Path planning method and system for performing vertebral puncture using an orthopedic surgical robot

Abstract

The present application relates to the technical field of vertebral puncture, and particularly relates to a vertebral puncture path planning method and system using an orthopedic surgery machine, which acquires a multi-modal image parameter set of a patient's vertebral region in real time, performs three-dimensional structure feature clustering processing, and generates a vertebral structure clustering feature set. The clustering feature set is input into an adaptive spatial planning network model to dynamically predict and construct a puncture path evolution trend sequence. Based on the sequence, a risk critical region is identified, a collision warning signal is triggered and a dynamic response is monitored to obtain a path avoidance response behavior data set. Through multi-dimensional coupling analysis of the response data and the evolution trend, a high-risk anatomical region is determined to form a regional risk degree identification data set. Finally, according to the identification data, a priority is divided and adjusted to generate an adaptive path adjustment instruction set, and a precise and safe vertebral puncture path planning result is output.

Description

Technical Field

[0001] This invention relates to the field of vertebral puncture technology, and in particular to a method and system for path planning of vertebral puncture using orthopedic surgical equipment. Background Technology

[0002] With the rapid development of minimally invasive spinal surgery techniques, vertebral puncture has been widely used in the treatment of vertebral compression fractures, vertebral tumors, and vertebroplasty. Traditional puncture procedures rely heavily on the surgeon's experience and hand-eye coordination, posing risks such as low positioning accuracy, long radiation exposure time, and accidental puncture of important nerves or blood vessels. The difficulty of the procedure increases significantly, especially in patients with complex anatomy or osteoporosis. In recent years, the introduction of orthopedic surgical robotic systems has provided a new solution to improve puncture accuracy and surgical safety. However, existing robot-assisted puncture systems are mostly based on preoperative static images for path planning, making it difficult to respond in real time to dynamic anatomical shifts caused by changes in patient position, tissue deformation, or instrument interaction during surgery. This leads to a mismatch between the preset path and the actual anatomical environment, increasing the risk of puncture.

[0003] Furthermore, current path planning methods generally lack the ability to fine-tune modeling and risk prediction of vertebral internal structural heterogeneity (such as uneven bone density distribution and areas of cortical bone weakness), making it difficult to achieve personalized safety path optimization. Although some systems have introduced intraoperative image update mechanisms, their path adjustment strategies are mostly passive corrections, lacking forward-looking prediction and proactive avoidance mechanisms for potential risk areas, resulting in delayed response and insufficient adjustments.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a path planning method and system for vertebral puncture using orthopedic surgical robots. This aims to solve the technical problem that existing orthopedic surgical robot vertebral puncture path planning methods are unable to integrate multimodal images to achieve real-time perception and risk prediction of vertebral structural heterogeneity and dynamic anatomical changes, resulting in insufficient path safety and adaptability.

[0006] To achieve the above objectives, the present invention provides a path planning method for vertebral puncture using an orthopedic surgical machine, the method comprising: A multimodal image parameter set of the patient's vertebral body region is acquired in real time, and the image parameter set is subjected to three-dimensional structural feature clustering processing to generate a vertebral body structure clustering feature set; The clustering feature set of the vertebral structure is input into an adaptive spatial planning network model to dynamically predict the evolution trend of the puncture path at key anatomical nodes and construct a path evolution trend sequence. Based on the path evolution trend sequence, risk critical regions in the puncture path are identified, and dynamic response monitoring is performed on the collision warning signal triggered in the critical regions to obtain a path avoidance response behavior dataset. A multi-dimensional coupling analysis is performed on the path avoidance response behavior dataset and the path evolution trend sequence to identify high-risk areas and form a regional risk level identification dataset. The path planning priority is divided into gradients based on the regional risk level identification dataset. An adaptive path adjustment instruction set is generated according to the priority, and the path planning result of the orthopedic surgical robot vertebral puncture is output.

[0007] Optionally, the vertebral body structure clustering feature set includes vertebral body density distribution curves, anatomical structure coupling strength grouping identifiers, and three-dimensional anatomical node topological relationships; the path evolution trend sequence includes path direction angle time series, puncture probability density function, and anatomical node avoidance continuity features; the path avoidance response behavior dataset includes path deviation peak amplitude, early warning response lag time features, and avoidance direction vector; the regional risk level identifier dataset includes high-risk anatomical node numbers, local path deviation index, and structural stability failure markers; and the orthopedic surgical robot vertebral puncture path planning results include path adjustment command type, robotic arm execution parameters, and target avoidance area coordinates.

[0008] Optionally, the steps for obtaining the cone structure clustering feature set are as follows: The bone density distribution, anatomical node spacing, and structural coupling stability of the patient's vertebral region are acquired in real time using multimodal imaging equipment to obtain the original set of image parameters; Based on the bone density data in the original parameter set of the image, the structural coupling coefficient between adjacent anatomical nodes is calculated, and clustering is performed according to the coefficient similarity threshold to generate anatomical structural coupling group identifiers. The bone density distribution characteristics, coupling stability, and node spacing fluctuation range of each group in the anatomical structure coupling group identifier are extracted to construct a vertebral structure cluster feature set.

[0009] Optionally, the steps for obtaining the path evolution trend sequence are as follows: The vertebral structure cluster feature set is discretized according to anatomical spatial coordinates, and spatial coding and avoidance trend identifier are assigned to each anatomical node to generate the puncture path state evolution time sequence. Based on the puncture path state evolution time series, an adaptive spatial planning network model is used to dynamically model the path direction angle change rate and path dwell time between nodes, and output the path state transition density function. Based on the path state transition density function, the highest probability safe path is selected as the main trend path, the continuous change sequence of direction angle in the path is extracted, and the path evolution trend sequence is generated by combining the dissected node number.

[0010] Optionally, the steps for obtaining the path avoidance response behavior dataset are as follows: Based on the path evolution trend sequence, the second derivative of the puncture path direction angle is calculated, and critical regions exceeding the preset risk threshold are identified and integrated into path risk critical regions. For the critical risk region of the path, a pulsed avoidance control signal is applied to the robot arm, and the time difference between the signal triggering time and the robot arm response start time is measured to generate a dynamic response time series. Record the maximum path offset, the dynamic trajectory of the avoidance direction, and the variance of the response process in the dynamic response time series to obtain a path avoidance response behavior dataset.

[0011] Optionally, the steps for obtaining the regional risk level identification dataset are as follows: Align the path avoidance response behavior dataset with the path evolution trend sequence according to the dissected node number, calculate the time offset between the path offset duration and the path state transition cycle, and generate node response offset features. Based on the node response offset characteristics, kernel density estimation is used to analyze the distribution dispersion of each node response offset. A risk identification threshold is set according to the dispersion confidence interval, and candidate nodes for high-risk areas are output. Based on the candidate nodes in the high-risk areas, the node numbers, corresponding anatomical spatial coordinates, and risk level indexes are extracted. The risk node distribution density is aggregated by vertebral anatomical region, and the structural risk index of each region is marked to establish a regional risk level identification dataset.

[0012] Optionally, the steps for obtaining the vertebral puncture path planning results of the orthopedic surgical robot are as follows: Based on the aforementioned regional risk level identification dataset, the weighted comprehensive risk level score is calculated by integrating two core indicators: the path direction angle offset and the offset magnitude of the dissected nodes. Based on the weighted comprehensive risk score, the mean and coefficient of variation of the global score distribution are calculated, and the planning priority labels are divided according to the risk gradient interval to generate the path planning priority distribution. Based on the path planning priority distribution, the anatomical node coordinates and priority labels corresponding to each priority are mapped to the robot control system to generate a gradient path adjustment instruction set and output the vertebral puncture path planning result of the orthopedic surgical robot.

[0013] Furthermore, to achieve the above objectives, the present invention also provides a path planning system for vertebral puncture using an orthopedic surgical machine, the system comprising: The image clustering module is used to acquire multimodal image parameter sets of the patient's vertebral body region in real time, and perform three-dimensional structural feature clustering processing on the image parameter sets to generate a vertebral body structure clustering feature set. The model prediction module is used to input the clustering feature set of the vertebral structure into the adaptive spatial planning network model, dynamically predict the evolution trend of the puncture path of key anatomical nodes, and construct the path evolution trend sequence. The risk warning module is used to identify the risk critical area in the puncture path based on the path evolution trend sequence, perform dynamic response monitoring on the collision warning signal triggered in the critical area, and obtain a path avoidance response behavior dataset. The coupling analysis module is used to perform multi-dimensional coupling analysis on the path avoidance response behavior dataset and the path evolution trend sequence, identify high-risk dissection areas, and form a regional risk level identification dataset. The path adjustment module is used to perform gradient division of path planning priorities based on the regional risk level identification dataset, generate an adaptive path adjustment instruction set according to the priorities, and output the path planning results of the orthopedic surgical robot vertebral puncture.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a path planning device for vertebral puncture using an orthopedic surgical machine. The device includes: a memory, a processor, and a path planning program for vertebral puncture using an orthopedic surgical machine stored in the memory and executable on the processor. The path planning program for vertebral puncture using an orthopedic surgical machine is configured to implement the steps of the path planning method for vertebral puncture using an orthopedic surgical machine as described above.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a path planning program for vertebral puncture using an orthopedic surgical machine. When the path planning program for vertebral puncture using an orthopedic surgical machine is executed by a processor, it implements the steps of the path planning method for vertebral puncture using an orthopedic surgical machine as described above.

[0016] This invention provides a path planning method for vertebral puncture using orthopedic surgical equipment. The method achieves refined modeling of the anatomical heterogeneity within the vertebral body by fusing multimodal imaging data and performing three-dimensional structural feature clustering. It dynamically predicts the evolution trend of the puncture path using an adaptive spatial planning network model, improving the foresight and personalization of the path planning. By identifying critical risk areas and monitoring path avoidance response behavior, a closed-loop control mechanism of prediction, early warning, response, and evaluation is constructed, enhancing the system's real-time adaptability to dynamic anatomical changes. Furthermore, through multi-dimensional coupling analysis, regional risk indicators and gradient control instructions are generated, enabling accurate identification and priority intervention of high-risk areas. This significantly improves the safety, accuracy, and intelligence of vertebral puncture path planning, effectively reducing surgical risks such as neurovascular injury, and has good clinical application value and promising prospects for wider application. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the path planning method for vertebral puncture using an orthopedic surgical machine according to the present invention. Figure 2 This is a structural block diagram of an embodiment of the path planning system for vertebral puncture using an orthopedic surgical machine according to the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the path planning method for vertebral puncture using an orthopedic surgical machine according to the present invention.

[0021] In one embodiment, the path planning method for vertebral puncture using an orthopedic surgical machine includes: Step S100: Real-time acquisition of multimodal image parameter set of the patient's vertebral body region, and three-dimensional structural feature clustering processing of the image parameter set to generate vertebral body structure clustering feature set.

[0022] The vertebral body region can be the local spatial area of ​​the vertebra to be punctured and its surrounding anatomical structures within the human spine. It can serve as the target anatomical area for path planning and puncture operation, providing the spatial scope for image acquisition and risk assessment. The multimodal image parameter set can be a complementary set of image data about the patient's vertebral body region acquired from two or more medical imaging modalities (such as CT, MRI, X-ray, and ultrasound). Furthermore, the multimodal image parameter set can be acquired synchronously or asynchronously using intraoperative or preoperative multimodal imaging equipment and registered and fused to form parametric data in a unified coordinate system. This provides multidimensional anatomical information such as vertebral bone density, soft tissue boundaries, and the course of blood vessels and nerves, supporting heterogeneous modeling. For example, the multimodal image parameter set may include, but is not limited to, one or more subsets of CT image parameters, MRI image parameters, and intraoperative ultrasound image parameters.

[0023] Three-dimensional structural feature clustering can be an unsupervised or semi-supervised clustering process that uses voxel or point cloud data from a multimodal image parameter set to perform clustering based on structural similarity (such as density, texture, and boundary curvature). This can be used to divide a continuous anatomical space into sub-regions with homogeneous structural characteristics, revealing the heterogeneous distribution within the vertebral body. In an exemplary embodiment, three-dimensional structural feature clustering can be achieved through density-texture joint clustering based on K-means++ or boundary-aware region segmentation based on graph cut. The vertebral body structural clustering feature set can be a labeled three-dimensional dataset representing the structural attributes of different anatomical sub-regions within the vertebral body, generated after three-dimensional structural feature clustering. Furthermore, the vertebral body structural clustering feature set can be output by a clustering algorithm, which assigns a category label and corresponding feature vector (such as mean density, variance, and gradient direction) to each spatial location. This provides a refined semantic map of the anatomical environment for subsequent path prediction, distinguishing structures such as cortical bone, cancellous bone, and weak areas. For example, the vertebral body structure cluster feature set may include, but is not limited to, one or more of the following: high-density bone region feature subset, low-density bone region feature subset, and cortical bone thin area feature subset.

[0024] Real-time acquisition of multimodal image parameters of the patient's vertebral body region can be achieved by simultaneously or sequentially acquiring various image data of the patient's vertebral body region using multiple imaging devices during or before surgery. Furthermore, real-time acquisition of multimodal image parameters of the patient's vertebral body region can be achieved through combined acquisition of intraoperative C-arm CT and optical navigation systems, or by registration and fusion of preoperative high-resolution CT and intraoperative ultrasound images, thereby obtaining a dynamic image foundation encompassing multidimensional anatomical information such as bone density, soft tissue, and blood vessels. Three-dimensional structural feature clustering processing of the image parameter set can be performed by mapping the multimodal image parameter set to a unified three-dimensional space and using clustering algorithms to group voxels or point clouds based on structural similarity. In a specific embodiment, three-dimensional structural feature clustering processing of the image parameter set can be achieved through density-texture joint clustering based on K-means++ or boundary-aware region segmentation based on graph cut (GraphCut), thereby enabling semantic analysis of the anatomical heterogeneity within the vertebral body. Generating a cluster feature set of vertebral structures can be achieved by encoding the clustering results into a labeled 3D feature map, with each spatial location accompanied by a structural category and statistical features, thereby enabling the construction of a refined anatomical environment model that can be used for path prediction.

[0025] Step S200: Input the vertebral structure clustering feature set into the adaptive spatial planning network model to dynamically predict the evolution trend of the puncture path of key anatomical nodes and construct the path evolution trend sequence.

[0026] The adaptive spatial planning network model can be a deep learning-based temporal-spatial joint prediction model capable of dynamically adjusting the puncture path evolution prediction strategy based on the current anatomical state. Furthermore, the adaptive spatial planning network model can be trained end-to-end using historical puncture trajectories, anatomical deformation data, and successful / failed cases, possessing online fine-tuning capabilities. This allows for dynamic and personalized prediction of the future trajectory of the puncture path near key anatomical nodes. For example, the adaptive spatial planning network model may include, but is not limited to, one or more of the following: a graph neural network path prediction module, a recurrent neural network temporal evolution module, and an attention mechanism spatial focusing module. In an exemplary embodiment, the adaptive spatial planning network model can receive a vertebral structure clustering feature set as input and output a path evolution trend sequence; its prediction results are adjusted by feedback from a path avoidance response behavior dataset. Specifically, the adaptive spatial planning network model can adopt a deep neural network architecture based on an encoder-decoder structure, including a feature extraction module composed of multiple three-dimensional convolutional layers, a long short-term memory network layer for capturing the spatiotemporal dependencies of the path, and a fully connected output layer for generating continuous motion parameters. Its input dimension matches the feature vector dimension of the vertebral structure cluster feature set, and the output dimension corresponds to the spatial coordinate sequence of key anatomical nodes and the path evolution probability distribution. The model training method adopts a strategy combining supervised learning based on historical surgical datasets and reinforcement learning based on simulated environments. It uses the Adam optimizer to minimize the mean square error between the actual trajectory of the predicted path and the expert demonstration trajectory, and simultaneously introduces a simulated surgical collision penalty as a loss function term to continuously iteratively optimize the network parameters. Its core evaluation logic lies in comprehensively calculating the multi-objective weighted score of the minimum safe distance between the predicted path and the organs at risk, the path curvature smoothness index, and the predicted value of bone material resistance, aiming to screen out the optimal puncture path evolution trend that maximizes the surgical safety boundary under the premise of minimally invasive constraints.

[0027] Key anatomical nodes can be anatomical locations adjacent to important nerves, blood vessels, or bony landmarks along the vertebral puncture path, such as the medial wall of the pedicle or the entrance to the neural foramen. These can be used as key areas of focus for path evolution prediction, determining the puncture safety boundaries and angle tolerances. The puncture path evolution trend can be a dynamic pattern of path deviation direction and magnitude that the puncture instrument may undergo due to tissue interaction or deformation under a given initial puncture direction and current anatomical environment. This can be used to reflect the potential trajectory of the path in the dynamic intraoperative environment for prospective risk assessment. The path evolution trend sequence can be an ordered set of puncture path evolution trends output by an adaptive spatial planning network model, arranged by time or spatial step size. This can be used to provide the probability distribution of future multi-step path evolution, supporting the identification of critical risk areas. For example, the path evolution trend sequence may include, but is not limited to, one or more of short-term, medium-term, and long-term evolution subsequences.

[0028] Inputting the vertebral structure clustering feature set into an adaptive spatial planning network model can be achieved by using a 3D feature map as the network input tensor, activating the model's internal spatial-temporal reasoning mechanism, thereby initiating personalized path evolution prediction based on the current anatomical state. Dynamic prediction of the puncture path evolution trend at key anatomical nodes can involve the model simulating multiple possible path deviation scenarios within the neighborhood of the key anatomical node and outputting a probability distribution. Furthermore, dynamic prediction of the puncture path evolution trend at key anatomical nodes can be achieved through multi-trajectory prediction based on Monte Carlo sampling or continuous path evolution modeling based on deterministic differential equations, thus allowing for advance prediction of the path's potential trajectory in a dynamic environment. Constructing a path evolution trend sequence can be achieved by sorting the dynamic prediction results by time or spatial steps, forming an ordered set of path evolutions, thereby providing temporal input for identifying critical risk areas.

[0029] Step S300: Based on the path evolution trend sequence, identify the risk critical area in the puncture path, perform dynamic response monitoring on the collision warning signal triggered in the critical area, and obtain a path avoidance response behavior dataset.

[0030] The critical risk region in the puncture path can be a transitional area in the path evolution trend sequence before entering a high-risk anatomical structure (such as the boundary of nerve, blood vessel, or cortical bone fracture). It can serve as a warning trigger point to activate the dynamic response monitoring mechanism to avoid actual invasion of the high-risk area. For example, the critical risk region in the puncture path can include, but is not limited to, one or more of the following: a nerve-adjacent critical region, a blood vessel-adjacent critical region, or a cortical bone fracture critical region. The collision warning signal can be a logical or numerical warning symbol generated by the system when the path evolution trend approaches the critical risk region. It can be used to activate the dynamic response monitoring module, initiate avoidance behavior simulation, or adjust the process in real time. Dynamic response monitoring can be the process of real-time capture and recording of the avoidance actions (such as angle fine-tuning or needle insertion pause) performed by the puncture instrument after receiving the warning signal. This data can be used to generate behavioral data reflecting the effectiveness of the system's response for closed-loop evaluation and replanning.

[0031] The path avoidance response behavior dataset can be a collection of multi-dimensional response behavior records, including puncture instrument pose adjustment, force feedback changes, and time delays, collected during dynamic response monitoring. It can be used as input for multi-dimensional coupled analysis to evaluate the effectiveness and residual risk of avoidance strategies. For example, the path avoidance response behavior dataset may include, but is not limited to, subsets of pose adjustment behavior, force feedback behavior, and time response behavior. Identifying the critical risk region in the puncture path based on the path evolution trend sequence can detect the transition region before the first entry into a high-risk boundary in the path evolution trend. Furthermore, identifying the critical risk region in the puncture path based on the path evolution trend sequence can be achieved through critical point detection based on distance field thresholds or boundary identification based on risk gradient mutations, thereby enabling prospective localization of risk areas.

[0032] Dynamic response monitoring of collision warning signals triggered in critical areas can involve the system issuing a warning and initiating real-time monitoring of the robot's avoidance behavior when a risky critical area is identified. In one specific embodiment, dynamic response monitoring of collision warning signals triggered in critical areas can be achieved by executing avoidance actions and recording responses in a virtual simulation environment, or by monitoring actual responses through force sensors and pose feedback during real punctures, thereby establishing a closed-loop behavior from warning to response. Obtaining a path avoidance response behavior dataset can involve collecting and structured storage of multi-dimensional response data such as robot pose, force perception, and time during the avoidance process, thus providing empirical behavioral evidence for multi-dimensional coupled analysis.

[0033] Step S400: Perform multi-dimensional coupling analysis on the path avoidance response behavior dataset and the path evolution trend sequence to identify high-risk areas and form a regional risk level identification dataset.

[0034] Multidimensional coupling analysis can be an analytical method that correlates path avoidance response behavior datasets with path evolution trend sequences across dimensions such as spatial location, temporal phase, and mechanical interaction. It can be used to comprehensively assess the actual risk level of each anatomical region, transcending the limitations of single imaging or prediction dimensions. Furthermore, multidimensional coupling analysis can be achieved by constructing a spatiotemporal-mechanical tensor and performing principal component analysis, or by using graph neural networks to fuse multi-source heterogeneous data for risk scoring. High-risk anatomical regions can be local vertebral areas identified by multidimensional coupling analysis as having a high probability of neurovascular injury or bone structure damage. These can be used as core judgment objects for regional risk identification, guiding path priority allocation. Regional risk identification datasets can be three-dimensional labeled data sets that assign continuous or graded risk values ​​to various spatial locations of the vertebral body. They can be used to quantify the puncture risk of different regions, supporting gradient path control. For example, regional risk identification datasets can include, but are not limited to, one or more subsets of nerve injury risk, vascular puncture risk, and bone structure failure risk.

[0035] Multi-dimensional coupling analysis of path avoidance response behavior datasets and path evolution trend sequences can be performed by modeling the correlation and quantifying the risks of the two sets of data in spatial, temporal, and mechanical dimensions. Furthermore, this multi-dimensional coupling analysis can be achieved by constructing a spatiotemporal-mechanical tensor and performing principal component analysis, or by using graph neural networks to fuse multi-source heterogeneous data for risk scoring, thus enabling multi-dimensional fusion and objective quantification of risk assessment. Identifying high-risk anatomical regions can be done by identifying anatomical regions with risk levels exceeding a preset threshold or relatively significant risks based on the coupling analysis results, thereby clarifying high-risk locations requiring priority intervention. Creating a regional risk level label dataset can be achieved by labeling the judgment results in a three-dimensional anatomical space as continuous numerical values ​​or hierarchical labels, thereby generating a risk map that can be used for path regulation.

[0036] Step S500: Based on the regional risk level identification dataset, the path planning priority is divided into gradients, an adaptive path adjustment instruction set is generated according to the priority, and the path planning result of the orthopedic surgical robot vertebral puncture is output.

[0037] The path planning priority can be an intervention priority order assigned to each segment or candidate path in the puncture path based on the regional risk level identifier. This can be used to determine the order of path adjustments, ensuring that high-risk areas are avoided or optimized first. Gradient partitioning can be a strategy that divides the path planning priority into multiple non-discrete levels based on the continuous numerical distribution of the regional risk level identifier dataset. This can be used to achieve fine-grained risk response and avoid overly conservative or aggressive decisions caused by binary (safe / dangerous) decision-making. Furthermore, gradient partitioning can be achieved through segmented priority mapping based on risk percentiles or priority weight allocation based on a continuous risk gradient function. The adaptive path adjustment instruction set can be a set of specific control commands generated based on gradient priority to drive the orthopedic surgical robot to perform puncture path correction. This can be used to convert risk assessment results into executable robot motion commands to complete closed-loop control. For example, the adaptive path adjustment instruction set may include, but is not limited to, one or more of the following: a subset of angle fine-tuning instructions, a subset of needle insertion depth limitation instructions, and a subset of path replanning trigger instructions. The vertebral puncture path planning result of the orthopedic surgical robot can be the final puncture path scheme output after the above-mentioned full process, including the starting point, target point, intermediate trajectory and risk avoidance strategy, which can be used to directly guide the orthopedic surgical robot to perform safe and accurate vertebral puncture operations.

[0038] Prioritizing path planning based on a regional risk level dataset can be done by dividing path adjustment requirements into multiple priority levels according to the magnitude of the risk level. Furthermore, this priority prioritization can be achieved through segmented priority mapping based on risk percentiles or priority weight allocation based on a continuous risk gradient function, enabling refined hierarchical control of risk response. Generating an adaptive path adjustment instruction set based on priority can translate priority information into specific robot control commands, such as angle corrections and needle insertion speed limits. In an exemplary embodiment, generating the adaptive path adjustment instruction set based on priority can be achieved by generating joint space commands through inverse kinematics or by selecting the optimal avoidance action sequence based on a reinforcement learning strategy, thus completing the transformation from risk assessment to execution control. Outputting the vertebral puncture path planning results for the orthopedic surgical robot can be done by outputting the final path scheme in a standard format (such as DICOM-RT or a robot-specific protocol) to the surgical robot control system, thereby providing a directly executable and safe puncture path.

[0039] For example, in the scenario of vertebroplasty for osteoporotic patients, the path planning method for vertebral puncture using an orthopedic surgical robot in this embodiment can be as follows: Intraoperative C-arm CT and ultrasound are used to acquire a multimodal image parameter set of the patient's L1 vertebral region. After three-dimensional structural feature clustering, a weak cortical bone area is identified on the anterior wall of the vertebral body. This vertebral structure clustering feature set is input into an adaptive spatial planning network model, predicting that the puncture path tends to deviate towards the neural foramen when approaching the medial wall of the pedicle. The system identifies that this deviated path is about to enter the nerve-adjacent critical zone, triggering a collision warning signal. An angle fine-tuning response is simulated in a virtual environment, and pose and force feedback data are recorded to form a path avoidance response behavior dataset. Through multi-dimensional coupling analysis, it is found that the risk of nerve injury in this area is significantly higher than in other areas, generating a high-risk marker. Based on this, the path planning priority is gradient-divided, prioritizing the restriction of the puncture angle and deviating it outward by 2°. Finally, an adaptive path adjustment instruction set is generated to guide the robot to safely complete the puncture.

[0040] In one embodiment, the vertebral body structure clustering feature set includes vertebral body density distribution curves, anatomical structure coupling strength grouping identifiers, and three-dimensional anatomical node topological relationships. The path evolution trend sequence includes path direction angle time series, puncture probability density function, and anatomical node avoidance continuity features. The path avoidance response behavior dataset includes path deviation peak amplitude, early warning response lag time features, and avoidance direction vector. The regional risk level identifier dataset includes high-risk anatomical node numbers, local path deviation index, and structural stability failure markers. The orthopedic surgical robot vertebral puncture path planning results include path adjustment command type, robotic arm execution parameters, and target avoidance area coordinates.

[0041] The vertebral body structure clustering feature set can be a structured 3D feature set generated after 3D structural feature clustering processing, containing vertebral body density distribution curves, anatomical structure coupling strength grouping identifiers, and 3D anatomical node topological relationships. This set can provide a multidimensional semantic expression of the vertebral body's internal heterogeneity, supporting path prediction models in understanding bone density gradients, structural connectivity strength, and spatial topological constraints. In an exemplary embodiment, the vertebral body structure clustering feature set can be obtained by performing 3D structural feature clustering processing on an image parameter set. Further, this processing can employ spectral clustering combined with graph neural networks to extract the topological relationships between 3D anatomical nodes, or utilize a Gaussian mixture model to fit the bone density distribution and generate a continuous density curve, thereby achieving refined, multidimensional modeling of the vertebral body's internal structural heterogeneity, surpassing the traditional homogeneity assumption. For example, the vertebral body structure clustering feature set can include, but is not limited to, one or more of the following: vertebral body density distribution curves, anatomical structure coupling strength grouping identifiers, and 3D anatomical node topological relationships. The path evolution trend sequence can be a multi-dimensional temporal feature set describing the future direction of the puncture path, generated by an adaptive spatial planning network model. It can provide dynamic evolutionary basis for identifying critical risk areas and support modeling nonlinear and nondeterministic path deviations. In one specific embodiment, the path evolution trend sequence can be obtained by inputting a vertebral structure clustering feature set into an adaptive spatial planning network model to dynamically predict the puncture path evolution trend of key anatomical nodes. Furthermore, this prediction can be achieved by modeling the dynamic influence between anatomical nodes using a spatiotemporal graph convolutional network, or by combining a variational autoencoder to generate the uncertainty interval of the puncture probability density function, thereby realizing a forward-looking, probabilistic prediction of the path deviation trend and supporting personalized path evolution modeling. For example, the path evolution trend sequence may include, but is not limited to, one or more of the following: path orientation angle temporal sequence, puncture probability density function, and anatomical node avoidance continuity features.

[0042] The critical risk region can be a transitional area identified in the path evolution trend sequence, located between the safety boundary and high-risk anatomical structures. It can serve as an early warning trigger point to activate an avoidance response mechanism and prevent actual entry into the high-risk area. In an exemplary embodiment, the critical risk region can be obtained by identifying the critical risk region in the puncture path based on the path evolution trend sequence. Furthermore, this identification can be based on gradient mutation detection of the local path deviation index, or by using structural stability failure markers to inversely deduce the critical entry point, thereby establishing a proactive early warning mechanism to avoid passive correction and improve the system's proactive response. For example, the critical risk region can include one or more of the following: the cortical bone thinning-neural foramen transition zone, the critical deviation region of the medial wall of the pedicle, and the pre-risk zone for vascular perforation near the endplate. The regional risk label dataset can be a quantitative risk label set generated through multi-dimensional coupling analysis, containing high-risk anatomical node numbers, local path deviation indices, and structural stability failure markers. This can be used to upgrade risk assessment from binary judgment to continuous gradient expression, supporting refined priority control. In one specific embodiment, the regional risk level identification dataset can be obtained by performing multi-dimensional coupled analysis of path avoidance response behavior datasets and path evolution trend sequences to determine high-risk dissection areas. Furthermore, this analysis can construct a tensor fusion model to calculate a local path deviation index, or analyze the correlation strength between avoidance behavior and structural failure markers through causal reasoning, thereby generating regional risk level identifiers containing fine-grained indicators such as structural stability failure markers, achieving an upgraded risk quantification. For example, the regional risk level identification dataset may include, but is not limited to, one or more of the following: high-risk dissection node numbers, local path deviation indices, and structural stability failure markers.

[0043] Based on a regional risk level dataset, path planning priorities can be prioritized using a gradient approach. This can be achieved by dividing path adjustment needs into multiple priority levels based on continuous risk values ​​such as high-risk node numbers and local path deviation indices. Furthermore, this prioritization can generate discrete priority levels based on risk quantile mapping, or use a soft threshold function to generate continuous priority weights for the path optimization objective function. This allows for differentiated, priority-driven intervention in high-risk areas, avoiding a one-size-fits-all approach to risk mitigation.

[0044] Taking vertebral tumor biopsy as an example, the path planning method for vertebral puncture using orthopedic surgical equipment in this embodiment can be as follows: Intraoperative CT and MRI fused images are acquired, and a vertebral body structure clustering feature set is generated through three-dimensional structural feature clustering. This feature set includes vertebral body density distribution curves (showing low density in the tumor area), anatomical structure coupling strength grouping identifiers (showing weakened connection between the pedicle and vertebral body), and three-dimensional anatomical node topological relationships (showing spatial proximity between the neural foramen and the tumor). This feature set is input into an adaptive spatial planning network model to predict that the puncture path will continuously deflect inwards as it approaches the pedicle. The puncture probability density function shows a secondary peak near the neural foramen, while the avoidance continuity feature indicates a high risk of angle abrupt changes. Based on this, the system identifies the critical risk region between the medial wall of the pedicle and the neural foramen. In the simulated avoidance, the system records that the peak amplitude of the path deviation is large and the response lag time is long. Combined with the evolution trend, multi-dimensional coupling analysis is performed to determine that there are structural stability failure markers and high local path deviation indices in this region, generating a regional risk level label. Finally, based on this label, the path priority is divided into gradients, prioritizing the restriction of the medial deviation angle and guiding the path along the bone ridge with high coupling strength, outputting a safe puncture plan.

[0045] In one embodiment, the steps for obtaining the cone structure clustering feature set are as follows: By using multimodal imaging equipment to collect real-time data on bone density distribution, anatomical node spacing, and structural coupling stability in the vertebral region of the patient, a set of raw image parameters is obtained.

[0046] Among these, multimodal imaging equipment can be an integrated imaging system capable of synchronously or asynchronously acquiring multiple medical imaging modalities (such as CT, ultrasound, and optical navigation) to support real-time acquisition of multidimensional parameters such as bone density, spacing, and stability. Bone density distribution can be the spatial distribution pattern of bone mineral density at different spatial locations within the vertebral body, used to reflect bone strength and structural integrity, and to identify osteoporotic areas or areas of weak cortical bone. Anatomical node spacing can be the Euclidean distance or geodesic distance between two key anatomical landmarks in the vertebral body (such as the medial wall of the pedicle or the center of the endplate), used to characterize the spatial geometric relationship of local anatomical structures and their potential relative displacement trends during surgery. Structural coupling persistence stability can be an indicator of the ability of adjacent anatomical structures to maintain their relative position and mechanical association when subjected to external disturbances (such as puncture force or changes in body position), used to quantify the deformation resistance of local areas of the vertebral body under dynamic conditions during surgery and to predict the risk of path deviation.

[0047] Real-time acquisition of bone density distribution, anatomical node spacing, and structural coupling stability in the vertebral region using multimodal imaging equipment allows for the simultaneous acquisition of multi-source data reflecting bone density, anatomical geometry, and structural stability during surgery. Furthermore, this operation can be achieved by combining intraoperative CT with an optical surface tracking system, calculating node spacing and stability through registration, or by fusing ultrasound elastography with low-dose CT to simultaneously estimate bone density and local stiffness. This enables dynamic quantitative perception of the vertebral body's internal biomechanical-structural heterogeneity.

[0048] The raw image parameter set can be an unprocessed dataset containing raw measurements of bone density distribution, anatomical node spacing, and structural coupling stability acquired by multimodal imaging equipment. This dataset serves as the initial input for subsequent structural coupling coefficient calculations and clustering. Alternatively, the raw image parameter set can be obtained by integrating the acquired multidimensional raw measurements into a structured dataset within a unified coordinate system, thus providing standardized input for structural coupling coefficient calculations.

[0049] Based on bone density data from the original image parameter set, the structural coupling coefficient between adjacent anatomical nodes is calculated, and clustering is performed according to the coefficient similarity threshold to generate anatomical structural coupling group identifiers.

[0050] Bone density data can be numerical voxel or regional measurements representing bone density distribution from the raw image parameters, used to calculate the structural coupling strength between adjacent anatomical nodes. Adjacent anatomical nodes can be two key anatomical landmarks with direct connections or functional relationships in the anatomical topology, serving as the basic unit for calculating the structural coupling coefficient. The structural coupling coefficient can be a dimensionless value calculated based on bone density data, representing the strength of the mechanical association between adjacent anatomical nodes, used to quantify the structural synergy between bone tissues in different regions, supporting clustering. Furthermore, the structural coupling coefficient can be derived comprehensively from bone density gradient, node spacing change rate, and local stiffness models. The coefficient similarity threshold can be a preset or adaptive similarity boundary value used to determine whether two structural coupling coefficients belong to the same cluster group, used to control the clustering granularity and ensure that structures within a group have similar biomechanical properties. Clustering can be a set of anatomical sub-regions with homogeneous mechanical-structural properties, divided according to the similarity of the structural coupling coefficient, used as the basic unit for forming anatomical structural coupling group identifiers. Anatomical structure coupling grouping identifiers can be three-dimensional label maps that mark the vertebral body space. Each label corresponds to a cluster group and its structural coupling attributes, which are used to provide structural partitions with biomechanical significance to construct a high-dimensional vertebral body structure clustering feature set. Furthermore, anatomical structure coupling grouping identifiers can include, but are not limited to, high-coupling stability grouping identifiers, medium-coupling fluctuation grouping identifiers, and low-coupling deformable grouping identifiers.

[0051] The structural coupling coefficient between adjacent anatomical nodes is calculated based on bone density data from the original image parameter set. This can be achieved by deriving the mechanical correlation strength between adjacent nodes using bone density gradients, local stiffness models, and the rate of change of node spacing. Further, this operation can be implemented by estimating local coupling stiffness using finite element inversion methods and normalizing it to a coupling coefficient, or by generating coupling coefficients based on graph neural networks learning the joint mapping relationship between bone density and spacing. This transforms static density information into a mechanically meaningful measure of structural correlation. Clustering is then performed based on a coefficient similarity threshold. This can involve dividing the structural coupling coefficients into several clusters according to the similarity threshold, with each cluster corresponding to an anatomical sub-region with homogeneous mechanical properties. Further, this operation can be achieved by using the DBSCAN algorithm for density clustering and automatically determining the number of groups, or by using hierarchical clustering combined with the elbow rule to select the optimal number of groups. This allows for the identification of anatomical regions with similar biomechanical response patterns. Finally, anatomical structural coupling group identifiers are generated. This can be achieved by labeling the clustering results in 3D on the vertebral space, forming a structural coupling semantic map, thereby establishing a physically meaningful representation of anatomical regions.

[0052] Bone density distribution characteristics, coupling stability and node spacing fluctuation range of each group in the anatomical structure coupling group identifier are extracted to construct a vertebral structure cluster feature set.

[0053] Among these features, bone density distribution characteristics can be statistical features of bone density within the anatomical structure coupling group (such as mean, variance, and gradient direction), used to describe the bone strength characteristics of the group. Coupled stability can be a quantitative indicator of the ability of the anatomical structure coupling group to maintain structural relationships under dynamic perturbations, used to reflect the possibility of deformation or displacement of the group during surgery. Node spacing fluctuation range can be the range of changes in the spacing between adjacent anatomical nodes within the anatomical structure coupling group under multiple measurements or simulated perturbations, used to characterize the dynamic uncertainty of the geometric relationship of the group. Extracting the bone density distribution characteristics, coupled stability, and node spacing fluctuation range of each group in the anatomical structure coupling group identifier can be achieved by statistically summarizing the original parameters within each group and extracting representative feature vectors. Furthermore, this operation can be achieved by calculating the mean and standard deviation of bone density within the group, the coefficient of variation of the stability time series, the range of spacing fluctuations, or by fitting the probability distribution of parameters within the group and extracting the distribution parameters as features, thereby constructing a high-dimensional, multi-attribute fused structural description. Constructing a vertebral structure cluster feature set can be achieved by integrating the multidimensional features of each group into a unified three-dimensional feature dataset, with each spatial location having attached structural coupling attributes. This can generate a refined anatomical model with embedded dynamic response patterns for subsequent path prediction.

[0054] For example, in the scenario of vertebroplasty for patients with osteoporosis and weak posterior vertebral walls, the path planning method for vertebral puncture using orthopedic surgical equipment in this embodiment can be as follows: During the operation, bone density distribution, pedicle-endplate distance, and deformation stability of local structures under respiratory disturbances in the L2 vertebral body region are acquired by combining C-arm CT and ultrasound elastography to form a raw set of image parameters; the system calculates the structural coupling coefficient between adjacent anatomical nodes (such as the left pedicle and the superior endplate) based on the bone density gradient and the rate of change of distance, and finds that the coupling coefficient in the posterior wall region is significantly lower than that in the anterior column; based on adaptive design... By setting a coefficient similarity threshold, the vertebral body is divided into a highly coupled anterior column group, a moderately coupled lateral wall group, and a low-coupled posterior wall group, generating anatomical structure coupling group identifiers. Further, the mean bone mineral density (significantly lower in the posterior wall group), coupling stability (large fluctuations in the posterior wall group), and node spacing fluctuation range (up to 3.2 mm in the posterior wall group) of each group are extracted to construct a vertebral body structure cluster feature set. After this feature set is input into an adaptive spatial planning network model, the model identifies a high risk of rupture in the posterior wall region, actively shifts the puncture path outward and limits the needle insertion depth, effectively avoiding puncture of the posterior wall of the vertebral body and damage to the spinal cord.

[0055] In one embodiment, the steps for obtaining the path evolution trend sequence are as follows: The vertebral structure cluster feature set is discretized according to anatomical spatial coordinates, and spatial coding and avoidance trend label are assigned to each anatomical node to generate the puncture path state evolution time series. The anatomical spatial coordinates can be a three-dimensional spatial coordinate system established based on the patient's vertebral anatomical structure, used to uniformly represent the positional relationships of various anatomical elements. In this embodiment, the anatomical spatial coordinates provide a standardized spatial reference framework for the vertebral structure cluster feature set, supporting discretization and encoding. Discretization can be the process of dividing a continuous vertebral structure cluster feature set into a finite number of discrete units according to a preset spatial granularity. Further, discretizing the vertebral structure cluster feature set according to the anatomical spatial coordinates can be achieved by dividing the continuous vertebral structure cluster feature set into discrete voxels or grid units of fixed size or adaptive granularity under the anatomical spatial coordinate system. For example, this operation can be achieved by using a uniform voxel grid for regular discretization, or by adaptively dividing non-uniform node density based on structural gradient, thereby realizing the transformation from continuous anatomical structure to a computable set of nodes.

[0056] Anatomical nodes can be basic anatomical units formed by discretization, possessing clear spatial locations and structural attributes. In an exemplary embodiment, anatomical nodes serve as the basic computational units for path state evolution and temporal modeling. Spatial coding can be a unique identifier assigned to each anatomical node, encoding its positional information in anatomical spatial coordinates. Furthermore, spatial coding can enable rapid node localization and precise tracking of path states. Avoidance trend indicators can be risk response tags attached to anatomical nodes, indicating the trend that the node should be avoided or traversed with caution in path planning. In a specific embodiment, avoidance trend indicators are embedded with preliminary risk assessment results, guiding path evolution towards a safe area. The puncture path state evolution temporal sequence can be a state sequence composed of anatomical nodes with spatial coding and avoidance trend indicators arranged in temporal or spatial order. Furthermore, the puncture path state evolution temporal sequence can construct a basic representation of path states with spatiotemporal semantics, supporting dynamic modeling. For example, the puncture path state evolution temporal sequence can include initial state subsequences, intermediate transition state subsequences, target proximity state subsequences, etc.

[0057] Assigning a spatial code and avoidance trend identifier to each anatomical node can be achieved by generating a unique code based on the node's position in the anatomical spatial coordinates and associating it with the risk attributes of its respective cluster group to generate an avoidance trend identifier. Furthermore, this operation can be implemented by using Z-order curve encoding to maintain spatial locality, or by mapping a regional risk level identifier dataset to generate a three-valued avoidance trend (avoidance / caution / passability), thereby endowing the node with localization capabilities and risk-oriented semantics. Generating the temporal sequence of puncture path state evolution can be achieved by arranging the coded and labeled anatomical nodes in an ordered state sequence according to the puncture logic (e.g., from the extradermal to the vertebral body center), thus constructing a basic representation of the path state with spatiotemporal semantics.

[0058] Based on the puncture path state evolution time series, an adaptive spatial planning network model is used to dynamically model the path direction angle change rate and path dwell time between nodes, and output the path state transition density function. The path orientation angle change rate can be the rate at which the angle of the puncture path changes between adjacent anatomical nodes. In one embodiment, the path orientation angle change rate reflects the curvature and turning abruptness of the path, used to assess operative feasibility and tissue damage risk. The path dwell time can be the expected or simulated length of time the puncture instrument remains at a certain anatomical node or local area. Further, the path dwell time characterizes operative stability or potential jamming risk; a high dwell time may indicate high resistance or high risk. The path state transition density function can be a continuous probability density function describing the probability of a safe transition from the current anatomical node to the next node. In this embodiment, the path state transition density function is jointly output by the adaptive spatial planning network model based on the orientation angle change rate, dwell time, and node risk attributes. Further, the path state transition density function can quantify the probability distribution of path feasibility, supporting optimal path selection. For example, the path state transition density function may include a high-safety transition probability sub-distribution, a medium-risk transition probability sub-distribution, and a low-feasibility transition probability sub-distribution.

[0059] An adaptive spatial programming network model is employed to dynamically model the rate of change of path orientation angles and path dwell time between nodes. This can be achieved by inputting the time series of puncture path state evolution into the model and jointly learning the impact of the rate of change of orientation angles and dwell time on path safety. Furthermore, this operation can be further refined by using gated recurrent units (GRUs) to model temporal dependencies and output a joint probability distribution, or by constructing a graph attention network, treating nodes as graph nodes and using edge weights to reflect transition feasibility. This allows for multi-dimensional joint modeling of path dynamic behavior. The output path state transition density function can be the model outputting the probability density distribution of safe transitions from any node to its neighboring nodes, thus providing a probabilistic and quantitative basis for path feasibility.

[0060] Based on the path state transition density function, the highest probability safe path is selected as the main trend path. The continuous change sequence of direction angle in the path is extracted, and the path evolution trend sequence is generated by combining the dissected node number.

[0061] The highest probability safe path can be the path trajectory with the highest cumulative transition probability in the path state transition density function. In one embodiment, the highest probability safe path serves as a candidate for the main trend path, ensuring comprehensive safety at the geometric, mechanical, and anatomical levels. The main trend path can be a dominant path selected through screening, representing the most likely safe puncture direction. Furthermore, the main trend path serves as the basis trajectory for generating the path evolution trend sequence. The continuous change sequence of azimuth angles can be a continuous numerical sequence formed by sequentially arranging the puncture azimuth angles between nodes in the main trend path. In an exemplary embodiment, the continuous change sequence of azimuth angles characterizes the spatial dynamics of the path and is used for risk threshold identification. The anatomical node number can be a unique serial number assigned to discretized anatomical nodes according to spatial or topological order. Furthermore, the anatomical node number supports the structured organization and retrospective analysis of the path sequence.

[0062] The highest-probability safe path is selected as the main trend path based on the path state transition density function. This can be achieved through dynamic programming or maximum likelihood path search algorithms, finding the path with the highest cumulative probability in the density function space. This operation is implemented by using the Viterbi algorithm for optimal path backtracking or by exploring the high-probability path space using Monte Carlo tree search, thus ensuring that the selected path has the highest safety under multidimensional constraints. Extracting the continuous change sequence of orientation angles in the path can be done by calculating the puncture orientation angles between adjacent nodes along the main trend path, forming a continuous numerical sequence, thereby capturing the dynamic characteristics of the path's spatial orientation. Combining the dissection node numbers to generate a path evolution trend sequence can be done by binding the continuous change sequence of orientation angles with the corresponding dissection node numbers, forming structured time-series data, thus generating a high-fidelity prediction sequence that can be used for risk critical area identification and real-time adjustment.

[0063] Taking thoracic vertebral body tumor puncture biopsy as an example, the path planning method for vertebral body puncture using orthopedic surgical equipment in this embodiment can be as follows: The system discretizes the constructed vertebral body structure cluster feature set into 1mm³ nodes in the T7 vertebral body anatomical spatial coordinates, assigns a Z-order spatial code to each node and adds an avoidance trend label (such as marking the area near the costotransverse joint as "avoidance"); after generating the puncture path state evolution time sequence, the adaptive spatial planning network model dynamically models and finds that: if the path directly passes through the medial side of the pedicle, the change rate of the direction angle increases sharply and the dwell time is significantly prolonged, indicating that there is a risk of nerve compression; the path state transition density function output by the model shows that the path offset outward by 2° has the highest cumulative safety probability; the system selects this path as the main trend path, extracts its continuous change sequence of direction angle (the change is gradual, and the maximum change rate is <5° / mm), and binds it with the anatomical node number to generate a path evolution trend sequence; the subsequent steps identify a change point of direction angle in the original planned path as a risk critical area based on this sequence, trigger an early warning and complete the avoidance to avoid damage to the intercostal nerve.

[0064] In one embodiment, the steps for obtaining the path avoidance response behavior dataset are as follows: Based on the path evolution trend sequence, the second derivative of the puncture path direction angle is calculated to identify critical regions that exceed the preset risk threshold and integrate them into the path risk critical region. For critical areas of path risk, pulsed avoidance control signals are applied to the robotic arm, and the time difference between the signal triggering time and the robotic arm response start time is measured to generate a dynamic response time series. Record the maximum path offset, the dynamic trajectory of the avoidance direction, and the variance of the response process in the dynamic response time series to obtain a path avoidance response behavior dataset.

[0065] The puncture path direction angle is the angle of the puncture instrument's travel direction relative to the reference coordinate system at a point on the path. It can be used to describe the local direction of the path, and its variation characteristics reflect the path curvature and turning behavior. In this embodiment, the puncture path direction angle is obtained by analyzing the path evolution trend sequence, serving as the basic input for subsequent geometric change analysis. The second derivative value can be the second derivative of the puncture path direction angle with respect to the path arc length or time, characterizing the rate of change of path curvature. It can be used to quantify the degree of abrupt changes in path geometry and to identify high-risk turning areas. For example, the second derivative value can be estimated by using the central difference method to estimate the second derivative of the discrete direction angle sequence, or by fitting a continuous direction angle function through spline interpolation and then analytically differentiating it. The preset risk threshold can be a set boundary value used to determine whether the second derivative value constitutes a critical risk state. It can be used as a criterion for identifying critical risk areas of the path and controlling the warning sensitivity. In an exemplary embodiment, the preset risk threshold can be pre-calibrated based on the safety margin of the anatomical structure and the kinematic characteristics of the instrument, and stored in the system parameter library.

[0066] The path risk critical region can be a local segment in the path evolution trend sequence where the second derivative value exceeds a preset risk threshold. It can be used to identify high-risk transitional regions where geometrical abrupt changes in the path may trigger neurovascular injury. Furthermore, the path risk critical region can include, but is not limited to, one or more of the following: high curvature abrupt change critical region, sharp turning critical region, and complex deformation critical region. Calculating the second derivative value of the puncture path direction angle based on the path evolution trend sequence can be achieved by numerically differentiating the direction angle data in the path evolution trend sequence to obtain its second derivative with respect to the path parameters. Further, this operation can be achieved by estimating the second derivative of the discrete direction angle sequence using the central difference method, or by fitting a continuous direction angle function using spline interpolation and then analytically differentiating it, thereby quantifying the rate of change of path curvature and identifying geometrical abrupt change regions. Identifying critical regions exceeding the preset risk threshold can be achieved by comparing the calculated second derivative value with the preset risk threshold and marking the exceeding segment. In a specific embodiment, this operation can be achieved by detecting whether the local peak value continuously exceeds the threshold using a sliding window, or by combining the first derivative of the direction angle to jointly determine the complex risk state, thus enabling the mathematical localization of potentially high-risk regions.

[0067] Integrating into a path risk critical region can involve merging all out-of-threshold segments, denoising them, and mapping them back to anatomical space to form continuous risk region markers, thereby generating a structured set of risk regions that can be used to trigger avoidance tests. The robotic arm can be the end effector of an orthopedic surgical robot performing vertebral puncture procedures, capable of receiving control signals and performing actual puncture or avoidance actions. In an exemplary embodiment, the robotic arm has six degrees of freedom of motion, and its end-effector pose can be fed back in real time via joint encoders or an external navigation system. The pulsed avoidance control signal can be a brief, directional disturbance-type control command applied to the robotic arm to trigger a simulated obstacle avoidance response. It can be used to actively stimulate the system's avoidance behavior to observe its dynamic response characteristics, rather than waiting for an actual collision. Furthermore, the pulsed avoidance control signal can include, but is not limited to, angle offset pulse signals, torque disturbance pulse signals, and velocity transient pulse signals.

[0068] The signal trigger moment can be the precise time point at which the pulsed avoidance control signal is sent to the robot control system, and can be used as a starting reference for dynamic response delay calculation. In one specific embodiment, the signal trigger moment is synchronously recorded by a high-precision clock inside the main control system. The robotic arm response start moment can be the starting time point at which the robotic arm begins to generate detectable pose changes, and can be used as a termination reference for dynamic response delay calculation. For example, the robotic arm response start moment can be determined by a sudden change in the force sensor signal or a pose change threshold of the optical tracking system. The time difference can be the interval between the signal trigger moment and the robotic arm response start moment, and can be used to quantify the response delay characteristics of the system control link. In this embodiment, the time difference is precisely measured through a high-precision clock synchronization mechanism. The dynamic response time series can be a sequence formed by arranging the time differences measured in multiple pulse tests in temporal or spatial order, and can be used to characterize the changes in the system's real-time response capability at different anatomical positions. Furthermore, the dynamic response time series can include, but is not limited to, low-latency response subsequences, medium-latency response subsequences, and high-latency response subsequences.

[0069] Applying pulsed avoidance control signals to the robotic arm in critical path risk areas can involve injecting brief directional disturbance commands at the corresponding path points within the identified critical risk areas to simulate emergency obstacle avoidance. Furthermore, this operation can be performed by applying pulse signals and recording digital responses in a virtual simulation environment, or by applying small safety pulses and monitoring actual mechanical responses in a real puncture-paused state, thereby actively stimulating the system's avoidance behavior to observe its dynamic performance. Measuring the time difference between the signal trigger moment and the robotic arm's response initiation moment can be achieved by synchronously recording the timestamps of the control signal issuance and the first change in the robotic arm's pose using a high-precision clock and calculating the difference. In one specific embodiment, this operation determines the response initiation moment based on a sudden change in the force sensor signal or based on a pose change threshold of the optical tracking system, thus obtaining empirical data on the system control delay. Generating a dynamic response time series can be achieved by organizing the time differences obtained from multiple tests into an ordered sequence according to path location or test order, thereby constructing a spatiotemporal distribution model of the system's response capability.

[0070] The maximum path deviation can be the maximum spatial distance by which the puncture path deviates from the planned trajectory during the avoidance response. It can be used to assess whether the magnitude of the obstacle avoidance action is within the safety tolerance, avoiding excessive deviation that could lead to new risks. In an exemplary embodiment, the maximum path deviation is obtained directly by an external optical navigation system. The avoidance direction dynamic trajectory can be the spatial trajectory of the actual movement of the puncture instrument during the avoidance response. It can be used to reflect the spatial rationality and anatomical compatibility of the avoidance path. In this embodiment, the avoidance direction dynamic trajectory is reconstructed from the robot's end-effector pose sequence. The response process fluctuation variance can be a statistical measure of the dispersion of the path deviation or control output around the mean during the avoidance response. It can be used to characterize the stability and repeatability of the system's avoidance actions. For example, the response process fluctuation variance is obtained by calculating the variance of the deviation sequence using a sliding window. Recording the maximum path deviation, the avoidance direction dynamic trajectory, and the response process fluctuation variance in the dynamic response time series can be done by simultaneously collecting pose data during the avoidance response and calculating these three key performance indicators. Furthermore, this operation can be achieved by inferring the end-effector trajectory and calculating the offset and variance using the robot's joint encoder, or by directly measuring the end-effector spatial trajectory using an external optical navigation system. This allows for a comprehensive characterization of the amplitude, direction, and stability features of the avoidance behavior. Obtaining a path avoidance response behavior dataset can be achieved by structurally storing the maximum path offset, the dynamic trajectory of the avoidance direction, and the variance of the response process as a multidimensional dataset. This provides empirical behavioral evidence for multidimensional coupled analysis and supports closed-loop control.

[0071] For example, in a scenario of dynamic verification during lumbar pedicle puncture, the path planning method for vertebral puncture using an orthopedic surgical machine in this embodiment can be as follows: The system calculates the second derivative of the direction angle of the L3 pedicle puncture path based on the path evolution trend sequence, and finds a curvature abrupt change (second derivative reaches 0.8 rad / mm²) before entering the pedicle isthmus, exceeding the preset risk threshold of 0.6; this section is integrated into the path risk critical area. The system then applies a pulsed avoidance control signal with a 5° outward deflection to the robotic arm, and records the signal trigger time and the actual start time of the robotic arm deflection, measuring a time difference of 85 ms, which is included in the dynamic response time series. At the same time, the optical navigation system records that the maximum path deviation is 1.2 mm (less than the safety tolerance of 2 mm), the dynamic trajectory of the avoidance direction smoothly avoids the neural foramen, and the variance of the response process fluctuation is 0.04 mm² (indicating stable execution). The three indicators together constitute the path avoidance response behavior dataset. Subsequent multi-dimensional coupling analysis confirmed that although the area is a geometric critical zone, the system has sufficient avoidance capabilities and does not require major replanning. It can pass safely by only fine-tuning the needle entry angle.

[0072] In one embodiment, the steps for obtaining the regional risk level identification dataset are as follows: Align the path avoidance response behavior dataset and the path evolution trend sequence according to the dissected node number, calculate the time offset of the path offset duration and the path state transition cycle, and generate node response offset features. Based on the node response offset characteristics, kernel density estimation is used to analyze the distribution dispersion of the response offset of each node. The risk identification threshold is set according to the dispersion confidence interval, and candidate nodes for high-risk areas are output. Based on candidate nodes in high-risk areas, node numbers, corresponding anatomical spatial coordinates, and risk level indexes are extracted. Risk node distribution density is aggregated by vertebral anatomical region, and structural risk index of each region is labeled to establish a regional risk level label dataset.

[0073] The path deviation duration can be the length of time the puncture path deviates from the planned trajectory during the obstacle avoidance response. It can reflect the time characteristics of the system maintaining obstacle avoidance status, and a long deviation may indicate a high-drag or high-risk area. In an exemplary embodiment, the path deviation duration can be obtained by recording the time interval between the first deviation of the actual path trajectory from the planned trajectory and its reconvergence. Furthermore, the path deviation duration can be used in conjunction with the path state transition period to characterize the local control dynamics. The path state transition period can be the average time interval between significant changes in adjacent states (such as orientation angle and curvature) in the path evolution trend sequence. It can be used to characterize the dynamic complexity of the path, and short-cycle transitions indicate high-frequency disturbances or structural instability. For example, the path state transition period can be obtained by detecting state abrupt change points through a sliding window and calculating their average interval. In a specific embodiment, the path state transition period can serve as a benchmark indicator for the dynamic response rhythm of the prediction model.

[0074] The time offset is the difference or phase difference between the duration of the path offset and the path state transition period. It is used to quantify the degree of time mismatch between the predicted dynamics and the actual response, serving as an indicator of local control difficulty. In this embodiment, the time offset can be obtained by calculating the absolute time difference or a normalized relative ratio (offset duration / transition period) in conjunction with the context. Furthermore, the time offset can form a mapping relationship with node response offset features. The node response offset features can be multi-dimensional feature vectors constructed based on the time offset, characterizing the system response and predicted mismatch at each dissected node. These features can be used to fuse the differences between dynamic prediction and measured response for risk quantification modeling. In a specific embodiment, the node response offset features can include, but are not limited to, one or more of the following: low mismatch stability features, medium mismatch fluctuation features, and high mismatch hysteresis features. Aligning the path avoidance response behavior dataset and the path evolution trend sequence by dissected node number can be achieved by using a common dissected node number as a key to precisely match the two sets of heterogeneous data in the spatial-temporal dimension. Furthermore, this operation can achieve O(1) node alignment through hash mapping, or use interpolation alignment to process non-completely overlapping node sequences, thereby achieving spatiotemporal synchronization between predicted behavior and measured response.

[0075] Calculating the time offset between the duration of path offset and the path state transition period can be achieved by obtaining the duration of offset and the state transition period for each node separately, and calculating the difference between the two or the normalized phase difference. Furthermore, this operation can be implemented by calculating the absolute time difference as the offset, or by calculating the relative proportion (offset duration / transition period) as a dimensionless offset index, thereby quantifying the degree of dynamic mismatch between prediction and response. Generating node response offset features can be achieved by combining the time offset with other relevant indicators (such as volatility variance, maximum amplitude) into a feature vector, thereby constructing a multidimensional response mismatch characterization. Kernel density estimation can be a nonparametric probability density estimation method used to infer the continuous distribution pattern from a finite sample. It can be used to model the distribution of node response offsets without assumptions, avoiding normality constraints. For example, kernel density estimation can fit the sample distribution using a Gaussian kernel function and output a smooth probability density curve.

[0076] The dispersion of the response offset distribution can be a statistical measure of the diffusion of the response offset distribution (such as bandwidth, entropy, or standard deviation) obtained through kernel density estimation. It can be used to measure the consistency of the system's response at a specific node; higher dispersion indicates greater risk. In an exemplary embodiment, the dispersion can be quantified using bandwidth or Shannon entropy obtained from kernel density estimation. The dispersion confidence interval can be a reliable range of the dispersion statistic calculated based on the sample distribution. It can be used to provide a statistical basis for the risk identification threshold, improving cross-patient adaptability. In this embodiment, the dispersion confidence interval can be obtained through Bootstrap resampling or Bayesian posterior distribution estimation. The risk identification threshold can be a dynamically set boundary value based on the dispersion confidence interval, used to determine whether a node belongs to high risk. It can be used to achieve adaptive risk screening and avoid misjudgments caused by fixed thresholds. In a specific embodiment, the risk identification threshold can be set as the upper limit of the 95% confidence interval.

[0077] Candidate nodes for high-risk regions can be a set of anatomical nodes whose response offset distribution dispersion exceeds the risk identification threshold, and can be used as the input basis for calculating the structural risk index. For example, candidate nodes for high-risk regions can include, but are not limited to, one or more of the following: highly discrete neural neighbor nodes, highly discrete osteoporotic nodes, and highly discrete interface transition nodes. Kernel density estimation is used to analyze the distribution dispersion of the response offset of each node. This can be achieved by fitting the kernel density to the response offset of each node in multiple tests or neighborhood sampling, and calculating the distribution bandwidth or entropy. Further, this operation can be implemented by using a Gaussian kernel function for density estimation and calculating the standard deviation, or by using adaptive kernel bandwidth to optimize the accuracy of local density estimation, thereby quantifying response stability without assumptions. The risk identification threshold is set based on the dispersion confidence interval, which can be a dynamic threshold calculated based on the upper limit of the 95% confidence interval from the dispersion samples. Further, this operation can be implemented by estimating the confidence interval through Bootstrap resampling, or by combining it with a Bayesian posterior distribution to set an adaptive threshold, thereby achieving individualized anatomical risk identification.

[0078] Outputting high-risk region candidate nodes can be done by filtering nodes whose dispersion exceeds a threshold and marking them as high-risk candidates, thereby focusing on anatomical locations requiring key intervention. The risk level index can be a quantitative risk level identifier assigned to the high-risk region candidate nodes, which can be used to support subsequent grouping, aggregation, and priority ranking. In one embodiment, the risk level index can be represented using integer levels (e.g., 1–5) or continuous values ​​(e.g., 0–1). Extracting the node number, corresponding anatomical spatial coordinates, and risk level index based on the high-risk region candidate nodes can be a structured extraction of three key attributes from the candidate node set, thus providing standardized input for region aggregation. The vertebral body anatomical region can be a sub-region of the vertebral body divided according to anatomical function or structure (e.g., pedicle, anterior wall of the vertebral body, superior endplate, etc.), which can be used as a spatial unit for risk node aggregation. For example, the vertebral body anatomical region can be predefined by preoperative image segmentation and assigned a unique region identifier.

[0079] Risk node distribution density can be the number or weighted density of candidate nodes in high-risk areas within a unit anatomical region, reflecting the overall risk concentration in that region. In one specific embodiment, risk node distribution density can be obtained through voxel counting or kernel smoothing weighted density estimation. Aggregating risk node distribution density by vertebral anatomical region can involve grouping high-risk nodes by predefined anatomical regions (e.g., pedicles) and calculating the number or weighted density per unit volume. Further, this operation can be achieved by calculating density using voxel counting or kernel smoothing weighted density estimation, thereby elevating the assessment from point risk to regional risk. The structural risk index can be a continuous numerical indicator characterizing the overall risk level of a vertebral anatomical region, calculated based on risk node distribution density, and can be used to provide quantifiable and comparable risk data for gradient path control. For example, the structural risk index can include, but is not limited to, one or more of the following: low-risk stable index area, medium-risk fluctuation index area, and high-risk clustering index area. Labeling the structural risk index of each region can be achieved by mapping the risk node distribution density to a continuous risk index (e.g., 0–1 or 1–5 levels). Furthermore, this operation can generate continuous indices through linear normalization or by setting piecewise mapping functions based on clinical priors, thereby generating interpretable and comparable regional risk quantification indicators. Establishing a regional risk labeling dataset can be achieved by integrating the structural risk indices of each vertebral anatomical region into a unified dataset using three-dimensional annotations, thus providing a high-dimensional risk map for gradient path priority allocation.

[0080] Taking multi-segment osteoporotic vertebroplasty as an example, the path planning method for vertebral puncture using an orthopedic surgical machine in this embodiment can be as follows: The system aligns the path avoidance response behavior dataset of L1–L4 with the path evolution trend sequence according to the anatomical node number. It is found that the duration of path offset in the L2 pedicle region (120ms) is significantly longer than its path state transition period (80ms), with a time offset of 40ms. Kernel density estimation shows that the dispersion of the response offset distribution in this region (bandwidth = 0.35) exceeds the upper limit of the 95% confidence interval (0.28). The nodes were marked as candidate nodes for high-risk areas. The system extracted their node numbers, spatial coordinates, and risk level indexes, aggregated them according to the pedicle anatomical region, and calculated that the distribution density of risk nodes in the right pedicle of L2 was 3.2 nodes / cm³, which was much higher than that of L3 (1.1 nodes / cm³). Based on this, the structural risk index of the right pedicle of L2 was marked as 0.87 (high risk), while that of L3 was 0.32 (low risk). The final generated regional risk label dataset guided the path planning to prioritize avoiding the high-risk area of ​​L2 and select a more lateral approach, successfully avoiding nerve damage.

[0081] In one embodiment, the steps for obtaining the vertebral puncture path planning results of the orthopedic surgical robot are as follows: Based on the regional risk level identification dataset, the weighted comprehensive risk level score is calculated by integrating two core indicators: the path direction angle offset and the offset magnitude of the dissected nodes. The path orientation angle offset can be the difference between the actual orientation angle of the puncture path and the originally planned orientation angle during the avoidance response. It can reflect the severity of the path turning adjustment and be used to assess the operational safety margin. In one embodiment, the path orientation angle offset can be obtained by comparing the intraoperative real-time navigation system with the preoperative planned path. Furthermore, the path orientation angle offset can be used in conjunction with the regional risk level indicator in multi-dimensional risk modeling. The offset amplitude can be the maximum distance the puncture path deviates spatially from the originally planned trajectory. It can be used to characterize the spatial magnitude of obstacle avoidance actions; excessive amplitude may introduce new risks. For example, the offset amplitude can be extracted from intraoperative optical or electromagnetic tracking data using a three-dimensional spatial trajectory reconstruction algorithm. In a specific embodiment, the offset amplitude and the path orientation angle offset together constitute a quantitative representation of the robot's dynamic response characteristics. The weighted comprehensive risk score can be a weighted composite risk quantification index that integrates the regional risk level indicator, the path orientation angle offset, and the offset amplitude. It can be used to unify anatomical vulnerability and robot dynamic response characteristics to generate a multi-dimensional risk score. Furthermore, the weighted comprehensive risk score may include, but is not limited to, one or more of the following: low comprehensive risk score, medium comprehensive risk score, and high comprehensive risk score.

[0082] Based on a regional risk level dataset, the path orientation angle offset and offset magnitude of the dissected nodes are fused. This can be achieved by concatenating or weighting the regional risk level labels and dynamic response indicators at the dissected node level. Furthermore, this operation can be implemented using linear weighting (i.e., summing the regional risk level labels multiplied by weight coefficient α, the path orientation angle offset multiplied by weight coefficient β, and the offset magnitude multiplied by weight coefficient γ) or by using a multilayer perceptron to learn nonlinear fusion weights, thereby constructing a comprehensive risk representation that integrates the dissection and execution dimensions. Calculating the weighted comprehensive risk score can be achieved by applying a fusion formula to each dissected node to output a single risk score value. Furthermore, this operation enables the generation of sortable and comparable node-level risk quantification indicators.

[0083] Based on the weighted comprehensive risk score, the mean and coefficient of variation of the global score distribution are calculated, and the planning priority labels are divided according to the risk gradient interval to generate the path planning priority distribution. The global score distribution can be a statistical distribution set composed of weighted comprehensive risk scores of all anatomical nodes, which can be used to characterize the overall risk pattern and local heterogeneity of the patient's vertebral body. The mean can be the arithmetic mean of the global score distribution, which can be used to reflect the overall average risk level of the vertebral body. In a specific embodiment, the mean can be obtained by traversing the risk scores of all anatomical nodes and calculating the arithmetic mean. The coefficient of variation can be the ratio of the standard deviation of the global score distribution to the mean, which can be used to measure the relative dispersion of the risk distribution, quantify the intensity of risk heterogeneity, and guide the granularity of gradient interval division. For example, the coefficient of variation can be obtained by first calculating the standard deviation and then dividing by the mean. The risk gradient interval can be a continuous risk level interval dynamically divided based on the mean and coefficient of variation, which can be used to achieve adaptive, non-uniform risk grading and avoid the rigidity of fixed thresholds. Furthermore, the risk gradient interval can include, but is not limited to, low-risk stable intervals, medium-risk transitional intervals, and high-risk clustering intervals. The planning priority label can be an intervention priority identifier assigned to each anatomical node according to the risk gradient interval, which can be used to guide the order of priority in path adjustment. In one exemplary embodiment, planning priority labels are automatically assigned by mapping node scores to corresponding gradient intervals. The path planning priority distribution can be a priority-labeled node risk priority map organized by the anatomical space, which can be used to provide structured input for gradient instruction generation.

[0084] The mean and coefficient of variation of the global score distribution based on the weighted comprehensive risk score can be calculated by taking the mean μ and standard deviation σ of all node scores, and then calculating the coefficient of variation CV = σ / μ. Furthermore, this operation can be achieved by using truncated means to reduce the influence of outliers or by using robust coefficients of variation (such as MAD / median) to improve noise resistance, thus adaptively characterizing the individual risk pattern of patients. Priority labels are planned according to risk gradient intervals, which can be achieved by dynamically setting interval boundaries based on μ and CV (e.g., low-risk intervals are 0 to μ minus 0.5 times CV, medium-risk intervals are μ minus 0.5 times CV to μ plus 1 times CV, and high-risk intervals are μ plus 1 times CV and above). Furthermore, this operation can be achieved by automatically determining the optimal number of intervals using K-means clustering or by combining clinical priors to set a minimum high-risk proportion, thus achieving individualized, non-uniform risk stratification. Generating the path planning priority distribution can be achieved by assigning each anatomical node to a corresponding gradient interval and assigning a priority label, forming a spatial priority map. Furthermore, this operation enables the construction of a structured priority control basis.

[0085] Based on the path planning priority distribution, the coordinates of the anatomical nodes corresponding to each priority and the priority labels are mapped to the robot control system to generate a gradient path adjustment instruction set and output the vertebral puncture path planning results of the orthopedic surgical robot.

[0086] The anatomical node coordinates can be the three-dimensional position of the anatomical node in the vertebral anatomical coordinate system, which can be used as the spatial positioning basis for the robot control system to adjust the path. The priority label can be a planning priority category identifier bound to the anatomical node, which can be used to determine the constraint strength of the corresponding node in the path adjustment. The robot control system can be the underlying control hardware and software system that receives the path planning results and drives the orthopedic surgical robot to perform puncture operations, which can be used to convert priority information into actual motion commands. The gradient path adjustment command set can be a set of robot control commands generated according to the priority distribution, applying differentiated constraints to different risk areas, which can be used to achieve intelligent control of strong constraints in high-risk areas and efficient passage in low-risk areas. Furthermore, the gradient path adjustment command set can include, but is not limited to, high-priority forced detour commands, medium-priority angle restriction commands, and low-priority free passage commands.

[0087] Mapping the coordinates and priority labels of anatomical nodes corresponding to each priority level to the robot control system based on the path planning priority distribution can convert priority-coordinate pairs into path constraint parameters recognizable by the robot control protocol. This operation can be achieved by publishing priority-based path point messages via ROS topics or writing them into the robot controller's dedicated path configuration file, thus completing the semantic transformation from risk assessment to control execution. Generating a gradient path adjustment instruction set can be achieved by generating differentiated instructions based on priority labels: high-priority areas are subject to strong constraints (such as maximum deflection angle limits and minimum feed rates), while low-priority areas allow for free optimization. Furthermore, this operation can be achieved by triggering virtual potential field repulsion in high-priority areas or enabling impedance control mode in medium-priority areas, thereby achieving differentiated robot behavior control based on risk perception. Outputting the vertebral puncture path planning results of the orthopedic surgical robot can be achieved by encapsulating the gradient path adjustment instruction set into a standard format and transmitting it to the surgical robot execution unit. Furthermore, this operation enables the provision of a final puncture plan that combines safety, efficiency, and individual adaptability.

[0088] Taking vertebroplasty for multi-segment compression fractures of the thoracolumbar region as an example, the path planning method for vertebral puncture using an orthopedic surgical machine in this embodiment can be as follows: Based on a regional risk level dataset, the system integrates the directional angle offset (8°) and offset amplitude (1.8mm) of the L1 pedicle region to calculate a weighted comprehensive risk score of 0.76; the global score distribution has a mean of 0.42 and a coefficient of variation of 0.65, and based on this, the risk level gradient intervals are dynamically divided: [0–0.35) is low risk, [0.35–0.65) is medium risk, [0.65–0.65) is medium risk, and [0.65–0.65) is low risk. (5–∞) is considered high risk; the L1 pedicle is assigned a high priority label, while the T12 anterior column score is 0.28 and is marked as low priority; after the path planning priority distribution is generated, the system maps it to the robot control system: the maximum needle insertion angle is limited by ±3° and the speed is ≤0.5mm / s in the high priority area, while ±10° and 1.5mm / s are allowed in the low priority area; the final output gradient path adjustment instruction set enables the robot to carefully fine-tune to avoid the neural foramen in the L1 area, while performing rapid straight-line puncture at T12, with no warning triggers throughout the process, reducing the operation time by 22%.

[0089] In addition, refer to Figure 2 To achieve the above objectives, the present invention also provides a path planning system for vertebral puncture using an orthopedic surgical machine, the system comprising: The image clustering module 10 is used to acquire multimodal image parameter sets of the patient's vertebral body region in real time, and perform three-dimensional structural feature clustering processing on the image parameter sets to generate a vertebral body structure clustering feature set. The model prediction module 20 is used to input the clustering feature set of the vertebral structure into the adaptive spatial planning network model, dynamically predict the evolution trend of the puncture path of key anatomical nodes, and construct the path evolution trend sequence. Risk warning module 30 is used to identify risk critical areas in the puncture path based on the path evolution trend sequence, perform dynamic response monitoring on the collision warning signal triggered in the critical area, and obtain a path avoidance response behavior dataset. The coupling analysis module 40 is used to perform multi-dimensional coupling analysis on the path avoidance response behavior dataset and the path evolution trend sequence, determine high-risk dissection areas, and form a regional risk level identification dataset. The path adjustment module 50 is used to perform gradient division of path planning priorities based on the regional risk level identification dataset, generate an adaptive path adjustment instruction set according to the priorities, and output the path planning result of the orthopedic surgical robot vertebral puncture.

[0090] Other embodiments or specific implementations of the path planning system for vertebral puncture using orthopedic surgical machines described in this invention can be referred to the above-described method embodiments, and will not be repeated here.

[0091] Furthermore, to achieve the above objectives, the present invention also provides a path planning device for vertebral puncture using an orthopedic surgical machine. The device includes: a memory, a processor, and a path planning program for vertebral puncture using an orthopedic surgical machine stored in the memory and executable on the processor. The path planning program for vertebral puncture using an orthopedic surgical machine is configured to implement the steps of the path planning method for vertebral puncture using an orthopedic surgical machine as described above.

[0092] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a path planning program for vertebral puncture using an orthopedic surgical machine. When the path planning program for vertebral puncture using an orthopedic surgical machine is executed by a processor, it implements the steps of the path planning method for vertebral puncture using an orthopedic surgical machine as described above.

[0093] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for path planning in vertebral puncture using an orthopedic surgical machine, characterized in that, The method includes: A multimodal image parameter set of the patient's vertebral body region is acquired in real time, and the image parameter set is subjected to three-dimensional structural feature clustering processing to generate a vertebral body structure clustering feature set; The clustering feature set of the vertebral structure is input into an adaptive spatial planning network model to dynamically predict the evolution trend of the puncture path at key anatomical nodes and construct a path evolution trend sequence. Based on the path evolution trend sequence, risk critical regions in the puncture path are identified, and dynamic response monitoring is performed on the collision warning signal triggered in the critical regions to obtain a path avoidance response behavior dataset. A multi-dimensional coupling analysis is performed on the path avoidance response behavior dataset and the path evolution trend sequence to identify high-risk areas and form a regional risk level identification dataset. The path planning priority is divided into gradients based on the regional risk level identification dataset. An adaptive path adjustment instruction set is generated according to the priority, and the path planning result of the orthopedic surgical robot vertebral puncture is output.

2. The path planning method for vertebral puncture using an orthopedic surgical machine as described in claim 1, characterized in that, The vertebral body structure clustering feature set includes vertebral body density distribution curves, anatomical structure coupling strength grouping identifiers, and three-dimensional anatomical node topological relationships. The path evolution trend sequence includes path direction angle time series, puncture probability density function, and anatomical node avoidance continuity features. The path avoidance response behavior dataset includes path deviation peak amplitude, early warning response lag time features, and avoidance direction vector. The regional risk level identifier dataset includes high-risk anatomical node numbers, local path deviation index, and structural stability failure markers. The orthopedic surgical robot vertebral puncture path planning results include path adjustment command type, robotic arm execution parameters, and target avoidance area coordinates.

3. The path planning method for vertebral puncture using an orthopedic surgical machine as described in claim 1, characterized in that, The specific steps for obtaining the clustering feature set of the cone structure are as follows: The bone density distribution, anatomical node spacing, and structural coupling stability of the patient's vertebral region are acquired in real time using multimodal imaging equipment to obtain the original set of image parameters; Based on the bone density data in the original parameter set of the image, the structural coupling coefficient between adjacent anatomical nodes is calculated, and clustering is performed according to the coefficient similarity threshold to generate anatomical structural coupling group identifiers. The bone density distribution characteristics, coupling stability, and node spacing fluctuation range of each group in the anatomical structure coupling group identifier are extracted to construct a vertebral structure cluster feature set.

4. The path planning method for vertebral puncture using an orthopedic surgical machine as described in claim 3, characterized in that, The specific steps for obtaining the path evolution trend sequence are as follows: The vertebral structure cluster feature set is discretized according to anatomical spatial coordinates, and spatial coding and avoidance trend identifier are assigned to each anatomical node to generate the puncture path state evolution time sequence. Based on the puncture path state evolution time series, an adaptive spatial planning network model is used to dynamically model the path direction angle change rate and path dwell time between nodes, and output the path state transition density function. Based on the path state transition density function, the highest probability safe path is selected as the main trend path, the continuous change sequence of direction angle in the path is extracted, and the path evolution trend sequence is generated by combining the dissected node number.

5. The path planning method for vertebral puncture using an orthopedic surgical machine as described in claim 4, characterized in that, The specific steps for obtaining the path avoidance response behavior dataset are as follows: Based on the path evolution trend sequence, the second derivative of the puncture path direction angle is calculated, and critical regions exceeding the preset risk threshold are identified and integrated into path risk critical regions. For the critical risk region of the path, a pulsed avoidance control signal is applied to the robot arm, and the time difference between the signal triggering time and the robot arm response start time is measured to generate a dynamic response time series. Record the maximum path offset, the dynamic trajectory of the avoidance direction, and the variance of the response process in the dynamic response time series to obtain a path avoidance response behavior dataset.

6. The path planning method for vertebral puncture using an orthopedic surgical machine as described in claim 5, characterized in that, The specific steps for obtaining the regional risk level identification dataset are as follows: Align the path avoidance response behavior dataset with the path evolution trend sequence according to the dissected node number, calculate the time offset between the path offset duration and the path state transition cycle, and generate node response offset features. Based on the node response offset characteristics, kernel density estimation is used to analyze the distribution dispersion of each node response offset. A risk identification threshold is set according to the dispersion confidence interval, and candidate nodes for high-risk areas are output. Based on the candidate nodes in the high-risk areas, the node numbers, corresponding anatomical spatial coordinates, and risk level indexes are extracted. The risk node distribution density is aggregated by vertebral anatomical region, and the structural risk index of each region is marked to establish a regional risk level identification dataset.

7. The path planning method for vertebral puncture using an orthopedic surgical machine as described in claim 6, characterized in that, The specific steps for obtaining the vertebral puncture path planning results of the orthopedic surgical robot are as follows: Based on the aforementioned regional risk level identification dataset, the weighted comprehensive risk level score is calculated by integrating two core indicators: the path direction angle offset and the offset magnitude of the dissected nodes. Based on the weighted comprehensive risk score, the mean and coefficient of variation of the global score distribution are calculated, and the planning priority labels are divided according to the risk gradient interval to generate the path planning priority distribution. Based on the path planning priority distribution, the anatomical node coordinates and priority labels corresponding to each priority are mapped to the robot control system to generate a gradient path adjustment instruction set and output the vertebral puncture path planning result of the orthopedic surgical robot.

8. A path planning system for vertebral puncture using an orthopedic surgical machine, characterized in that, The system includes: The image clustering module is used to acquire multimodal image parameter sets of the patient's vertebral body region in real time, and perform three-dimensional structural feature clustering processing on the image parameter sets to generate a vertebral body structure clustering feature set. The model prediction module is used to input the clustering feature set of the vertebral structure into the adaptive spatial planning network model, dynamically predict the evolution trend of the puncture path of key anatomical nodes, and construct the path evolution trend sequence. The risk warning module is used to identify the risk critical area in the puncture path based on the path evolution trend sequence, perform dynamic response monitoring on the collision warning signal triggered in the critical area, and obtain a path avoidance response behavior dataset. The coupling analysis module is used to perform multi-dimensional coupling analysis on the path avoidance response behavior dataset and the path evolution trend sequence, identify high-risk dissection areas, and form a regional risk level identification dataset. The path adjustment module is used to perform gradient division of path planning priorities based on the regional risk level identification dataset, generate an adaptive path adjustment instruction set according to the priorities, and output the path planning results of the orthopedic surgical robot vertebral puncture.

9. A path planning device for vertebral puncture using an orthopedic surgical machine, characterized in that, The device includes: a memory, a processor, and a path planning program for vertebral puncture using an orthopedic surgical machine, stored in the memory and executable on the processor, the path planning program for vertebral puncture using an orthopedic surgical machine being configured to implement the steps of the path planning method for vertebral puncture using an orthopedic surgical machine as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a path planning program for vertebral puncture using an orthopedic surgical machine, which, when executed by a processor, implements the steps of the path planning method for vertebral puncture using an orthopedic surgical machine as described in any one of claims 1 to 7.

Images