Spinal screw placement preoperative planning quality control method, electronic device and computer storage medium

Abstract

The application provides a spine screw placement preoperative planning quality control method, an electronic device and a computer storage medium. By constructing boundary models of key risk objects such as pedicle cortex, spinal canal, intervertebral foramen and superior articular facet joint surface, combining preoperative screw placement planning data to calculate quantitative indicators such as minimum safety distance, predicted breaking amount and breaking direction, and inputting structured feature vectors and local image features of the planning channel into a risk assessment model to output risk assessment results, the safety hazards of screw placement can be accurately identified for complex cases such as pedicle morphology variation, degeneration or deformity. At the same time, the deficiencies of the existing planning system, such as the lack of quantitative quality control and grading early warning mechanism, can be made up. High-risk planning schemes can be prompted in time, and objective basis for path correction can be provided, so as to improve the accuracy, safety and standardization of spine screw placement preoperative planning.

Description

Technical Field

[0001] This application relates to the field of communication technology, and more specifically, to a method for quality control of preoperative planning for spinal screw placement, an electronic device, and a computer storage medium. Background Technology

[0002] In spinal screw placement surgery, the accuracy of screw entry point, direction, and size selection is crucial. Preoperative planning typically relies on surgeons' manual interpretation and experience based on CT images, leading to strong subjectivity and difficulty in ensuring consistent planning quality. In complex cases with pedicle morphology variations, degeneration, or deformities, the planned trajectory may pose safety risks such as pedicle cortical breakthrough, proximity to the spinal canal / intervertebral foramen, and facet joint invasion. Furthermore, existing planning systems often lack quantitative quality control and graded early warning mechanisms for planning results, making it difficult to promptly identify high-risk plans and guide corrections. Summary of the Invention

[0003] The purpose of this application is to provide a method for quality control of preoperative planning for spinal screw placement, an electronic device, and a computer storage medium to solve the problems of existing technologies that rely on doctors to make manual interpretations and experience-based designs based on CT images, which have strong subjectivity and make it difficult to guarantee the consistency of planning quality.

[0004] This application provides a method for preoperative planning and quality control of spinal screw placement, including: Based on the three-dimensional anatomical model of the vertebral body and pedicle, a boundary model of the risk object is generated; wherein the risk object includes at least one of the following: pedicle cortical boundary, spinal canal boundary, intervertebral foramen boundary, superior articular facet boundary; the three-dimensional anatomical model is obtained by modeling based on medical images; Based on the boundary model and preoperative nail placement planning data, calculate quantitative indicators related to the risk object; among which, the quantitative indicators include at least one of the following: minimum safe distance, expected perforation amount, and expected perforation direction; the minimum safe distance represents the minimum distance between the planned channel and the risk object, and the expected perforation amount represents the value by which the planned channel exceeds the boundary of the risk object; The quantitative indicators are processed into structured feature vectors; the medical images are cropped based on the planned channel positions to obtain local image features; The risk assessment results are obtained by inputting structured feature vectors and local image features into the risk assessment model.

[0005] The above technical solution constructs boundary models of key risk objects such as the pedicle cortex, spinal canal, intervertebral foramen, and superior articular facet. It then calculates quantitative indicators such as the minimum safe distance, expected perforation amount, and perforation direction based on preoperative screw placement planning data. By integrating structured feature vectors and local image features of the planned pathway into the risk assessment model, it outputs risk assessment results. Compared to existing technologies that rely on doctors manually interpreting CT images and designing based on experience, this solution effectively overcomes the shortcomings of strong subjectivity and difficulty in ensuring consistent planning quality. It can accurately identify potential screw placement safety hazards in complex cases such as pedicle morphological variations, degeneration, or deformities. Furthermore, it compensates for the lack of quantitative quality control and graded early warning mechanisms in existing planning systems, promptly alerting high-risk planning schemes and providing objective basis for path correction, significantly improving the accuracy, safety, and standardization of preoperative spinal screw placement planning.

[0006] In some optional implementations, generating a boundary model of the risk object based on a three-dimensional anatomical model of the vertebral body and pedicle includes: A three-dimensional medical image segmentation network is used to automatically identify targets in a three-dimensional anatomical structure model, thereby obtaining a boundary model of the risk object.

[0007] In some optional implementations, the input to the three-dimensional medical image segmentation network is the patient's preoperative three-dimensional CT volume data, and the output of the three-dimensional medical image segmentation network is a voxel-level label map; Based on the label map, triangular mesh boundaries, voxel boundaries, and implicit boundary models are generated using the Marching Cubes algorithm, boundary tracking algorithm, and distance transformation algorithm. The boundaries of risk objects are represented using a signed distance field. d_b(p) = s(p) · min_{q∈ Oh} ||p q|| 2 Where d_b(p) represents the distance from point p to the boundary of the risk object. Oh The signed distance; ||p q||2 represents the Euclidean distance; s(p) is the sign function, which takes a positive value when it is outside the boundary and a negative value when it is inside the boundary.

[0008] In the above technical solution, the risk object boundary model is based on the patient's preoperative 3D CT volume data. It uses a 3D medical image segmentation network to automatically identify the target structure and output voxel-level label maps. Then, the Marching Cubes algorithm, boundary tracking algorithm, and distance transformation algorithm are used to generate triangular mesh boundary, voxel boundary, and implicit boundary models, respectively. At the same time, a signed distance field is used to uniformly represent the risk object boundary. The sign function accurately distinguishes the internal and external positional relationship between spatial points and the boundary, and accurately quantifies the Euclidean distance from the spatial point to the risk boundary. This provides high-precision and standardized geometric boundary support for subsequent quantitative index calculations, ensuring the accuracy and stability of risk assessment.

[0009] In some alternative implementations, a three-dimensional medical image segmentation network is used, including 3D U-Net, V-Net, nnU-Net, or Swin UNETR.

[0010] In the above technical solution, the three-dimensional medical image segmentation network can adopt mature and efficient medical image segmentation models such as 3D U-Net, V-Net, nnU-Net or Swin UNETR. While ensuring the automatic segmentation accuracy of vertebral bodies, pedicles and various risk anatomical structures, it can adapt to spinal CT body data with different resolutions and different degrees of deformity, providing a stable and reliable voxel-level segmentation basis for subsequent boundary model construction, and improving the robustness and versatility of the entire preoperative planning and quality control process.

[0011] In some optional implementations, the risk assessment model includes: an input and preprocessing module, a feature extraction module, a feature fusion module, a prediction module, and an output and postprocessing module; The input and preprocessing module is used to receive structured feature vectors and local image data, and to normalize, handle missing data and vectorize the structured feature vectors, and to resample and crop the local image data. The feature extraction module is used to extract structured features and image features; The feature fusion module is used to fuse structured features and image features to form fused features; The prediction module is used to map fused features to risk probabilities or risk scores; The output and post-processing module is used to calibrate and determine the threshold for risk probability or risk score, and generate output values ​​for determining risk assessment results.

[0012] In the above technical solution, the risk assessment model consists of an input and preprocessing module, a feature extraction module, a feature fusion module, a prediction module, and an output and post-processing module. The input and preprocessing module can perform normalization, missing value processing, and vectorization encoding on the received structured feature vectors, and perform resampling and cropping operations on local image data. The feature extraction module and the feature fusion module realize efficient extraction of structured features and image features, and multimodal feature fusion, respectively. The prediction module maps the fused features to risk probability or risk score. The output and post-processing module generates standardized risk assessment results through probability calibration and threshold determination, realizing accurate and stable automated risk assessment of spinal screw placement planning schemes.

[0013] In some alternative implementations, preoperative screw placement planning data includes the entry point, axial direction, length, and diameter of each screw.

[0014] In some optional implementations, after obtaining the risk assessment results, the following steps are also included: If the risk assessment results are not satisfactory, the preoperative nail placement planning data is constrained and optimized within the preset feasible domain to obtain the corrected nail placement planning data.

[0015] In the above technical solution, if the risk assessment results are not up to standard, that is, if there is a high safety risk in the current screw placement plan, the preoperative screw placement plan data (including screw entry point, axis direction, length and diameter) will be constrained and optimized within the preset feasible domain. By adjusting the planning parameters, risks such as pedicle cortex perforation and spinal canal invasion are avoided, and finally the corrected screw placement plan data is output.

[0016] In some alternative implementations, a feasible domain is preset, including at least one of the following: The entry point is located within the operable bone surface area posterior to the target vertebral body; A safe bone passage through the target pedicle in the axial direction; The screw length shall not exceed the anterior cortical boundary; The screw diameter shall not exceed the minimum safe width of the pedicle local area minus the preset safety margin.

[0017] In the above technical solution, the preset feasible domain clearly defines the safety constraint boundary of the screw placement plan, which includes at least the following core requirements: the entry point must be limited to the posterior vertebral body within the operable bone surface area to ensure the accessibility and stability of screw placement; the axial direction must pass through the safe bone channel of the target pedicle to avoid the risk of pedicle cortical perforation; the screw length must not exceed the anterior cortical boundary to prevent the screw from penetrating the vertebral body and damaging surrounding tissues; the screw diameter must be controlled within the range of the minimum safe width of the pedicle local area minus the preset safety margin to ensure the integrity of the pedicle bone structure. This provides a clear and quantifiable safety standard for the constraint optimization of the screw placement plan, ensuring that the modified plan not only conforms to clinical operating procedures but also minimizes surgical risks.

[0018] In some optional implementations, the preoperative nail placement planning data is constrained and optimized within a preset feasible region, including: The objective function is constructed with the goal of minimizing the amount of modification and maximizing the reduction in risk as the optimization objectives: J(x) = α·C_change(x, x 0 ) + b·C_risk(x) + c·C_feasible(x) ; Wherein, the preoperative screw placement planning data is x0 = [e0, u0, l0, r0], and the corrected screw placement planning data is x = [e, u, l, r]; e0 is the original in-point coordinates; u0 is the original unit direction vector; l0 is the original screw length; r0 is the original screw radius; e is the corrected in-point coordinates; u is the corrected unit direction vector; l is the corrected screw length; r is the corrected screw radius; J(x) represents the comprehensive cost function of candidate solution x; α, β, γ represent weight coefficients; C_change(x, x0) represents the cost of modification relative to the original plan; C_risk(x) represents the residual risk cost of the corrected solution; C_feasible(x) represents the penalty cost when violating hard constraints. Within the feasible region, neighborhood sampling, grid search, and heuristic constraint optimization are performed to generate modified pinning planning data for one or more schemes where the objective function J(x) is less than a threshold.

[0019] In the above technical solution, when optimizing the preoperative nail placement planning data within the preset feasible domain, the core optimization objectives are to minimize the amount of modification to the original plan and maximize the reduction in nail placement risk. A comprehensive cost objective function is constructed, and then candidate solutions with a comprehensive cost function less than a preset threshold are selected. Finally, the corresponding corrected nail placement planning data is output, which not only achieves effective risk control but also preserves the rationality of the original plan to the greatest extent, taking into account both safety and clinical operability.

[0020] In some alternative implementations, the cost of modification is: C_change(x, x 0 ) = λ 1·||e e 0 || 2 + λ 2 ·Df + λ 3 ·|l l 0 | + λ 4 ·|r r 0 | ; Among them, ||e e0||2 represents the offset distance of the entry point; Δφ represents the change in the angle between the corrected axis and the original axis; |l l0| represents the change in length; |r r0| represents the change in radius; λ1, λ2, λ3, and λ4 represent the weighting coefficients of each modified dimension; Residual risk cost: C_risk(x) = μ 1 ·g 1 (D_min(x)) + μ 2 ·g 2 (P_pen(x)) + μ 3 ·g 3 (T) ; Where D_min(x) represents the minimum safe distance corresponding to the modified protocol; P_pen(x) represents the expected perforation amount corresponding to the modified protocol; T represents the clinical sensitivity level corresponding to the risk object type; μ1, μ2, and μ3 represent weighting coefficients; and g1(·), g2(·), and g3(·) represent normalized mapping functions.

[0021] In the above technical solution, the cost of modification is quantified and planned from four dimensions: entry point, direction, length, and radius, to ensure the convenience of surgical operation; the cost of residual risk is quantified and planned from the perspectives of safe distance, perforation risk, and tissue sensitivity, to ensure the safety of surgery.

[0022] An electronic device provided in this application includes a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the machine-readable instructions, when executed by the processor, perform any of the methods described above.

[0023] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of any of the methods described above. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart illustrating the steps of a preoperative planning and quality control method for spinal screw placement, as provided in this application embodiment; Figure 2 A schematic diagram illustrating the expected penetration amount provided in the embodiments of this application; Figure 3 This is a schematic diagram of the risk distance alarm and graded stability determination process provided in the embodiments of this application.

[0026] Some of the reference numerals in the specific implementation methods are as follows: 10 - Hyperboundary length; 20 - Hyperboundary volume; 30 - Crossing depth. Detailed Implementation

[0027] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0028] Please refer to Figure 1 , Figure 1 A flowchart illustrating the steps of a preoperative planning and quality control method for spinal screw placement, provided in this application embodiment, includes: Step 100: Generate a boundary model of the risk object based on the three-dimensional anatomical model of the vertebral body and pedicle; wherein the risk object includes at least one of the following: pedicle cortical boundary, spinal canal boundary, intervertebral foramen boundary, superior articular facet boundary; the three-dimensional anatomical model is obtained by modeling based on medical images; Specifically, preoperative three-dimensional medical imaging data of the patient's spine (e.g., DICOM CT sequence) and preoperative screw placement planning data are acquired. The planning data includes at least one of the following: the insertion point, axial direction, length, and diameter of each screw; the planning data can be derived from manual planning by the physician, imported from existing navigation planning software, or output from other automated planning modules. The three-dimensional medical imaging data is preprocessed and anatomically modeled to obtain a structural model containing at least the vertebral body and pedicles. This modeling process can employ deep learning-based segmentation models, traditional methods based on statistical shape / threshold growth, or a combination thereof. Based on the structural model, a boundary model of risk objects is generated. Risk objects include at least the pedicle cortical boundary; preferably, they may also include the spinal canal boundary, intervertebral foramen boundary, and superior articular facet boundary, for subsequent risk calculation and early warning interpretation. The generation of risk object boundaries includes the following process: First, the CT body data is standardized, denoised, anisotropically resampled, and enhanced with bone tissue; second, the anatomical structures such as the vertebral body, pedicle, spinal canal, and intervertebral foramen are segmented; finally, the segmentation results are converted into a boundary model for geometric calculation.

[0029] Step 200: Based on the boundary model and preoperative nail placement planning data, calculate the quantitative indicators related to the risk object; wherein, the quantitative indicators include at least one of the following: minimum safe distance, expected perforation amount, and expected perforation direction; the minimum safe distance represents the minimum distance between the planned channel and the risk object, and the expected perforation amount represents the value by which the planned channel exceeds the boundary of the risk object; The minimum safe distance is calculated by taking the finite planning channel corresponding to the planned screw path as the calculation object and determining the minimum distance between the planned channel and the boundary of the risk object. When the planned channel comes into contact with or intersects with the risk object, the minimum safe distance is recorded as zero. Specifically, when dragging the nail, a boundary with a color mark will be generated directly. This boundary can be made into a reverse magnetic attraction form. If it exceeds this boundary, it will be difficult to drag it out, and the distance value between the current nail tip and this boundary will be given.

[0030] Estimated breach volume: An indicator used to quantify the extent to which a planned passage exceeds the safety boundary of a risky object. The estimated breach volume includes at least one of the following: breach depth 30, over-boundary length 10, and over-boundary volume 20. Figure 2 As shown; The expected direction of penetration is used to aid in explanation and visual annotation.

[0031] Step 300: Process the quantitative indicators into structured feature vectors; crop the medical images based on the planned channel positions to obtain local image features; Step 400: Input the structured feature vector and local image features into the risk assessment model to obtain the risk assessment results.

[0032] The above technical solution constructs boundary models of key risk objects such as the pedicle cortex, spinal canal, intervertebral foramen, and superior articular facet. It then calculates quantitative indicators such as the minimum safe distance, expected perforation amount, and perforation direction based on preoperative screw placement planning data. By integrating structured feature vectors and local image features of the planned pathway into the risk assessment model, it outputs risk assessment results. Compared to existing technologies that rely on doctors manually interpreting CT images and designing based on experience, this solution effectively overcomes the shortcomings of strong subjectivity and difficulty in ensuring consistent planning quality. It can accurately identify potential screw placement safety hazards in complex cases such as pedicle morphological variations, degeneration, or deformities. Furthermore, it compensates for the lack of quantitative quality control and graded early warning mechanisms in existing planning systems, promptly alerting high-risk planning schemes and providing objective basis for path correction, significantly improving the accuracy, safety, and standardization of preoperative spinal screw placement planning.

[0033] In some alternative implementations, a boundary model of the risk object is generated based on a three-dimensional anatomical model of the vertebral body and pedicle, including: A three-dimensional medical image segmentation network is used to automatically identify the target structure. The model input is the patient's preoperative three-dimensional CT volume data, and the output is a voxel-level labeled map. The labeled map distinguishes at least the vertebral body bone region, pedicle region, spinal canal region, and intervertebral foramen region.

[0034] Based on the labeled image, triangular mesh boundaries, voxel boundaries, and implicit boundary models are generated using the Marching Cubes algorithm, boundary tracking algorithm, and distance transformation algorithm, for use in subsequent minimum distance, penetration amount, and direction calculations.

[0035] For regions with uncertain boundary identification (regions where the model has low confidence in its judgment of the boundary position), a safety margin is extended outward based on the segmentation confidence, and / or the region is marked as a manually reviewed region to improve the robustness and interpretability of the quality control results.

[0036] The boundary of the risk object is represented by a signed distance field, which is defined as follows: d_b(p) = s(p) · min_{q∈ Oh} ||p q|| 2; In the formula: d_b(p) represents the distance from point p to the boundary of the risk object. Oh The signed distance, in mm; ||p q||2 represents the Euclidean distance in mm; s(p) is a sign function, taking a positive value when located outside the boundary and a negative value when located inside the boundary. This distance field allows for the direct determination of the spatial relationship between the planned passage and the risk boundary.

[0037] In the above technical solution, the risk object boundary model is based on the patient's preoperative 3D CT volume data. It uses a 3D medical image segmentation network to automatically identify the target structure and output voxel-level label maps. Then, the Marching Cubes algorithm, boundary tracking algorithm, and distance transformation algorithm are used to generate triangular mesh boundary, voxel boundary, and implicit boundary models, respectively. At the same time, a signed distance field is used to uniformly represent the risk object boundary. The sign function accurately distinguishes the internal and external positional relationship between spatial points and the boundary, and accurately quantifies the Euclidean distance from the spatial point to the risk boundary. This provides high-precision and standardized geometric boundary support for subsequent quantitative index calculations, ensuring the accuracy and stability of risk assessment.

[0038] In some alternative implementations, a three-dimensional medical image segmentation network is used, including 3D U-Net, V-Net, nnU-Net, or Swin UNETR.

[0039] In the above technical solution, the three-dimensional medical image segmentation network can adopt mature and efficient medical image segmentation models such as 3D U-Net, V-Net, nnU-Net or Swin UNETR. While ensuring the automatic segmentation accuracy of vertebral bodies, pedicles and various risk anatomical structures, it can adapt to spinal CT body data with different resolutions and different degrees of deformity, providing a stable and reliable voxel-level segmentation basis for subsequent boundary model construction, and improving the robustness and versatility of the entire preoperative planning and quality control process.

[0040] In some optional implementations, the risk assessment model receives structured feature vectors and local image features, and outputs a risk score or risk probability. The risk assessment model is a machine learning model or a deep learning model, and the model type includes at least a classification or regression model based on structured risk quantification indicators, a deep learning model based on local image data, and a fusion model based on structured features and image features. Classification or regression models based on structured risk quantification indicators include random forest models, gradient boosting tree models, or multilayer perceptron models. Deep learning models based on local image data include two-dimensional convolutional neural networks or three-dimensional convolutional neural networks. The fusion model fuses the structured features and image features to output a risk score or risk probability.

[0041] In some optional implementations, the risk assessment model includes: an input and preprocessing module, a feature extraction module, a feature fusion module, a prediction module, and an output and postprocessing module; The input and preprocessing module is used to receive structured feature vectors and local image data, and to normalize, handle missing data and vectorize the structured feature vectors, and to resample and crop the local image data. The feature extraction module is used to extract structured features and image features; The feature fusion module is used to fuse structured features and image features to form fused features; The prediction module is used to map fused features to risk probabilities or risk scores; The output and post-processing module is used to calibrate and determine the threshold for risk probability or risk score, and generate output values ​​for determining risk assessment results.

[0042] In the above technical solution, the risk assessment model consists of an input and preprocessing module, a feature extraction module, a feature fusion module, a prediction module, and an output and post-processing module. The input and preprocessing module can perform normalization, missing value processing, and vectorization encoding on the received structured feature vectors, and perform resampling and cropping operations on local image data. The feature extraction module and the feature fusion module realize efficient extraction of structured features and image features, and multimodal feature fusion, respectively. The prediction module maps the fused features to risk probability or risk score. The output and post-processing module generates standardized risk assessment results through probability calibration and threshold determination, realizing accurate and stable automated risk assessment of spinal screw placement planning schemes.

[0043] The training process for the risk assessment model is as follows: The training samples consist of image data, risk object boundary data, planned channel parameters, a set of risk quantification indicators, and training labels. The image data is medical image data of the target anatomical region. The risk object boundary data is determined by the segmentation results. The planned channel parameters include entry point coordinates, channel direction vector, channel length, and channel diameter. The set of risk quantification indicators is generated by risk quantification indicator building units. The training labels are risk assessment result labels, including safe, low-risk, and high-risk, with risk score labels being continuous numerical values. The training labels are obtained through annotation by clinical experts or generated by a rule-based model and then corrected by clinical experts.

[0044] The training process includes sample construction, sample partitioning, model training, and model validation. Sample partitioning involves dividing the model into training, validation, and test sets based on case dimensions. During model training, structured feature vectors and local image features are used as model inputs, and risk assessment result labels are used as supervision signals. A loss function is employed to measure the difference between the model output and the supervision signals, and the model parameters are iteratively updated until the loss converges. Model validation involves calculating evaluation metrics on the validation set and selecting target model parameters. Evaluation metrics include classification accuracy, recall, AUC, or regression error. After training, the target model is used to output risk scores or risk probabilities for new planned channels.

[0045] Risk assessment results are generated based on the risk score / risk probability and preset grading rules. The grading rules include at least: threshold determination: mapping the risk score / probability or quantitative indicator to multiple levels of risk; stability determination: used to suppress frequent jumps in risk assessment results near the threshold, preferably including a hysteresis threshold strategy and a window confirmation strategy.

[0046] In some alternative implementations, preoperative screw placement planning data includes the entry point, axial direction, length, and diameter of each screw.

[0047] In some optional implementations, after obtaining the risk assessment results, the following steps are also included: If the risk assessment results are not satisfactory, the preoperative nail placement planning data is constrained and optimized within the preset feasible domain to obtain the corrected nail placement planning data.

[0048] In the above technical solution, if the risk assessment results are not up to standard, that is, if there is a high safety risk in the current screw placement plan, the preoperative screw placement plan data (including screw entry point, axis direction, length and diameter) will be constrained and optimized within the preset feasible domain. By adjusting the planning parameters, risks such as pedicle cortex perforation and spinal canal invasion are avoided, and finally the corrected screw placement plan data is output.

[0049] In some alternative implementations, a feasible domain is preset, including at least one of the following: The entry point is located within the operable bone surface area posterior to the target vertebral body; A safe bone passage through the target pedicle in the axial direction; The screw length shall not exceed the anterior cortical boundary; The screw diameter shall not exceed the minimum safe width of the pedicle local area minus the preset safety margin.

[0050] In the above technical solution, the preset feasible domain clearly defines the safety constraint boundary of the screw placement plan, which includes at least the following core requirements: the entry point must be limited to the posterior vertebral body within the operable bone surface area to ensure the accessibility and stability of screw placement; the axial direction must pass through the safe bone channel of the target pedicle to avoid the risk of pedicle cortical perforation; the screw length must not exceed the anterior cortical boundary to prevent the screw from penetrating the vertebral body and damaging surrounding tissues; the screw diameter must be controlled within the range of the minimum safe width of the pedicle local area minus the preset safety margin to ensure the integrity of the pedicle bone structure. This provides a clear and quantifiable safety standard for the constraint optimization of the screw placement plan, ensuring that the modified plan not only conforms to clinical operating procedures but also minimizes surgical risks.

[0051] In some optional implementations, the preoperative nail placement planning data is constrained and optimized within a preset feasible region, including: The objective function is constructed with the goal of minimizing the amount of modification and maximizing the reduction in risk as the optimization objectives: J(x) = α·C_change(x, x 0 ) + b·C_risk(x) + c·C_feasible(x) ; Wherein, the preoperative screw placement planning data is x0 = [e0, u0, l0, r0], and the corrected screw placement planning data is x = [e, u, l, r]; e0 is the original in-point coordinates; u0 is the original unit direction vector; l0 is the original screw length; r0 is the original screw radius; e is the corrected in-point coordinates; u is the corrected unit direction vector; l is the corrected screw length; r is the corrected screw radius; J(x) represents the comprehensive cost function of candidate solution x; α, β, γ represent weight coefficients; C_change(x, x0) represents the cost of modification relative to the original plan; C_risk(x) represents the residual risk cost of the corrected solution; C_feasible(x) represents the penalty cost when violating hard constraints. Within the feasible region, neighborhood sampling, grid search, and heuristic constraint optimization are performed to generate modified pinning planning data for one or more schemes where the objective function J(x) is less than a threshold.

[0052] In the above technical solution, when optimizing the preoperative nail placement planning data within the preset feasible domain, the core optimization objectives are to minimize the amount of modification to the original plan and maximize the reduction in nail placement risk. A comprehensive cost objective function is constructed, and then candidate solutions with a comprehensive cost function less than a preset threshold are selected. Finally, the corresponding corrected nail placement planning data is output, which not only achieves effective risk control but also preserves the rationality of the original plan to the greatest extent, taking into account both safety and clinical operability.

[0053] In some alternative implementations, the cost of modification is: C_change(x, x 0 ) = λ 1 ·||e e 0 || 2 + λ 2 ·Df + λ 3 ·|l l 0 | + λ 4 ·|r r 0 | ; Among them, ||e e0||2 represents the offset distance of the entry point; Δφ represents the change in the angle between the corrected axis and the original axis; |l l0| represents the change in length; |r r0| represents the change in radius; λ1, λ2, λ3, and λ4 represent the weighting coefficients of each modified dimension; Residual risk cost: C_risk(x) = μ 1 ·g 1 (D_min(x)) + μ 2 ·g 2 (P_pen(x)) + μ 3 ·g 3 (T) ; Where D_min(x) represents the minimum safe distance corresponding to the modified protocol; P_pen(x) represents the expected perforation amount corresponding to the modified protocol; T represents the clinical sensitivity level corresponding to the risk object type; μ1, μ2, and μ3 represent weighting coefficients; and g1(·), g2(·), and g3(·) represent normalized mapping functions.

[0054] In the above technical solution, the cost of modification is quantified and planned from four dimensions: entry point, direction, length, and radius, to ensure the convenience of surgical operation; the cost of residual risk is quantified and planned from the perspectives of safe distance, perforation risk, and tissue sensitivity, to ensure the safety of surgery.

[0055] In one specific embodiment, the steps for preoperative planning and quality control of spinal screw placement include: Step (1): Import the patient's spinal CT image data (DICOM sequence) into the preoperative planning workstation, and import or generate preoperative screw placement planning data. The planning data includes at least one of the following: the entry point, axis direction, length, and diameter of each screw.

[0056] Step (2): Call the anatomical structure modeling module to preprocess and model the CT images to obtain the vertebral body and pedicle structure model, and generate the pedicle cortical boundary model; preferably, the spinal canal and intervertebral foramen boundary model are generated at the same time.

[0057] The pedicle cortex, spinal canal, and intervertebral foramen were selected as risk targets. Patient CT scan data was input into a pre-trained 3D segmentation network to obtain multi-class label images. Subsequently, a boundary extraction algorithm was used to convert the label images into corresponding boundary models. The pedicle cortex boundary was generated from the outer contour of the bony region, while the spinal canal and intervertebral foramen boundaries were generated from the boundaries of their respective anatomical cavities. These boundary models were uniformly mapped to the patient coordinate system, facilitating spatial geometric calculations with the planned trajectory.

[0058] Step (3): Calculate the risk quantification index for the planned axis of each screw: calculate the minimum safe distance between the planned channel and the pedicle cortex boundary and / or the expected perforation amount; when risk objects such as the spinal canal / intervertebral foramen are configured, calculate the corresponding index simultaneously.

[0059] For each screw, its planned path is represented as a cylindrical region extending along the axial direction from the entry point, with a radius corresponding to the screw's radius. If the set of discrete points {p} is defined by the trajectory centerline... iLet} (i = 1, 2, …, N) represent the planned centerline, then the minimum safe distance is defined as follows: D_min = min_{i = 1, …, N} d_b(p i ) In the formula: D_min represents the minimum safe distance from the centerline of the planned channel to the boundary of the risk object, in mm; N represents the number of sampling points, a dimensionless quantity; d_b(p i ) represents the signed distance from the i-th sampling point to the boundary of the risk object, in mm. When D_min>0, the planned passage is located on the safe side; when D_min≤0, the planned passage has touched or entered the risk boundary.

[0060] The expected penetration depth is defined as the maximum depth by which the planned passage extends beyond the boundary of the risk object. P_pen = max(0, min_{i = 1, …, N} d_b(p i )) In the formula: P_pen represents the expected penetration depth in mm. When the planned channel does not cross the boundary, P_pen = 0; when the planned channel enters the interior of the risk object, P_pen is the intrusion depth. The transboundary length L_pen (mm) and transboundary volume V_pen (mm³) are also calculated to further characterize the transboundary extent.

[0061] The risk scoring uses the following comprehensive evaluation function: R = w 1 ·f 1 (D_min) + w 2 ·f 2 (P_pen) + w 3 ·f 3 (θ) + w 4 ·f 4 (T) In the formula: R represents the risk score, which is a dimensionless quantity; w1, w2, w3, and w4 represent weighting coefficients, all of which are dimensionless quantities; θ represents the angle corresponding to the penetration direction, in degrees; T represents the organizational sensitivity coefficient corresponding to the risk object type, which is a dimensionless quantity; f1(·), f2(·), f3(·), and f4(·) represent normalized mapping functions, the output of which is a dimensionless quantity. Through the above formula, the minimum safe distance, penetration amount, directional factors, and organizational sensitivity are unified and integrated into a single risk score.

[0062] Step (4): Input the above indicators into the risk assessment module to obtain the risk score or risk probability of each screw.

[0063] Step (5): Generate risk assessment results according to the preset rules of the graded early warning module. The graded rules include threshold determination; and further perform stability determination (for example, adopt a hysteresis threshold strategy to avoid repeated changes in risk assessment results near the critical threshold).

[0064] Step (6) Output graded early warning information: For screws whose risk assessment results do not meet the standards, display the name of the risk object, the risk assessment result, quantitative indicators (such as minimum distance, penetration amount) and directional prompt information, and highlight them in the three-dimensional view or tomographic view.

[0065] Step (7): Generate a quality control report from the quality control results of all screws. The report can be exported, printed or archived.

[0066] The main risk distance alarm and graded stability determination process is as follows: Figure 3 As shown.

[0067] In another specific embodiment, after step (6), the following steps are included: When the risk assessment result of a certain screw fails to meet the standard, the correction suggestion generation module is activated to perform candidate search or optimization for the screw within the feasible domain, and generate multiple correction planning schemes. The schemes include adjustments to at least one of the entry point, axis direction, length, and diameter.

[0068] First, several candidate offset points are generated within the 3D local neighborhood around the original entry point, with offset step sizes set to 0.5 mm and 1.0 mm. Simultaneously, several candidate directions are generated around the original axis, with angle step sizes set to 1° to 3°. For the length and diameter parameters, discrete sampling is performed within the allowable value range. This results in a finite number of candidate parameter combinations. Combinations that do not satisfy the feasible region constraints are then eliminated, yielding a set of candidate modified planning parameters.

[0069] Candidate solutions are ranked using the minimum modification criterion (e.g., weighted summation of entry point offset, orientation angle change, length change, and diameter change, with the minimum value given priority), and recommended solutions and corresponding expected risk reduction effects are output.

[0070] For each candidate scheme x^(k), its entry point offset distance, direction change angle, length change, diameter change, corrected minimum safe distance, and expected perforation amount are calculated, and these indicators are substituted into the comprehensive objective function J(x^(k)). Sorting J(x^(k)) from smallest to largest, the top-ranked schemes are output, along with the risk reduction value ΔR of each scheme compared to the original plan and a description of the main modifications, to facilitate rapid comparison and decision-making by physicians.

[0071] Doctors can choose to accept the recommended treatment plan or continue to make manual adjustments on the interface. They can also perform quality control and graded warnings on the updated plan to form a closed loop.

[0072] One possible structure of the electronic device provided in this application includes: a processor, a memory, and a communication interface, which are interconnected and communicate with each other via a communication bus and / or other forms of connection mechanism (not shown).

[0073] The memory includes one or more, which may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor and other possible components can access the memory, reading and / or writing data therein.

[0074] A processor may be one or more, and can be an integrated circuit chip with signal processing capabilities. The aforementioned processors can be general-purpose processors, including Central Processing Units (CPUs), Microcontroller Units (MCUs), Network Processors (NPs), or other conventional processors; they can also be special-purpose processors, including Neural-network Processing Units (NPUs), Graphics Processing Units (GPUs), Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Furthermore, when there are multiple processors, some may be general-purpose processors, while others may be special-purpose processors.

[0075] A communication interface includes one or more devices that can be used to communicate directly or indirectly with other devices to exchange data. The communication interface may include interfaces for wired and / or wireless communication.

[0076] One or more computer program instructions may be stored in the memory, and the processor may read and execute these computer program instructions to implement the methods provided in the embodiments of this application.

[0077] Electronic devices can also include more or fewer components, or different structures. Each component can be implemented using hardware, software, or a combination thereof. Electronic devices can be physical devices, such as PCs, laptops, tablets, mobile phones, servers, embedded devices, etc., or they can be virtual devices, such as virtual machines, virtualized containers, etc. Furthermore, electronic devices are not limited to a single device; they can also be a combination of multiple devices or a cluster of a large number of devices.

[0078] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of any of the methods described above.

[0079] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0080] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0082] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0083] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for quality control of preoperative planning for spinal screw placement, characterized in that, include: Based on the three-dimensional anatomical model of the vertebral body and pedicle, a boundary model of the risk object is generated; wherein, the risk object includes at least one of the following: pedicle cortical boundary, spinal canal boundary, intervertebral foramen boundary, and superior articular facet boundary; the three-dimensional anatomical model is obtained by modeling based on medical images; Based on the boundary model and preoperative nail placement planning data, quantitative indicators related to the risk object are calculated; wherein, the quantitative indicators include at least one of the following: minimum safe distance, expected perforation amount, and expected perforation direction; the minimum safe distance represents the minimum distance between the planned channel and the risk object, and the expected perforation amount represents the value by which the planned channel exceeds the boundary of the risk object; The quantitative indicators are processed into structured feature vectors; The medical images are cropped based on the planned channel locations to obtain local image features; The structured feature vectors and local image features are input into the risk assessment model to obtain the risk assessment results.

2. The method as described in claim 1, characterized in that, The process of generating a boundary model of the risk object based on the three-dimensional anatomical model of the vertebral body and pedicle includes: A three-dimensional medical image segmentation network is used to automatically identify targets in a three-dimensional anatomical structure model, thereby obtaining a boundary model of the risk object.

3. The method as described in claim 2, characterized in that, The input to the three-dimensional medical image segmentation network is the patient's preoperative three-dimensional CT volume data, and the output of the three-dimensional medical image segmentation network is a voxel-level label map; Based on the labeled image, triangular mesh boundaries, voxel boundaries, and implicit boundary models are generated using the Marching Cubes algorithm, boundary tracking algorithm, and distance transformation algorithm. The boundary of the risk object is represented using a signed distance field. d_b(p) = s(p) · min_{q∈ Ω} ||p q|| 2 Where d_b(p) represents the distance from point p to the boundary of the risk object. Ω The signed distance; ||p q||2 represents the Euclidean distance; s(p) is the sign function, which takes a positive value when it is outside the boundary and a negative value when it is inside the boundary.

4. The method as described in claim 1, characterized in that, The risk assessment model includes: an input and preprocessing module, a feature extraction module, a feature fusion module, a prediction module, and an output and postprocessing module; The input and preprocessing module is used to receive structured feature vectors and local image data, and to normalize, handle missing data and vectorize the structured feature vectors, and to resample and crop the local image data. The feature extraction module is used to extract structured features and image features; The feature fusion module is used to fuse structured features and image features to form fused features; The prediction module is used to map the fused features to risk probabilities or risk scores. The output and post-processing module is used to calibrate and determine the threshold of the risk probability or risk score, and generate output values ​​for determining the risk assessment results.

5. The method as described in claim 1, characterized in that, After obtaining the risk assessment results, the process also includes: If the risk assessment results are not satisfactory, the preoperative nail placement planning data is constrained and optimized within the preset feasible domain to obtain the corrected nail placement planning data.

6. The method as described in claim 5, characterized in that, The preset feasible domain includes at least one of the following: The entry point is located within the operable bone surface area posterior to the target vertebral body; A safe bone passage through the target pedicle in the axial direction; The screw length shall not exceed the anterior cortical boundary; The screw diameter shall not exceed the minimum safe width of the pedicle local area minus the preset safety margin.

7. The method as described in claim 6, characterized in that, The constraint optimization of preoperative nail placement planning data within a preset feasible domain includes: The objective function is constructed with the goal of minimizing the amount of modification and maximizing the reduction in risk as the optimization objectives: J(x) = α·C_change(x, x 0 ) + β·C_risk(x) + γ·C_feasible(x) ； Wherein, the preoperative screw placement planning data is x0 = [e0, u0, l0, r0], and the corrected screw placement planning data is x = [e, u, l, r]; e0 is the original in-point coordinates; u0 is the original unit direction vector; l0 is the original screw length; r0 is the original screw radius; e is the corrected in-point coordinates; u is the corrected unit direction vector; l is the corrected screw length; r is the corrected screw radius; J(x) represents the comprehensive cost function of candidate solution x; α, β, γ represent weight coefficients; C_change(x, x0) represents the cost of modification relative to the original plan; C_risk(x) represents the residual risk cost of the corrected solution; C_feasible(x) represents the penalty cost when violating hard constraints. Within the feasible region, neighborhood sampling, grid search, and heuristic constraint optimization are performed to generate modified pinning planning data for one or more schemes where the objective function J(x) is less than a threshold.

8. The method as described in claim 7, characterized in that, The cost of the changes: C_change(x, x 0 ) = λ 1 ·||e e 0 || 2 + λ 2 ·Δφ + λ 3 ·|l l 0 | + λ 4 ·|r r 0 | ； Among them, ||e e0||2 represents the offset distance of the entry point; Δφ represents the change in the angle between the corrected axis and the original axis; |l l0| represents the change in length; |r r0| represents the change in radius; λ1, λ2, λ3, and λ4 represent the weight coefficients of each modified dimension; The cost of the residual risk: C_risk(x) = μ 1 ·g 1 (D_min(x)) + μ 2 ·g 2 (P_pen(x)) + μ 3 ·g 3 (T) ； Where D_min(x) represents the minimum safe distance corresponding to the modified protocol; P_pen(x) represents the expected perforation amount corresponding to the modified protocol; T represents the clinical sensitivity level corresponding to the risk object type; μ1, μ2, and μ3 represent weighting coefficients; and g1(·), g2(·), and g3(·) represent normalized mapping functions.

9. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when executed by the processor, perform the method as described in any one of claims 1-8.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-8.

Images