. Based on the distance from the actual acquired intraoperative fluoroscopic imaging source to the detector and the physical size of the detector array pixels, the fluoroscopic projection matrix of the virtual camera is constructed, and a three-dimensional to two-dimensional coordinate geometric projection calculation is performed. ;in, This represents the two-dimensional virtual projection coordinate vector generated by the projection of the cortical bone periphery; This represents the camera intrinsic parameter matrix that reflects the focal length of the perspective device and the spatial offset of the principal point; This represents the initial rotation matrix set based on the preoperative patient positioning prediction; Represents the initial translation vector associated with the initial rotation; This represents the three-dimensional spatial coordinate vector of the cortical bone periphery in an absolute reference frame. After generating a preliminary two-dimensional virtual projection contour, the system calls an iterative nearest-point alignment algorithm to calculate the spatial displacement residual between the virtual contour and the actual anatomical contour. The optimal pose transformation matrix is ​​then solved by constructing and minimizing the error objective functional. Its solution expression is: ;in, This represents the alignment and rotation block matrix for the current step in the iterative loop process; This represents the alignment and translation block vector for the current step in the iterative loop process; This represents the total number of matching discrete point pairs established by searching on the two contour boundaries; k represents the iteration index sequence number of the matching point pair. This represents the coordinate vector of the k-th sampling point on the two-dimensional virtual projection contour; This represents the coordinate vector of the nearest neighbor target point determined by a traversal search based on Euclidean distance on a two-dimensional bone anatomical contour. After obtaining the accurate planar mapping relationship, relying on... The system performs coordinate transformation and parallel dimensionality reduction on the subchondral bone lesion area and anisotropic permeation tensor field, tiling it onto a two-dimensional perspective plane. Due to the overlapping of anatomical structures at different depths in perspective imaging, signal compression and attenuation occur on the plane. The system calculates the two-dimensional mapping weights of the feature vectors within the field based on the perspective projection depth attenuation function. The formula for calculating the mapping weight is: ;in, This indicates the display damping and rheological calculation weights assigned to the main permeation axial vector and the extreme value zone of the permeation gradient on the perspective plane; This represents the absolute spatial depth coordinates of an anisotropic permeation tensor field element in the virtual perspective camera reference frame; The reference depth constant represents the physical focal plane of the actual intraoperative fluoroscopic imaging. The depth attenuation adjustment coefficient, which is strongly correlated with the emission voltage of the X-ray tube of the fluoroscopic equipment, is used to compress the three-dimensional spatial mechanical features into a fluoroscopic background reference map carrying the anatomical depth resistance imprint after numerical modulation of the depth attenuation function.

[0036] Specifically, the process of identifying the rheological front profile of subchondral bone filling material in a continuous perspective sequence and calculating the local propagation velocity vector and front morphological curvature of the rheological front profile along the normal direction is as follows: High-density differential pixel sets are extracted from adjacent perspective frames in the continuous perspective sequence; the active profile model is driven to perform closed convergence evolution along the outer edge of the high-density differential pixel sets to output the rheological front profile; boundary nodes are discretized on the rheological front profile, and the outward-pointing normal direction field of the boundary nodes is extracted; the dense optical flow field of adjacent perspective frames in the continuous perspective sequence is calculated, and the dense optical flow field is projected onto the normal direction field to generate the local propagation velocity vector of each boundary node on the rheological front profile; the discrete spatial coordinates of the rheological front profile are extracted, and a second spatial derivative is performed on the discrete spatial coordinates to extract the spatial boundary bending features on the normal direction field to generate the front morphological curvature.

[0037] In this implementation scheme, for the quantitative extraction of the rheological front features of the filling material, pixel-level gray-level difference operations are performed on two adjacent frames in a continuous perspective sequence. Pixel connected regions with difference gray-level values ​​higher than the extreme value of background noise are extracted and defined as a high-density difference pixel set. The outer edge pixels of this pixel set are used as the initial evolution boundary to drive an active contour model based on region energy to extrapolate outwards until it converges to fit the outermost actual boundary of the current material, forming the rheological front contour. This rheological front contour is then uniformly discretized at fixed arc length intervals to obtain a series of boundary nodes. The closed sequence is obtained, and the outward normal direction field at each node is calculated based on the coordinates of adjacent nodes. Where q represents the discrete permutation index of the boundary nodes. Subsequently, the dense optical flow field of global pixels in two adjacent perspective images is calculated, and a functional equation is established with the objective of minimizing gray-level residuals and smoothing constraints. ;in, This represents the total energy functional for the optical flow field estimation; This represents the spatial grayscale gradient difference between two perspective images; This represents the velocity component of the dense optical flow field along the horizontal probe axis; This represents the velocity component of the dense optical flow field in the direction perpendicular to the probe axis; The time-domain first-order partial derivative of the grayscale sequence of a perspective image; This represents the smoothing constraint coefficient of the optical flow field. Its value is determined by calculating the reciprocal of the mean of the gray-level variance of the local region, and is used to balance the sharpness and coherence of the motion boundary. and Let represent the two integral independent variables of the perspective plane. Solving for the functional The minimized velocity components are projected onto the normal vectors of the corresponding nodes to generate locally propagating velocity vectors. Based on this, the discrete spatial coordinates of the rheological front profile in the planar coordinate system are extracted. The curvature of the leading edge is calculated by performing a second spatial differential using the central difference method. ;in, A curvature value that quantifies the sharpness of an outward bulge in a fluid stream; This represents the lateral coordinate projection component of the q-th boundary node; This represents a small regularization constant to prevent the denominator from being zero when differentiating along the edge of a straight line; Represents the longitudinal coordinate projection component of the q-th boundary node; and These represent the first and second order spatial difference operators along the arc length sequence of the rheological front profile, respectively.

[0038] Specifically, the process of constructing non-Newtonian rheological characterization parameters by combining the local spread velocity vector and the front morphology curvature, and locating the jet inrush node on the rheological front contour, is as follows: extract the modulus parameter of the local spread velocity vector and the peak parameter of the front morphology curvature; fuse the modulus parameter and the peak parameter using tensor product features to generate non-Newtonian rheological characterization parameters that characterize shear thinning and viscous fingering rheological properties; calculate the spatial distribution gradient of the non-Newtonian rheological characterization parameters along the rheological front contour; extract the spatial extremum coordinates where the spatial distribution gradient undergoes step distortion, and lock the spatial extremum coordinates as the jet inrush node.

[0039] In this implementation scheme, when locating rheological protrusion points with extremely high spillover risk, the local propagation velocity vectors on each boundary node are extracted. The Euclidean norm is recorded as a modulus parameter. Simultaneously, by filtering out the flat or concave portions of the leading edge curvature that are less than zero, the positive curvature value characterizing the sharp outward protrusions of the material is extracted as the peak parameter. Because bone cement-based filling materials exhibit significant non-Newtonian fluid shear thinning characteristics when flowing within porous bone trabeculae due to physical compression, manifested as a sharp increase in flow velocity accompanied by a sharpening of the fingertip morphology, the system accordingly... and Arrange the features into eigenvectors and perform matrix tensor product operations to obtain a second-order feature fusion tensor. Based on this fusion tensor, non-Newtonian rheological characterization parameters are constructed. ;in, This represents a multidimensional rheological physical quantity that comprehensively reflects the rapid velocity increase and spike growth under localized pressure. The trace operator is used to calculate the sum of the main diagonal elements of a second-order fusion tensor. The sensitivity weighting coefficients of the tensor trace are determined by extracting the two-dimensional mapping weights that coincide with the current perspective detection window space in the perspective background reference image. The mean of the region is obtained by comparing it with the mean of the global mapping weights of the baseline map; This represents the feature cross-coupling weight coefficient, and is set to... This is used to adaptively match the rheologically sensitive damping of materials in different osteoporosis regions. After obtaining the rheological parameters of each node on the contour, the discrete physical boundary index along the rheological front contour is used. Calculate the spatial distribution gradient of this characterization parameter. ;in, This indicates the steepness of the step change in parameters along a continuous edge profile. Traversing the entire rheological front profile, when a certain node position... The jump exceeds the preset step distortion threshold. Furthermore, when its second-order difference derivative on the boundary sequence is equal to zero, the spatial extremum coordinates of that node are extracted. In this calculation logic, the step distortion threshold... The specific determination method is as follows: The spatial distribution gradient values ​​of all boundary nodes of the entire closed rheological front contour within the current perspective frame are numerically sorted in ascending order. Based on the divergence distribution curve of high-density pixels within the current perspective frame, the gradient values ​​at the 95th percentile of the distribution curve are extracted as a dynamic comparison threshold to filter out background gradient noise generated by conventional flow. Finally, the spatial extreme value coordinate nodes that satisfy the above conditions are recorded as jet inrush nodes. The coordinates of these nodes represent the precise physical location where the local non-Newtonian fluid shear rate exceeds the bone porosity resistance.

[0040] Specifically, the process of mapping the jet intrusion node onto the perspective background reference image, calculating the angle between the local propagation velocity vector and the main permeation axis vector of the jet intrusion node, and simultaneously measuring the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient is as follows: Extract the two-dimensional spatial pixel coordinates of the jet intrusion node and register them into the two-dimensional coordinate system of the perspective background reference image; extract the main permeation axis vector that coincides with the two-dimensional spatial pixel coordinates in the two-dimensional coordinate system, perform the inverse cosine operation of the vector inner product of the local propagation velocity vector and the main permeation axis vector, and analytically generate the angle; extract the shortest Euclidean distance vector from the two-dimensional spatial pixel coordinates to the cortical periphery, and extract the shortest Euclidean distance vector from the two-dimensional spatial pixel coordinates to the two-dimensional mapped boundary of the extreme value zone of the permeation gradient; extract the magnitude of the shortest Euclidean distance vector respectively to generate the spatial approximation distance.

[0041] In this implementation scheme, for the geometric correlation feature calculation of the jet intrusion node in the global anatomy environment, the two-dimensional spatial pixel coordinates of the previously determined jet intrusion node in the current perspective frame are first extracted. Based on the previously calculated optimal pose transformation matrix, it is precisely projected onto the two-dimensional absolute coordinate system of the perspective background reference image to obtain the registration space coordinates. The system uses the registration spatial coordinates as the addressing index to extract the two-dimensional mapped principal permeation axis vector that coincides with its spatial coordinates from the perspective background reference map. To quantify the geometric overlap of the current bone cement rheological front protruding towards the most permeable channel, a local propagation velocity vector was executed. With two-dimensional mapping of the principal permeation axis vector The inverse cosine operation of the vector inner product is used to analytically generate the included angle. ;in, This represents the spatial angular deviation constant between the actual direction of fluid spread and the direction of the local minimum trabecular drag. This represents the static frame anti-zero division by bias constant; The Euclidean modulus of the velocity vector at the rheological front; This represents the Euclidean modulus of the axial permeation channel vector. Simultaneously, the system extracts the discrete boundary set of the cortical bone perimeter within the perspective background baseline image. Based on the Euclidean distance minimization criterion, the shortest Euclidean distance vector magnitude from the jet intrusion node to the periphery of the cortical bone is calculated, and the first spatial approximation distance is generated. ;in, This represents a quantitative distance that reflects the remaining penetration margin of the fluid from the physical boundary of the dense cortical bone. This represents the total number of coordinate nodes within the discrete boundary set of the cortical bone periphery; The discrete variable sequence number representing the coordinate node; Indicates the first The absolute two-dimensional coordinates of each discrete boundary node in the cortical bone are obtained. Similarly, the two-dimensional mapped boundary point set of the permeation gradient extremum zone is extracted, and the second spatial approximation distance from the jet intrusion node to the permeation gradient extremum zone is calculated using the same distance metric equation. ;in, This represents the physical spatial warning span of the fluid's imminent breach of the local micropore resistance abrupt change zone. Through the aforementioned feature mapping and geometric operations, the system transforms the discrete fluid dynamic transient state and static anatomically vulnerable constraints into strictly aligned spatial geometric correlation values.

[0042] Specifically, the process of constructing a temporal risk feature matrix by combining the included angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance, and then inputting the temporal risk feature matrix into a temporal prediction network to predict the probability of spillover of subchondral bone filling material breaking through the cortical periphery and the spillover arrival time is as follows: Following the timestamp dimension of the continuous perspective sequence, the extracted included angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance are sequentially spliced ​​to construct a temporal risk feature matrix containing latent temporal dimension features; the temporal risk feature matrix is ​​input into a temporal prediction network equipped with temporal convolutional kernels and gated memory units; the temporal convolutional kernels are called to extract the dynamic jump features of the temporal risk feature matrix at adjacent timestamps, and the gated memory units are called to establish the rheological evolution trend latent state across timestamps; the dynamic jump features and the rheological evolution trend latent state are fused and input into a nonlinear regression decoding module to output the spillover probability, and the spillover arrival time is extrapolated along the gradient descent direction of the rheological evolution trend latent state.

[0043] In this implementation scheme, for the data preprocessing and hardware-level protection mechanism of the time-series prediction network, the two-dimensional pixel matrix and its corresponding timestamp are extracted from the historical orthopedic surgical fluoroscopy sequences. The DICOM patient identification tag in the file header is removed, and a geometric affine transformation is performed to uniformly scale the pixel matrix to the reference resolution. At the same time, a hard-wired safety trigger threshold is set in the system's underlying microcontroller. When the first spatial approach distance is detected, a safety trigger threshold is set. When the value is less than or equal to the limit safety margin constant, the microcontroller bypasses the prediction network and directly sends an interrupt trigger signal to the power supply relay of the bone cement injection pump to perform a physical power-off operation. During the normal prediction process, the past is extracted according to the natural timestamp dimension of the continuous perspective sequence. Historical angle features, local spread velocity magnitude features, first spatial approximation distance, and second spatial approximation distance within each acquisition period are spliced ​​together in time series to generate a dimension of [missing information]. Time-series risk feature matrix The system will Load the temporal prediction network, call the temporal convolution kernel at the beginning of the network to perform sliding convolution operation along the time dimension, and extract the dynamic jump feature vector of adjacent time stamps. ;in, Indicates the first Characteristics of the rheological shear increment of the material under pressure at each time stamp; This represents the nonlinear rectification activation function; This represents the length of the receptive field dimension of a one-dimensional temporal sliding window; This indicates the first convolution operation module. Layer data purification weight matrix; This represents the bias correction vector that matches the weight matrix. Subsequently, the dynamically changing feature vector is input into a gated memory unit containing a reset and update architecture to establish the rheological evolution trend hidden state. The recursive iterative update equation of the hidden state is: ;in, This represents a comprehensive state latent variable at the current time point, encompassing the long-term pressure accumulation and instantaneous release trend within the interosseous space. An update gate variable representing the proportion of historical pressure information forgotten and the proportion of current jump information retained; Represents the normalized constant matrix; The Hadamard element-wise product operator represents the dimension. This indicates resetting the gating decision variable; This represents the state transition weight matrix that maps from the feature input to the hidden state channel; This represents the internal weight matrix indicating the transition from the hidden state of the previous time step to the current step. Ultimately, the resolved rheological evolution trend hidden state is... In the fully connected network transmitted to the nonlinear regression decoding module, the output of the normalized exponential function layer is mapped to the overflow probability in the numerical range from zero to one zero, and then along... Linear extrapolation is performed along the gradient descent differential direction on the time tensor plot to calculate the span step size of the hidden state approximating the critical surface of structural failure, which is then converted into the spillover arrival time. Please refer to [reference needed]. Figure 3 , Figure 3 The figure shows the time-series output curve of the multidimensional risk evolution after the time-series risk feature matrix is ​​input into the time-series prediction network in step S4. As shown in the figure, the horizontal axis of the graph represents the continuous perspective sequence time according to the natural timestamp dimension; the left vertical axis corresponds to the red solid line in the figure, representing the probability of spillover from the network regression output; the right vertical axis corresponds to the blue solid line in the figure, representing the arrival time of spillover calculated by linear extrapolation of the hidden state along the rheological evolution trend. In the first half of the time-series extrapolation (time nodes 1-5), the material overcomes capillary resistance uniformly in the interstitial space, the probability of spillover increases gradually and the arrival time decreases linearly in a step-like manner; when the evolution reaches the 6th perspective time series point, the spatial distribution gradient calculation of the rheological front profile triggers a step distortion, and the system marks the spatial extreme coordinates as the jet inrush node indicated by the black-framed circle in the figure. In response to this node, the red characteristic curve, representing the probability of spillover, instantly experiences a step-like nonlinear increase, with the value rapidly climbing above 0.85; simultaneously, the blue characteristic curve exhibits a step-like downward trend, indicating a nonlinear shortening of the spillover arrival time. This cross-curve evolution mechanism truly reflects the underlying control logic of the system to implement early interception and proactive risk prediction before the fluid material undergoes substantial anatomical transgression of the bone cortex.

[0044] Example 2; please refer to Figure 4A system for predicting the risk of extravasation of subchondral bone filler material in the knee joint, integrating MRI lesion localization and intraoperative fluoroscopic sequences, is used to execute the method for predicting the risk of extravasation of subchondral bone filler material in the knee joint, as described in the embodiments. The system includes: a tensor construction module for acquiring multi-sequence MRI images of the knee joint, extracting the subchondral bone lesion region and the cortical bone perimeter; analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion region, constructing an anisotropic permeability tensor field, and extracting the principal permeability axis vector and permeability gradient extreme value band within the anisotropic permeability tensor field; and a dimension reduction mapping module for acquiring continuous intraoperative fluoroscopic sequences of the knee joint, using the bone anatomical contour within the field of view as spatial registration anchor points, dimensionally reducing and mapping the subchondral bone lesion region, cortical bone perimeter, and anisotropic permeability tensor field to a two-dimensional fluoroscopic plane, generating a fluoroscopic image superimposed with the principal permeability axis vector and permeability gradient extreme value band. The system consists of a background baseline image and a rheological identification module. The rheological identification module identifies the rheological front profile of subchondral bone filler material in a continuous fluoroscopic sequence, calculates the local propagation velocity vector and front morphological curvature along the normal direction of the rheological front profile, and constructs non-Newtonian rheological characterization parameters by combining the local propagation velocity vector and front morphological curvature to locate jet intrusion nodes on the rheological front profile. A risk prediction module maps the jet intrusion nodes onto the fluoroscopic background baseline image, calculates the angle between the local propagation velocity vector of the jet intrusion node and the main permeation axis vector, and simultaneously measures the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient. A temporal risk feature matrix is ​​constructed by combining the angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance. This temporal risk feature matrix is ​​input into a temporal prediction network to predict the probability of subchondral bone filler material overflowing the cortical periphery and the overflow arrival time.

[0045] In this implementation scheme, the tensor construction module serves as the system's front-end data parsing unit, responsible for receiving and processing the low-level data of preoperative multi-sequence MRI images of the knee joint. This module contains a 3D reconstruction and matrix operation program, which schedules the microprocessor during system runtime to extract the spatial boundaries between the subchondral bone lesion region and the cortical bone periphery. Subsequently, the module reads voxel data within the lesion region, performs matrix analysis and eigenvalue decomposition calculations on the spatial texture distribution of the trabecular bone, generates a 3D anisotropic permeability tensor field, and encapsulates the obtained principal permeability axis vector and permeability gradient extremum band into a priori structure dataset carrying absolute 3D coordinates, directly writing it into system memory for subsequent process calls.

[0046] The dimensionality reduction mapping module is configured as a computational unit to perform cross-modal coordinate system transformation. This module continuously reads the continuous fluoroscopic sequence output by the intraoperative fluoroscopic equipment via the device communication interface, and runs an edge extraction algorithm to lock the bone anatomical contour within the field of view. Using this anatomical contour as the physical registration anchor point, the module retrieves the prior structure dataset generated by the tensor construction module to construct a projection transformation matrix from the three-dimensional absolute coordinate system to the current two-dimensional camera viewing angle. By performing coordinate transformation operations, the module synchronously tiles the lesion area, the perimeter of the bone cortex, and the anisotropic permeability tensor field onto the two-dimensional fluoroscopic plane, generating a fluoroscopic background reference map superimposed with direction vectors and resistance extrema bands, serving as the physical constraint base for subsequent fluid tracking.

[0047] The rheology recognition module performs real-time dynamic calculations of fluid dynamic parameters in perspective video streams. This module processes the input perspective sequence frame by frame, extracting the rheological front profile of the continuously expanding filling material through an active contour evolution model. Based on this, the module tracks the pixel displacement field and contour curvature along the normal to the front profile, calculating and extracting the local propagation velocity vector and the front morphological curvature. The module further performs tensor cross operations on these two physical parameters to construct non-Newtonian rheological characterization parameters, and in the background, it iterates through the spatial gradient distribution of these parameters on the contour, accurately extracting the pixel locations of gradient step distortion, thereby outputting the absolute coordinate positions of the jet intrusion nodes.

[0048] The risk prediction module is responsible for the system's data fusion mapping and neural network inference tasks. This module inputs the jet intrusion node data output from the rheology identification module into the perspective background baseline map, driving the geometric calculation program to simultaneously measure the geometric angle between the velocity vector and the main permeation axis vector, as well as the Euclidean approximation distance from the intrusion node to various anatomical constraint boundaries. The module then assembles the measured angle and distance data into a time-series risk feature matrix along the time axis and feeds it into a pre-loaded time-series prediction network to perform forward propagation calculations, outputting the spillover probability and remaining time.

[0049] This application provides a computer-readable storage medium, which includes non-volatile storage hardware such as ROM, RAM, or flash memory arrays, and internally stores a compiled executable computer program. When the system's central processing unit reads and runs the computer program, it will dynamically allocate running memory and computing resources to the aforementioned tensor construction, dimensionality reduction mapping, rheological identification, and risk prediction modules in sequence at the hardware bus level, thereby driving the underlying computing hardware to fully execute all data processing steps to implement the aforementioned method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences.

[0050] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0051] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0052] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0053] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0054] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0055] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences, characterized in that, Includes the following steps: S1. Obtain multi-sequence MRI images of the knee joint, extract the subchondral bone lesion area and the periphery of the bone cortex; analyze the spatial texture distribution of trabeculae within the subchondral bone lesion area, construct an anisotropic permeability tensor field, and extract the principal permeability axis vector and permeability gradient extreme value band within the anisotropic permeability tensor field. S2. Obtain the continuous fluoroscopic sequence of the knee joint during the operation. Using the bone anatomical contour in the field of view as the spatial registration anchor point, reduce the dimension of the subchondral bone lesion area, the periphery of the bone cortex and the anisotropic permeation tensor field to the two-dimensional fluoroscopic plane to generate a fluoroscopic background reference map superimposed with the main permeation axis vector and the permeation gradient extreme value zone. S3. Identify the rheological front profile of the subchondral bone filling material in a continuous fluoroscopic sequence, calculate the local propagation velocity vector and front morphological curvature of the rheological front profile along the normal direction; construct non-Newtonian rheological characterization parameters by combining the local propagation velocity vector and front morphological curvature, and locate the jet intrusion node on the rheological front profile. S4. Map the jet intrusion node onto the perspective background reference image, calculate the angle between the local propagation velocity vector of the jet intrusion node and the main permeation axis vector, and simultaneously measure the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient; combine the angle, the magnitude of the local propagation velocity vector and the spatial approximation distance to construct a time-series risk feature matrix, input the time-series risk feature matrix into the time-series prediction network, and predict the probability of the subchondral bone filling material breaking through the cortical periphery and the time of the overflow. The specific process of constructing non-Newtonian rheological characterization parameters by combining the local propagation velocity vector and the frontal morphology curvature, and locating the jet inrush node on the rheological front profile, is as follows: Extract the magnitude parameter of the local propagation velocity vector and the peak parameter of the frontal morphology curvature; The modulus parameter and the peak parameter are fused using tensor product features to generate non-Newtonian rheological characterization parameters that characterize shear thinning and viscous fingering rheological properties. Calculate the spatial distribution gradient of non-Newtonian rheological characterization parameters along the rheological front profile; Extract the spatial extreme coordinates of the step distortion of the spatial distribution gradient, and lock the spatial extreme coordinates as the jet inrush node.

2. The method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopy sequences according to claim 1, characterized in that: The specific process of acquiring multi-sequence MRI images of the knee joint and extracting the subchondral bone lesion region and the periphery of the bone cortex is as follows: Acquire proton density-weighted images and fat-suppressed sequence images of the knee joint, and perform spatial registration and fusion to generate a multimodal contrast voxel matrix; Extract the edema signal attenuation edge from the multimodal contrast voxel matrix, peel off the normal bone marrow fat signal region, and close and define the subchondral bone lesion region; The signal-free black band region outside the multimodal contrast voxel matrix is ​​tracked, and the continuity features of the inner and outer physical boundaries of the signal-free black band region are extracted. The three-dimensional closed bone cortex perimeter is then smoothly fitted and generated.

3. The method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopy sequences according to claim 1 or 2, characterized in that: The specific process of analyzing the spatial texture distribution of trabeculae within the subchondral bone lesion area, constructing an anisotropic permeability tensor field, and extracting the principal permeability axis vector and the extreme value band of the permeability gradient within the anisotropic permeability tensor field is as follows: Traverse the three-dimensional voxel blocks within the subchondral bone lesion region, extract the three-dimensional gray-level co-occurrence matrix of the three-dimensional voxel blocks, and construct the spatial structure tensor based on the local directional gradient of the three-dimensional gray-level co-occurrence matrix. Eigenvalue decomposition is performed on the spatial structure tensor, and the eigenvector corresponding to the largest eigenvalue is extracted as the spatial penetration dominance direction. The spatial penetration dominance direction is then mapped back to the subchondral bone lesion region to generate an anisotropic penetration tensor field. Calculate the spatial derivative of the anisotropic index within the anisotropic permeation tensor field, and extract the three-dimensional connected regions where the spatial derivative reaches a local maximum as the permeation gradient extreme zone; The normalized direction vector is extracted along the spatial permeability advantage direction inside the permeability gradient extreme zone and defined as the main permeability axis vector.

4. The method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopy sequences according to claim 1, characterized in that: The specific process of acquiring continuous fluoroscopic sequences of the knee joint during surgery, using the bone anatomical contours within the field of view as spatial registration anchor points, and dimensionally reducing and mapping the subchondral bone lesion area, the cortical bone perimeter, and the anisotropic permeation tensor field to a two-dimensional fluoroscopic plane to generate a fluoroscopic background reference map superimposed with the principal permeation axis vector and the permeation gradient extreme value zone is as follows: The anatomical edge contours of the medial and lateral femoral condyles and the tibial plateau in the field of view of a continuous perspective sequence are extracted by the edge detection operator, and the bone anatomical contours in a two-dimensional coordinate system are constructed. Based on the geometric parameters of intraoperative fluoroscopic imaging, a fluoroscopic projection matrix of the virtual camera is established, and the periphery of the bone cortex is projected onto the projection plane using the fluoroscopic projection matrix to generate a two-dimensional virtual projection contour. Perform iterative nearest-point spatial alignment between the anatomical contour and the two-dimensional virtual projection contour, and output the optimal pose transformation matrix; Based on the optimal pose transformation matrix, the subchondral bone lesion area and the anisotropic permeation tensor field are simultaneously reduced in dimension and mapped to a two-dimensional perspective plane. According to the perspective projection depth attenuation function, the main permeation axis vector and the extreme value of the permeation gradient are assigned two-dimensional mapping weights, and the perspective background baseline map is fused and output.

5. The method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopy sequences according to claim 1, characterized in that: The specific process for identifying the rheological front profile of subchondral bone filling material in a continuous fluoroscopic sequence and calculating the local propagation velocity vector and front morphological curvature along the normal direction of the rheological front profile is as follows: Extract the high-density difference pixel set of adjacent perspective frames in a continuous perspective sequence, drive the active contour model to perform closed convergence evolution along the outer edge of the high-density difference pixel set, and output the rheological front contour. Discretize the boundary nodes on the rheological front profile and extract the normal direction field pointing outward from the boundary nodes; Calculate the dense optical flow field of adjacent perspective frames in a continuous perspective sequence, project the dense optical flow field onto the normal direction field, and generate the local propagation velocity vector of each boundary node on the rheological front profile; Extract the discrete spatial coordinates of the rheological front profile, perform a second spatial differentiation on the discrete spatial coordinates, and extract the spatial boundary curvature features on the normal direction field to generate the front morphology curvature.

6. The method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopy sequences according to claim 1, characterized in that: The specific process of mapping the jet ingress node onto the fluoroscopic background reference image, calculating the angle between the local propagation velocity vector of the jet ingress node and the main permeation axis vector, and simultaneously measuring the spatial approximation distance from the jet ingress node to the periphery of the bone cortex and the extreme value zone of the permeation gradient is as follows: Extract the two-dimensional spatial pixel coordinates of the jet intrusion node, and register and map the two-dimensional spatial pixel coordinates into the two-dimensional coordinate system of the perspective background reference image; Extract the main permeation axis vector that coincides with the pixel coordinate space in the two-dimensional coordinate system, perform the inverse cosine operation of the vector inner product of the local propagation velocity vector and the main permeation axis vector, and generate the included angle analytically; Extract the shortest Euclidean distance vector from the two-dimensional spatial pixel coordinates to the periphery of the bone cortex, and extract the shortest Euclidean distance vector from the two-dimensional spatial pixel coordinates to the two-dimensional mapping boundary of the extreme value zone of the osmotic gradient; Extract the magnitude of the shortest Euclidean distance vector to generate the spatial approximation distance.

7. The method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopy sequences according to claim 1, characterized in that: A temporal risk feature matrix is ​​constructed by combining the included angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance. This temporal risk feature matrix is ​​then input into a temporal prediction network to predict the probability of subchondral bone filling material breaching the cortical periphery and the specific process of the spillover arrival time, as follows: Following the timestamp dimension of the continuous perspective sequence, the angles extracted in each step, the magnitude of the local propagation velocity vector, and the spatial approximation distance are sequentially spliced ​​to construct a temporal risk feature matrix containing latent temporal dimensions. The temporal risk feature matrix is ​​input into a temporal prediction network equipped with temporal convolutional kernels and gated memory units; The temporal convolution kernel is called to extract the dynamic jump features of the temporal risk feature matrix at adjacent time stamps, and the gated memory unit is called to establish the rheological evolution trend hidden state across time stamps; The nonlinear regression decoding module integrates dynamic jump characteristics and rheological evolution trend hidden state inputs, outputs the spillover occurrence probability, and extrapolates the spillover arrival time along the gradient descent direction of the rheological evolution trend hidden state.

8. A system for predicting the risk of extravasation of subchondral bone filler material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences, used to perform the method for predicting the risk of extravasation of subchondral bone filler material in the knee joint by integrating MRI lesion localization and intraoperative fluoroscopic sequences as described in any one of claims 1-7, characterized in that, include: The tensor construction module is used to acquire multi-sequence MRI images of the knee joint, extract the subchondral bone lesion area and the periphery of the bone cortex; analyze the spatial texture distribution of trabeculae within the subchondral bone lesion area, construct an anisotropic permeability tensor field, and extract the principal permeability axis vector and permeability gradient extreme value band within the anisotropic permeability tensor field. The dimension reduction mapping module is used to acquire continuous fluoroscopic sequences of the knee joint during surgery. Using the bone anatomical contours in the field of view as spatial registration anchor points, the subchondral bone lesion area, the periphery of the bone cortex, and the anisotropic permeation tensor field are dimensionally reduced and mapped to a two-dimensional fluoroscopic plane to generate a fluoroscopic background reference map superimposed with the main permeation axis vector and the permeation gradient extreme value zone. The rheology identification module is used to identify the rheological front profile of subchondral bone filling material in a continuous fluoroscopic sequence, calculate the local propagation velocity vector and front morphological curvature of the rheological front profile along the normal direction, and construct non-Newtonian rheological characterization parameters by combining the local propagation velocity vector and front morphological curvature to locate the jet intrusion node on the rheological front profile. The risk prediction module maps the jet intrusion node onto the perspective background baseline image, calculates the angle between the local propagation velocity vector of the jet intrusion node and the main permeation axis vector, and simultaneously measures the spatial approximation distance from the jet intrusion node to the cortical periphery and the extreme value zone of the permeation gradient. The module combines the angle, the magnitude of the local propagation velocity vector, and the spatial approximation distance to construct a temporal risk feature matrix. The temporal risk feature matrix is ​​then input into the temporal prediction network to predict the probability of subchondral bone filling material overflowing the cortical periphery and the overflow arrival time.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for predicting the risk of extravasation of subchondral bone filling material in the knee joint by fusing MRI lesion localization and intraoperative fluoroscopic sequences as described in any one of claims 1-7.

Images