Artificial intelligence-based spinal deformity correction effect prediction system

Abstract

The present application relates to the technical field of physiological health data processing, in particular to a spine deformity correction effect prediction system based on artificial intelligence. The system determines the local mechanical property field of the patient's spine position. Based on the correction surgery planning data, the deviation vector and the deviation distance generated by the surgery are determined, the deviation vector and the gradient vector of the voxel point in the local mechanical property field of the bone are analyzed, the crack propagation path generated by the surgery is simulated through the path optimization method, and the minimum crack propagation work is quantified. By comparing the accumulated path weight and the correction load energy, the load-resistance failure index which can accurately quantify the surgical risk is obtained, which is used for predicting the surgical risk. The present application accurately identifies the directional failure risk through strength stress analysis, outputs the minimum crack propagation work and effectively evaluates the surgical risk.

Description

Technical Field

[0001] This invention relates to the field of physiological health data processing technology, specifically to an artificial intelligence-based system for predicting the effect of spinal deformity correction. Background Technology

[0002] Spinal deformity correction surgery (such as scoliosis correction and kyphosis osteotomy) is a high-risk procedure in orthopedics. The core step of the surgery involves pressing a pre-bent metal rod into the screw cap implanted in the vertebral body, using the rod's rebound force to force the vertebral body back into place. However, this process introduces a huge mechanical load into the vertebral-screw interface. For patients with osteoporosis or age-related degenerative spine disease, their bone microstructure (trabeculae) often exhibits high spatial heterogeneity (i.e., anisotropy). If the rebound force generated by the correction operation happens to point towards areas of thin or cavitary bone within the vertebral body, it can easily cause the screw to cut into the bone, loosen, or even be pulled out, leading to serious complications such as proximal junctional failure (PJF).

[0003] Current techniques primarily rely on dual-energy X-ray absorptiometry (DXA) or CT measurements to assess bone quality. These methods typically output a regional average density value or T-score. This scalar statistical assessment method has significant limitations: it ignores the spatial distribution characteristics of bone microstructure and the directional coupling relationship between surgical corrective forces. For example, two regions may have the same average bone density, but one region may have a uniform distribution of trabeculae, while the other may have cavities oriented in the direction of force. Under the same corrective force, the former may be stable, while the latter may experience oriented structural collapse. Therefore, current techniques only determine the simple biomechanical characteristics of the spine's location, ignoring the imaging of force direction and bone distribution, leading to inaccurate predictions of surgical outcomes and a lack of reference value. Summary of the Invention

[0004] To address the problem that existing technologies only determine the simple biomechanical characteristics of the spine's location, neglecting the imaging of force direction and bone distribution, leading to inaccurate predictions of surgical outcomes and a lack of relevant technical references, the present invention aims to provide an artificial intelligence-based system for predicting the effects of spinal deformity correction. The specific technical solution adopted is as follows: This invention proposes an artificial intelligence-based system for predicting the effect of spinal deformity correction, the system comprising: The spinal CT image feature extraction module is used to determine the trabecular yield strength of each voxel point based on the grayscale information in the patient's spinal CT image and to construct the local mechanical property field of the bone. The corrective load energy analysis module is used to obtain the deviation vector and deviation distance between the coordinates of the key screws in the spine and the corresponding anchor point coordinates on the orthosis based on the corrective surgery planning data; obtain the corrective load energy according to the deviation distance and the rated stiffness coefficient of the orthosis; obtain the degree of collinearity between the gradient vector of each voxel point in the local mechanical property field of bone and the deviation vector, and obtain the structural vulnerability weight of each voxel point based on the degree of collinearity. The bone structure energy load analysis module is used to take the voxel points in contact with the screw as source nodes; and the voxel points within a preset radius centered on the screw axis as sink nodes; to perform path optimization between the source nodes and sink nodes and obtain the cumulative path weight corresponding to the optimal path; during the optimization process, the path weight between any two voxel points is obtained by the trabecular yield strength, structural fragility weight, and voxel spacing of the voxel points. The energy release-load comparison module is used to compare the accumulated path weight with the corrective load energy to obtain the load-resistance failure index, and to assess the risk of corrective surgery based on the load-resistance failure index.

[0005] Furthermore, the method for obtaining the yield strength of the trabecular bone includes: Using pre-calibrated phantom data from CT equipment, spinal CT images are converted into apparent bone mineral density images; based on empirical formulas in bone mechanics, bone mineral density is converted into the yield strength of the trabecular bone.

[0006] Furthermore, the method for quantifying the corrective load energy includes: The deviation distance is used as the displacement in the standard expression of elastic potential energy, and the corrective load energy is obtained by combining it with the rated stiffness coefficient.

[0007] Furthermore, the degree of collinearity is the dot product of the gradient vector and the deviation vector.

[0008] Furthermore, obtaining the structural vulnerability weight of each voxel point based on the degree of collinearity includes: If the degree of collinearity is greater than or equal to 0, the structural vulnerability weight is set to a positive integer 1; otherwise, the structural vulnerability weight is set to the sum of a positive integer 1 and the absolute value of the degree of collinearity.

[0009] Furthermore, the method for obtaining the path weight includes: For any two nodes, the ratio of the trabecular yield strength of the precursor node to the structural fragility weight is taken as the unit critical failure strength, the product of the unit critical failure strength and the voxel area is taken as the critical failure force, and the product of the critical failure force and the voxel spacing is taken as the path weight.

[0010] Furthermore, the method for obtaining the load-resistance failure index includes: The load-resistance failure index is obtained by taking the corrective load energy as the numerator, the sum of the accumulated path weights after value range adjustment and the foundation stability resistance energy as the denominator.

[0011] Furthermore, the assessment of corrective surgery risk based on the load-resistance failure index includes: If the load-resistance failure index is greater than or equal to a preset high-risk threshold, a high-risk signal is fed back; if the load-resistance failure index is greater than or equal to a preset low-risk threshold and less than a preset high-risk threshold, a critical risk signal is fed back; if the load-resistance failure index is less than a low-risk threshold, a low-risk signal is fed back.

[0012] Furthermore, the assessment of corrective surgery risk based on the load-resistance failure index further includes: If a high-risk signal is detected, the optimal path will be visually marked with a preset color, and the node with the minimum path weight on the optimal path will be marked, and the deviation vector will be visually displayed.

[0013] Furthermore, after obtaining the deviation distance, the process further includes: If the deviation distance is less than or equal to a preset distance threshold, a safety signal is fed back and the execution of subsequent modules is stopped; otherwise, the module algorithm continues to be executed.

[0014] The present invention has the following beneficial effects: This invention first utilizes existing CT image analysis methods to determine the local mechanical property field of the bone in the patient's spine. Then, based on corrective surgery planning data, it determines the deviation vector and deviation distance that will occur during the surgery, where the deviation distance quantifies the magnitude of the corrective load energy generated during the surgery. By performing collinearity analysis between the deviation vector and the gradient vector of voxel points in the local mechanical property field of the bone, the anatomical features shown in the CT images can be coupled with the surgical planning features. This mechanism can specifically quantify the degree of bone strength attenuation along the direction of force during surgery, thereby using structural vulnerability weights to quantify those hidden weak areas aligned with the direction of destructive force. Furthermore, through path optimization methods, it simulates the crack propagation path generated during surgery and quantifies the minimum crack propagation work. This feature characterizes the minimum energy threshold that must be overcome for the screw to undergo micro-loosening. Using bone spatial information to represent the bone structure's resistance to corrective load energy, by comparing the accumulated path weights with the corrective load energy, a load-resistance failure index that accurately quantifies surgical risk can be obtained, achieving effective prediction of corrective surgery risk. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a structural block diagram of an artificial intelligence-based spinal deformity correction effect prediction system provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an artificial intelligence-based spinal deformity correction effect prediction system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The specific solution of the artificial intelligence-based spinal deformity correction effect prediction system provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Please see Figure 1 The diagram illustrates a structural block diagram of an artificial intelligence-based spinal deformity correction effect prediction system according to an embodiment of the present invention. The system includes a spinal CT image feature extraction module 101, a correction load energy analysis module 102, a bone structure energy load analysis module 103, and an energy release-load comparison module 104.

[0021] It should be noted that the system proposed in this embodiment of the invention can run on existing medical artificial intelligence analysis systems. By calling CT image data from CT equipment, the data is processed and analyzed by an artificial intelligence chip. After obtaining the analysis results, they can be visualized through a visualization device, thereby assisting doctors in evaluating and judging corrective surgery. The spinal CT image feature extraction module 101 can directly read thin-slice spinal CT images from a PACS system via a standard DICOM interface, generating a local mechanical property field of bone through grayscale-intensity mapping, without the need for additional acquisition equipment. The corrective load energy analysis module 102 is compatible with structured data exported from mainstream surgical planning software, calculates the deviation vector between the screw and the anchor point and the corrective load energy, and calculates the structural vulnerability weight by combining the collinearity of the intensity gradient vector and the deviation vector. The bone structure energy load analysis module 103 is based on a general graph theory algorithm library, setting the screw contact voxel as the source node and the voxels within a preset radius range as the sink node, and performing path optimization to obtain the minimum cumulative path weight. The path weight is jointly determined by the voxel yield strength, the structural vulnerability weight, and the voxel spacing. Finally, the energy release-load comparison module 104 compares the accumulated path weights with the corrected load energy to generate a load-resistance failure index to assess surgical risk. The entire process is based on a general computing library and standard data protocols, can run in near real-time on clinical workstations, outputs standard format risk maps, and has clear engineering feasibility.

[0022] The spinal CT image feature extraction module 101 is used to determine the trabecular yield strength of each voxel point based on the grayscale information in the patient's spinal CT images and to construct a local mechanical property field of bone. The method of quantifying the mechanical characteristics of each voxel point based on the grayscale information in the spinal CT images is a technique known to those skilled in the art, i.e., the mechanical analysis of the prior art proposed in the background section. The trabecular yield strength of each voxel point in the obtained local mechanical property field of bone quantifies the basic ability of bone to resist mechanical damage at the corresponding anatomical location.

[0023] It should be noted that, in one specific implementation of this invention, to ensure a unified spatial standard for subsequent module algorithms, after calling the spinal CT images, the system proposed in this embodiment does not establish a patient CT anatomical space. This space uses the upper left pixel of the first slice of the CT scan sequence as the origin and follows the right-hand rule defined by the DICOM standard to establish a global three-dimensional Cartesian coordinate system, with the spatial units uniformly converted to millimeters. In this space, the system can determine the surgical target segment, such as the L1-L5 vertebrae, according to the doctor's specification or the automatic segmentation algorithm. To further avoid the influence of non-surgical areas on subsequent module algorithms, this embodiment extracts a cuboid region containing the target vertebrae and its bilateral pedicle regions, defined as the region of interest. Subsequent processing is performed within the region of interest; voxel points in other regions are not analyzed.

[0024] In one specific implementation of this invention, the spinal CT image feature extraction module 101 further includes denoising processing of the image data before extracting the mechanical features of voxel points. Since noise is inevitably introduced during the acquisition and reconstruction of the original CT images, and artifacts may occur if there are previous implants (such as metal rods) remaining in the patient's body, these high-frequency noises can cause severe oscillations or even errors in the calculation of bone density gradients in subsequent steps. Therefore, this invention employs three-dimensional anisotropic diffusion filtering or three-dimensional Gaussian smoothing filtering to preprocess the voxel points in the spinal CT images to achieve denoising and improve image quality. Both three-dimensional anisotropic diffusion filtering and three-dimensional Gaussian smoothing filtering are well-known technical fields to those skilled in the art. Taking three-dimensional Gaussian smoothing filtering as an example, the standard deviation of the three-dimensional Gaussian kernel function is set to a voxel spacing of 0.5 to 1.0. The original grayscale values ​​of the voxel points are convolved using the three-dimensional Gaussian kernel function. After processing, the macroscopic texture features of the trabecular bone are preserved while filtering out minute noise points, resulting in a grayscale image with continuous spatial distribution and smooth gradient changes.

[0025] Preferably, in this embodiment of the invention, spinal CT images are converted to apparent bone mineral density images using pre-calibrated phantom data from a CT device; based on empirical formulas in bone mechanics, bone mineral density is converted to the yield strength of the trabecular bone. As a specific example, in this embodiment of the invention, the formula for converting grayscale values ​​in spinal CT images to bone mineral density can be expressed as: in, Let be the bone mineral density of the nth voxel, 'a' be the data from the first phantom, and 'b' be the data from the second phantom. The grayscale value of the nth voxel in the spinal CT image. The first and second phantom data can be obtained by regression analysis of known density samples from a portable calibration phantom. If real-time phantom data is unavailable in certain implementation scenarios, empirical values ​​based on scanning parameters can be used, where the first phantom data can be set to 0.001 and the second phantom data to 1. The specific conversion formula logic is a well-known technique and will not be elaborated upon here.

[0026] The empirical formula for bone mechanics can be expressed as: in, Let A be the yield strength of the trabecular bone at voxel n, and B be the first empirical constant and the second empirical constant. This empirical formula is based on the classical empirical formula of Morgan-Keaveny mechanics. The yield strength of the trabecular bone represents the ultimate stress that the bone within voxel n can withstand before microstructural yielding or crushing. In this embodiment of the invention, considering that the lumbar spine is a cancellous bone region, the first empirical constant is set to 45 and the second empirical constant is set to 1.8. This setting can represent that the strength is approximately proportional to the square of the bone density, and that a small decrease in bone density will cause a significant decrease in strength. The specific logic of the formula is also a well-known technical means in the art and will not be elaborated here.

[0027] The corrective load energy analysis module 102 is used to analyze the planning data of the corrective surgery and pre-quantify the corrective load energy that the surgical outcome will generate on the spine. The core operation of spinal deformity correction is to press the pre-bent corrective rod into the screw cap. This process generates a huge rebound force, and the corrective load energy characterizes the energy contained in this rebound force generated by the surgical outcome.

[0028] The corrective load energy analysis module 102 can import digital surgical planning data, which typically includes the ideal screw implantation location and the pre-bending shape of the corrective rod set by the surgeon in the preoperative planning software. It should be noted that the corrective surgical planning data needs to be mapped to the same coordinate system space as the spinal CT image. Based on the corrective surgical planning data, the deviation vector and deviation distance between the coordinates of the key spinal screws and the corresponding anchor point coordinates on the orthosis are obtained. The coordinates of the key spinal screws can be selected from any of the spinal screw locations. In one specific implementation of this invention, the coordinates of the geometric center of the pedicle screw head at the uppermost fixed vertebra and other key segments are selected. The essential meaning of these coordinates is to select a reference point for deviation analysis, and therefore, they can be selected independently.

[0029] In one specific implementation of this invention, the method for obtaining the corresponding anchor point on the orthotic device includes: First, extract the three-dimensional spline curve of the pre-bent orthotic central axis from the orthotic surgical planning data. Then, perform a geometric search algorithm in three-dimensional space to find a point on the three-dimensional spline curve that minimizes the Euclidean distance from that point to the key screw coordinates. This point is the target anchor point for the orthotic to connect with the screw in an ideal state.

[0030] It should be noted that the methods for obtaining the deviation vector and deviation distance between two coordinate points in three-dimensional space are well-known to those skilled in the art. In this embodiment of the invention, the deviation vector is the coordinate of the key spinal screw pointing to the corresponding anchor point on the orthosis. Furthermore, for the convenience of subsequent collinearity analysis, the deviation vector is a unit vector, that is, it only represents the direction of movement of the screw due to the rebound effect of the orthosis, which tends to cause it to be pulled out, cut, or sink.

[0031] After the screws are forcibly inserted, the orthosis will exhibit an elastic rebound tendency to return to its original shape. The direction of this rebound force determines the specific path of action for bone destruction. The generated elastic force is strongly correlated with the deviation distance between the orthosis and the spine; therefore, the corrective load energy can be obtained based on the deviation distance and the orthosis's rated stiffness coefficient. The greater the corrective load energy obtained, the higher the destructive potential energy generated on the spine by the current surgical plan, and the stronger the driving force for internal fixation failure.

[0032] The gradient direction in the local mechanical property field of bone represents the direction of the fastest intensity change at the voxel location, i.e., the direction of bone strengthening. If the gradient vector and the deviation vector are collinear and in opposite directions (i.e., the angle is greater than 90 degrees), it indicates that the force direction points towards the bone weakening region; if the gradient vector and the deviation vector are collinear and in the same direction (i.e., the angle is less than 90 degrees), it indicates that the force direction points towards the bone strengthening region. Therefore, the corrective load energy analysis module 102 further obtains the degree of collinearity between the gradient vector and the deviation vector of each voxel in the local mechanical property field of bone, and obtains the structural vulnerability weight of each voxel based on the degree of collinearity. The structural vulnerability weight is used to quantify the spatial characteristics of bone at each voxel location.

[0033] Preferably, in some implementations of the present invention, considering that a small deviation distance will not have significant elastic potential energy, if the deviation distance is less than or equal to a preset distance threshold, a safety signal is fed back and subsequent module execution is stopped; otherwise, the module algorithm continues to be executed. The distance threshold can be set to a small value, such as 0.001 mm.

[0034] Preferably, in some implementations of the embodiments of the present invention, the method for quantifying the corrected load energy includes: The deviation distance is used as the displacement in the standard expression for elastic potential energy, and the corrective load energy is obtained by combining it with the rated stiffness coefficient. The standard expression for elastic potential energy is expressed by the formula: in, To correct for load energy, k is the rated stiffness coefficient. The deviation distance is given. This formula is a well-known existing formula in the art and will not be elaborated further. In this embodiment of the invention, a simplified linear elastic model is used to reflect the order of magnitude and trend of energy change. The rated stiffness coefficient can be estimated based on the material properties and geometric cross-sectional properties of the corrector, combined with the effective cantilever length of the corrector bar in the target segment. Where E is the elastic modulus in the material properties, for example, about 110 GPa for titanium alloys; and I is the moment of inertia of the cross section in the geometric properties. The effective cantilever length of the orthosis within the region of interest in the spine. All of the above parameters can be found and accessed in the orthopedic surgical planning data. The relationships between the parameters are well-known to those skilled in the art. Because k is estimated, the final corrective load energy does not represent an absolutely precise physical energy value, but rather an engineering index used to characterize the intensity of the surgical correction, and can only be used in the analysis process of the embodiments of this invention.

[0035] Preferably, in this embodiment of the invention, the degree of collinearity is the dot product of the gradient vector and the deviation vector. That is, the dot product represents the collinear components between the two vectors. If the two vectors are perpendicular, the dot product is 0, indicating that there is no intensity change in the direction of force. If the two are parallel, the absolute value of the dot product is the largest, indicating that the direction of force is directly opposite the direction of the most drastic intensity change, which may be the direction of increasing intensity (gradient direction) or the direction of decreasing intensity (negative gradient direction).

[0036] Furthermore, since the sign of the dot product result can significantly characterize the directional relationship between two vectors, if the dot product result is less than 0, the surface moves along the direction of the corrective rebound force, and the bone strength shows a decreasing trend, that is, the force is directed towards the osteoporotic or cavitary area. The larger the absolute value, the more severe the precipitous drop in strength in the forward direction, and the higher the risk of local forward collapse. If the dot product result is greater than or equal to 0, it indicates that the force is biased towards the bone-enhancing area, and the structure is relatively stable. Therefore, based on the degree of collinearity, the structural vulnerability weight of each voxel point is obtained, including: If the degree of collinearity is greater than or equal to 0, the structural vulnerability weight is set to a positive integer 1; otherwise, the structural vulnerability weight is set to the sum of a positive integer 1 and the absolute value of the degree of collinearity. That is, for every 1 MPa / mm decrease in sequential strength, the effective resistance at that location will be reduced by approximately half, thereby significantly improving the sensitivity of identifying weak areas at the edge of voids.

[0037] It should be noted that in other implementations of this invention, when the contribution level is less than 0, the absolute value of the collinearity can be multiplied by a preset sensitivity coefficient and then added to a positive integer 1 to obtain the structural vulnerability weight. The preset sensitivity coefficient allows adjustment of the algorithm's sensitivity, making it more sensitive to voxel points of certain vulnerable structures.

[0038] It should be noted that the gradient vector of each voxel point in the local mechanical property field of bone can be obtained by the three-dimensional central difference method. The specific method for obtaining the gradient vector is a well-known technique in the art and will not be elaborated here.

[0039] The bone structure energy load analysis module 103 is used to analyze the spinal bone structure's resistance to the elastic energy generated by the orthodontic appliance. Therefore, it is necessary to simulate the energy conduction path generated by the appliance, using the voxel points in contact with the screw as source nodes and the voxel points within a preset radius centered on the screw axis as sink nodes. Path optimization between these two types of points yields an optimal path that spatially delineates the continuous region where the bone is weakest and most prone to cascading collapse. Since damage always occurs along the path of least resistance, to achieve the effect of the optimal path, a path weight representing the energy transmission resistance between voxel points needs to be defined. That is, the path weight should characterize the damage resistance between two voxel points; the smaller the path weight, the easier the bone is to break, and the more likely the elastic force-induced wear cracks will occur along that location. Therefore, to obtain an effective path weight, it is necessary to effectively quantify the trabecular yield strength, structural fragility weight, and voxel spacing obtained from the above module. The greater the yield strength of the trabecular bone, the stronger the resistance of the voxel to elastic force, and the greater the corresponding path weight should be; the smaller the structural fragility weight, the more likely the voxel is located in a hard bone, and the greater the corresponding path weight should be; the larger the distance between voxels, the farther the distance between the two nodes, and the less likely it is to be a continuous conduction path, and the greater the corresponding path weight should be.

[0040] Preferably, in this embodiment of the invention, the method for obtaining path weights includes: For any two nodes, the ratio of the trabecular yield strength of the precursor node to the structural fragility weight is taken as the unit critical failure strength. The product of the unit critical failure strength and the voxel area is taken as the critical failure force. The product of the critical failure force and the voxel spacing is taken as the path weight. In the above process, it can be understood that the product of the voxel area and the voxel spacing is the bone volume involved in one step of crack propagation. The larger this volume is, the greater the difficulty of crack transmission. Therefore, the product with the unit critical failure strength can be regarded as the energy required to destroy a voxel, i.e., the crack transmission resistance.

[0041] It should be noted that the voxel area is the square of the voxel side length. The methods for obtaining the voxel area and the voxel spacing are conventional methods in 3D spatial analysis and will not be elaborated here.

[0042] In this embodiment of the invention, the path optimization algorithm can be either Dijkstra's algorithm or A algorithm*. By constructing a grid of source and sink nodes, the optimal path is the one with the smallest cumulative path weight from any point in the source node to any point in the sink node. Specific path optimization algorithms are well-known to those skilled in the art and will not be elaborated upon or limited here.

[0043] It should be noted that the preset radius range designed in the process of defining the confluence node in this embodiment of the invention can be set to 1.5 times the screw radius, and the voxel point on the cylindrical surface centered on the screw axis is finally selected as the confluence node. Furthermore, in other implementations of this embodiment of the invention, a vertebral cortical bone boundary node within a preset range in front of the screw can also be selected to prevent the risk of the screw sinking forward or cutting laterally. The specific selection object can be specifically set in the specific implementation scenario, and will not be elaborated or limited here.

[0044] Finally, the energy release-load comparison module 104 can be used to compare the accumulated path weight with the corrective load energy to assess the difference between the elastic force generated by the current surgical plan and the patient's bone resistance, thereby obtaining the load-resistance failure index. The risk of the corrective surgery is then assessed based on this load-resistance failure index. In other words, a higher load-resistance failure index indicates a more dangerous current surgical plan, requiring the physician to make specific adjustments based on the patient's condition.

[0045] Preferably, in this embodiment of the invention, considering that the corrective load energy is the nominal energy estimated based on a macroscopic linear elasticity model, while the accumulated path weight is based on the dissipated work accumulated by microscopic voxel paths, the two differ in physical scale and energy transfer mechanism. To establish a scientific quantitative relationship, certain parameters need to be introduced to adjust the value range of the data, thereby eliminating the influence of dimensions and obtaining a dimensionless load-resistance failure index. Specifically, this includes: taking the corrective load energy as the numerator, and using the sum of the accumulated path weight (after value range adjustment) and the foundation stability resistance energy as the denominator, to obtain the load-resistance failure index. The above process is expressed by the formula: in, The load-resistance failure index. To correct the load energy, This is the range adjustment coefficient. The basic stabilizing resistance energy is defined as follows: The range adjustment coefficient is used to correct the effective proportion of macroscopic device deformation energy converted into microscopic crack driving energy. This is because, physically, not all input elastic potential energy contributes to crack propagation; for example, some is absorbed by soft tissue deformation and heat dissipation. Therefore, the range adjustment coefficient will be less than 1. In this embodiment, it is set to 0.05, indicating that approximately 5% of the macroscopic orthotic rod potential energy is effectively converted into work driving microscopic bone crack propagation. The basic stabilizing resistance energy characterizes the basic stabilizing resistance provided by factors other than bone structure, such as screw friction with surrounding soft tissue and fibrous connections. This also prevents the denominator from being zero in the calculation; in this embodiment, the value is 0.001.

[0046] Preferably, in this embodiment of the invention, assessing the risk of corrective surgery based on the load-resistance failure index includes: If the load-resistance failure index is greater than or equal to the preset high-risk threshold, it indicates that the input effective destructive energy exceeds the resistance limit of the bone structure, predicting that instantaneous failure (such as screw pullout or fracture) will occur during or early after the operation, and feeding back a high-risk signal. At the same time, it is recommended that the doctor modify the surgical plan, such as changing the screw placement path or reducing the pre-bending of the corrective rod. If the load-resistance failure index is greater than or equal to the preset low-risk threshold and less than the preset high-risk threshold, it indicates that although the bone structure is not immediately destroyed, it is in a high-stress critical state and is very prone to fatigue failure under postoperative cyclic load, thus providing a critical risk signal. It is also recommended that the doctor strengthen the fixation, for example, by using bone cement for reinforcement. If the load-resistance failure index is less than the low-risk threshold, it indicates that the bone structure has a sufficient safety margin, the surgical plan is physically stable, and the feedback is a low-risk signal.

[0047] The low-risk threshold is lower than the high-risk threshold because the load-resistance failure index is essentially a ratio. Therefore, the low-risk threshold is set to 0.6 and the high-risk threshold is set to 1.

[0048] Furthermore, after assessing the risk of corrective surgery based on the load-resistance failure index, the procedure also includes: If a high-risk signal is detected, the optimal path will be visually marked with a preset color, and the node with the lowest path weight on the optimal path will be marked. The deviation vector will also be visualized. The preset color can be a striking color, such as red. Marking the node with the lowest path weight aims to identify weak points, making it easier for doctors to modify the surgical plan. Visualizing the deviation vector helps doctors understand the direction of force and analyze its source.

[0049] In summary, this invention determines the local mechanical property field of the bone at the patient's spinal position. Based on the surgical planning data, the deviation vector and deviation distance generated by the surgery are determined. The deviation vector and the gradient vector of the voxel point in the local mechanical property field of the bone are analyzed for collinearity. Through path optimization, the crack propagation path generated by the surgery is simulated and the minimum crack propagation work is quantified. By comparing the accumulated path weight with the corrective load energy, a load-resistance failure index that can accurately quantify surgical risk can be obtained for predicting surgical risk. This invention accurately identifies directional failure risk through strength-force forward analysis, outputs the minimum crack propagation work, and effectively assesses surgical risk.

[0050] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0051] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A spinal deformity correction effect prediction system based on artificial intelligence, characterized in that, The system includes: The spinal CT image feature extraction module is used to determine the trabecular yield strength of each voxel point based on the grayscale information in the patient's spinal CT image and to construct the local mechanical property field of the bone. The corrective load energy analysis module is used to obtain the deviation vector and deviation distance between the coordinates of the key screws in the spine and the corresponding anchor point coordinates on the orthosis based on the corrective surgery planning data; obtain the corrective load energy according to the deviation distance and the rated stiffness coefficient of the orthosis; obtain the degree of collinearity between the gradient vector of each voxel point in the local mechanical property field of bone and the deviation vector, and obtain the structural vulnerability weight of each voxel point based on the degree of collinearity. The bone structure energy load analysis module is used to take the voxel points in contact with the screw as source nodes; and the voxel points within a preset radius centered on the screw axis as sink nodes; to perform path optimization between the source nodes and sink nodes and obtain the cumulative path weight corresponding to the optimal path; during the optimization process, the path weight between any two voxel points is obtained by the trabecular yield strength, structural fragility weight, and voxel spacing of the voxel points. The energy release-load comparison module is used to compare the accumulated path weight with the corrective load energy to obtain the load-resistance failure index, and to assess the risk of corrective surgery based on the load-resistance failure index.

2. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, The method for obtaining the yield strength of the trabecular bone includes: Using pre-calibrated phantom data from CT equipment, spinal CT images are converted into apparent bone mineral density images; based on empirical formulas in bone mechanics, bone mineral density is converted into the yield strength of the trabecular bone.

3. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, The method for quantifying the corrective load energy includes: The deviation distance is used as the displacement in the standard expression of elastic potential energy, and the corrective load energy is obtained by combining it with the rated stiffness coefficient.

4. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, The degree of collinearity is the dot product of the gradient vector and the deviation vector.

5. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 4, characterized in that, The method of obtaining the structural vulnerability weight of each voxel point based on the degree of collinearity includes: If the degree of collinearity is greater than or equal to 0, the structural vulnerability weight is set to a positive integer 1; otherwise, the structural vulnerability weight is set to the sum of a positive integer 1 and the absolute value of the degree of collinearity.

6. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, The method for obtaining the path weight includes: For any two nodes, the ratio of the trabecular yield strength of the precursor node to the structural fragility weight is taken as the unit critical failure strength, the product of the unit critical failure strength and the voxel area is taken as the critical failure force, and the product of the critical failure force and the voxel spacing is taken as the path weight.

7. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, The method for obtaining the load-resistance failure index includes: The load-resistance failure index is obtained by taking the corrective load energy as the numerator, the sum of the accumulated path weights after value range adjustment and the foundation stability resistance energy as the denominator.

8. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, The assessment of corrective surgery risk based on the load-resistance failure index includes: If the load-resistance failure index is greater than or equal to a preset high-risk threshold, a high-risk signal is fed back; if the load-resistance failure index is greater than or equal to a preset low-risk threshold and less than a preset high-risk threshold, a critical risk signal is fed back; if the load-resistance failure index is less than a low-risk threshold, a low-risk signal is fed back.

9. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 8, characterized in that, The assessment of corrective surgery risk based on the load-resistance failure index also includes: If a high-risk signal is detected, the optimal path will be visually marked with a preset color, and the node with the minimum path weight on the optimal path will be marked, and the deviation vector will be visually displayed.

10. The artificial intelligence-based spinal deformity correction effect prediction system according to claim 1, characterized in that, After obtaining the deviation distance, the process also includes: If the deviation distance is less than or equal to a preset distance threshold, a safety signal is fed back and the execution of subsequent modules is stopped; otherwise, the module algorithm continues to be executed.

Images