Trans-pedicle puncture intelligent navigation system and equipment

Abstract

The invention discloses a transpedicular puncture intelligent navigation system and equipment, and the system comprises a sensing and modeling module which is used for achieving the real-time construction of a dynamic surgical area digital twinborn model and the prediction of the dynamic change trend in the surgical process through the fusion of multi-source and heterogeneous intraoperative information; the decision-making and planning module is used for comprehensively considering multiple complex targets such as safety, accuracy, stability and long-term curative effect through a digital twinborn model constructed based on the sensing and modeling module, and planning a globally optimal'intelligent 'puncture strategy capable of adapting to environmental changes in real time for the surgical puncture device; and the control and execution module is used for realizing submillimeter trajectory tracking precision and high-robustness precise operation by ensuring that the system is in a complex, changeable and uncertain real surgical physical environment. By means of the intelligent navigation system, the problems that a current transpedicular puncture navigation system cannot adapt to dynamic changes in an operation, lacks real-time feedback and closed-loop control capacity, and is single in path planning dimension can be solved; therefore, safety, efficiency, precision and simplicity of vertebral body operation puncture are remarkably improved, ray exposure is reduced, the infection probability is reduced, and the curative effect is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of percutaneous transpedicular puncture surgery, and particularly relates to a percutaneous transpedicular puncture intelligent navigation system and equipment. BACKGROUND

[0002] The precision and safety of traditional percutaneous transpedicular puncture surgery are highly dependent on the personal experience and spatial perception ability of the operator. In order to reduce the risk of surgery and improve the consistency of operation, a mechanical navigation device based on preoperative images appears in the prior art. The core principle of this technology is: before surgery, high-precision three-dimensional images of the patient's spine are obtained through computer tomography (CT) or magnetic resonance imaging (MRI), and key anatomical parameters of the target pedicle are measured, such as length, inner and outer diameter, and various inclination angles relative to the vertebral body. Subsequently, based on these static geometric data, a pre-set trigonometric function and a spherical center positioning algorithm are used to calculate an accurate puncture approach. Finally, these calculation results are mapped to a physical, compass-like guide device, and by selecting a specific guide hole on the compass and setting the corresponding puncture depth, the travel of the surgical instrument is mechanically guided. This technology also has a certain correction ability, when the initial deviation occurs in the first puncture positioning, the distance and angle of the deviation can be measured and substituted into the formula to recalculate and guide subsequent adjustments.

[0003] However, the prior art has the following problems: this background technology is based on a fundamental, but very fragile assumption in the real surgical environment, namely the "static rigid body model" assumption. It regards the patient's spine as a rigid object whose shape and position are completely fixed during surgery. This idealized model ignores the dynamics and complexity of the real surgical environment, thus bringing some technical limitations, such as the inability to adapt to dynamic changes during surgery, the lack of real-time feedback and closed-loop control capabilities, the single dimension of path planning, etc.

[0004] Therefore, there is an urgent need for a more intelligent intelligent navigation system to solve the current problems. SUMMARY

[0005] The percutaneous transpedicular puncture intelligent navigation system and equipment provided by the present application aims to build a new technical system that can perform intelligent and adaptive puncture path planning and high-robustness closed-loop control in a dynamic and uncertain real surgical environment by fusing multi-source heterogeneous information.

[0006] The system includes: a perception and modeling module, which integrates multi-source, heterogeneous intraoperative information to build a dynamic digital twin model of the surgical area in real time and predict dynamic changes during the surgical process; a decision-making and planning module, which, based on the digital twin model built by the perception and modeling module, comprehensively considers multiple complex objectives such as safety, accuracy, stability, and long-term efficacy to plan a globally optimal "intelligent" puncture strategy for the surgical puncture device that can adapt to environmental changes in real time; and a control and execution module, which ensures that the system achieves sub-millimeter-level trajectory tracking accuracy and highly robust precision operation in a complex, variable, and uncertain real surgical physical environment.

[0007] The intelligent navigation system for transpedicle puncture also employs a surgical strategy function π to minimize risk.

[0008] π = argmin_{π∈policyspace}J(π);

[0009] J(π)=E_{all possible surgical procedures} }[integral_{from the start to the end of the surgery}(L(state(t), action(t)))dt+Φ(final state)];

[0010] The input to the strategy function is all the sensing information of the system at any time t, and the output is the action instruction that the surgical puncture device should execute at that time.

[0011] The perception and modeling module includes: a multimodal data fusion and non-rigid 3D reconstruction unit, used to precisely drive the deformation of a high-precision preoperative 3D model during surgery using limited, low-dose real-time images to match the patient's actual dynamics; and a hybrid dynamic modeling unit based on physical laws and data-driven methods, used to predict trends during surgery.

[0012] The decision-making and planning module includes: a global path corridor planning unit based on optimal control theory, which plans the optimal surgical path by dynamically calculating the probabilistic distribution during the surgical process; and a sequential decision-making unit based on hierarchical deep reinforcement learning, which constructs an intelligent decision-making system by combining optimal control theory with hierarchical deep reinforcement learning.

[0013] The workflow of this multimodal data fusion and non-rigid 3D reconstruction unit is as follows:

[0014] Data-constrained projection projects the current 3D digital model onto the same viewing angle as the real-time C-arm image, based on a simulated X-ray imaging physics process, forming a simulated 2D image. Then, the difference between the simulated image and the real C-arm image is calculated, and this difference is used as a driving force to backpropagate and adjust the shape of the 3D model, ensuring that its projection best matches the actual observation.

[0015] Physical / anatomical constraint projection, in the adjustment of the last step, the three-dimensional model may appear unreasonable twist or deformation of anatomy. In this step, the system will be deformed model, projection to a "consistent with the anatomical and biomechanical laws" constraint space, and according to the built-in anatomical priori knowledge of unreasonable deformation smoothing and correction, to ensure the authenticity of the model.

[0016] Wherein, the perception and modeling module dynamically calculates the spatial displacement that should occur during the operation process through the optimal deformation field function, thereby dynamically evolving into a digital twin model consistent with the real situation, and the optimal deformation field function is:

[0017] Optimal deformation field (spatial coordinates, time) = argmin [data fidelity term + physical constraint term + spatial smoothing regularization term];

[0018] Data fidelity term = integral (|| projection operator (preoperative three-dimensional model + deformation field) - intraoperative two-dimensional image || 2 ) d image;

[0019] Physical constraint term = physical constraint weight x integral (differential algebraic system energy (deformation field, rate of change of deformation field with respect to time)) d time;

[0020] Spatial smoothing regularization term = regularization weight x integral (|| gradient operator (deformation field) || 2 ) d space.

[0021] Wherein, the control and execution module includes a disturbance rejection control unit based on dead zone self-adaption, which is used to realize fast, accurate and adaptive online inhibition of various unknown disturbances without increasing the complexity of the system, and ensure the high fidelity execution of the operation strategy in the physical layer.

[0022] The application also provides a transpedicular navigation device, which comprises any of the transpedicular intelligent navigation systems described above, and in addition, the device further comprises a puncture device electrically connected with the intelligent navigation system, and the intelligent navigation system controls the precise operation of the puncture device.

[0023] Wherein, the working steps of the navigation device are:

[0024] Firstly, the target position of the vertebral body lesion is determined according to the imaging, and the number of target points is determined according to the specific puncture needs of the clinic, and is input and marked in the transpedicular navigation device;

[0025] Then, the maximum range of bilateral pedicle puncture is calculated by the intelligent navigation system, i.e. the maximum projection range of bilateral pedicle in the vertebral body is determined, and whether the target lesion site in the vertebral body in the previous step is covered by the maximum projection range is determined, if not, the puncture method is abandoned, if yes, the next process is entered;

[0026] Then, the needle operation path is dynamically planned.

[0027] Finally, the puncture device performs a puncture action.

[0028] The puncture device comprises:

[0029] A guide needle for pedicle puncture;

[0030] A compass placed on the surface of the patient's back skin during surgery for guiding the puncture direction and position of the guide needle, the guide needle passing through the center of the compass and forming an initial positioning;

[0031] A stabilizing component for fixing the compass at a predetermined position and supporting the movement of the compass.

[0032] The positioning step of the compass is:

[0033] First, determine the puncture range (the length of the pedicle is L and the width is d):

[0034] When length / width = 2 (L / d = 2): at this time, the intersection of puncture is located at the midpoint of the pedicle midline, and the puncture range on the compass is exactly the inner circle range, without scaling;

[0035] When length / width > 2 (L / d > 2): the pedicle is relatively "slim", and the puncture range on the compass needs to be scaled down, and the calculation formula of the scaled radius is r = (d / L)*G, G represents the length of the guide needle;

[0036] When length / width < 2 (L / d < 2): the pedicle is relatively "short and thick", and the puncture range on the compass needs to be scaled up, and the calculation formula of the scaled radius is also r = (d / L)*G, G represents the length of the guide needle;

[0037] Second, handle the up and down offset:

[0038] If the intersection of the first puncture is not in the center of the pedicle midline, but has an upward or downward offset (distance a), then the target puncture point needs to be "translated" to compensate;

[0039] The translation distance calculation formula is c = (L / d)*a,

[0040] Translation direction rule: if the initial intersection point is above the midline, if the target point needs to be hit below, it needs to be translated to the right; if the target point needs to be hit above, it needs to be translated to the left;

[0041] If the initial intersection point is below the midline, the rule is completely opposite;

[0042] Thirdly, if there is an angle deviation (the center guide needle forms an angle A with the pedicle midline), the angle deviation needs to be corrected, and the compass needs to be translated so that the guide needle direction is parallel to the pedicle midline:

[0043] The distance of translation is calculated by the formula:

[0044]

[0045] T represents the distance the compass needs to be translated, G represents the length of the guide needle, and A represents the angle A formed by the center guide needle and the ideal pedicle midline;

[0046] The inner diameter of the compass is equal to G.

[0047] The beneficial effects of the present application are: unlike the prior art, the transpedicular puncture intelligent navigation system provided by the present application comprises: a perception and modeling module, which realizes real-time construction of a dynamic surgical area digital twin model and prediction of dynamic change trends in the surgical process by fusing multiple sources of intraoperative information; a decision and planning module, which, based on the digital twin model constructed by the perception and modeling module, comprehensively considers safety, accuracy, stability, long-term efficacy and other multiple complex objectives, and realizes planning of a globally optimal and environment-change-adaptive "intelligent" puncture strategy for a surgical puncture device; a control and execution module, which realizes sub-millimeter-level trajectory tracking accuracy and high-robustness precise operation in a complex and uncertain real surgical physical environment. Through the above intelligent navigation system and the navigation device applying the system, the problems of the current transpedicular puncture navigation system, such as inability to adapt to dynamic changes in the operation, lack of real-time feedback and closed-loop control capability, and single dimension of path planning, are solved, thereby significantly improving the safety, efficiency, accuracy, simplicity, reducing the exposure to radiation and reducing the infection rate of high vertebral body surgical puncture, and improving the efficacy. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0049] Figure 1 is a structural schematic diagram of an embodiment of the transpedicular puncture intelligent navigation system provided in the present application;

[0050] Figure 2 is a structural schematic diagram of an embodiment of the perception and modeling module provided in the present application;

[0051] Figure 3 is a workflow schematic diagram of an embodiment of the multi-modal data fusion and non-rigid three-dimensional reconstruction unit provided in the present application;

[0052] Figure 4 is a structural schematic diagram of an embodiment of the decision and planning module provided in the present application;

[0053] Figure 5 is a workflow schematic diagram of an embodiment of the transpedicular puncture navigation device provided in the present application. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It can be understood that the specific embodiments described herein are only used to explain the present application, rather than limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, rather than all the structures. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0055] Reference to "an embodiment" in this text means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase appears at various places in the specification intends to encompass the understanding that the particular embodiment described at that location is included in at least one embodiment of the present application. It is explicitly and implicitly understood that the embodiments described herein can be combined with other embodiments.

[0056] Now various spine minimally invasive surgery is popular, and is pursued by patients and doctors. The core technology and key link of minimally invasive surgery is safe, fast and accurate targeting puncture to the lesion site. The best path for safe and accurate puncture of intravertebral lesions is through the pedicle, avoiding other important tissues and organs. The pedicle structure is a kind of incomplete regular oblate cylindrical bone structure (the upper and lower diameter is the widest, and the front and back diameter is the narrowest), and the vertebral body is an incomplete regular cylindrical bone structure. The puncture range of the pedicle to the intravertebral lesion site is limited (the posterior and lower part of the vertebral body and a small amount of the posterior and upper part cannot be punctured through the pedicle), and in the puncturable part, accurate puncture needs to strictly plan the puncture route, which is extremely difficult to operate, especially for doctors with little experience. Moreover, the risk is also extremely high. Even with the guidance of intraoperative related equipment, the radiation exposure will increase significantly, the operation time will be significantly prolonged, the pollution probability will be greatly increased, and the safety will be severely tested.

[0057] The existing pedicle puncture surgery navigation system has the following four technical problems:

[0058] 1. Unable to adapt to intraoperative dynamic changes: In actual surgical procedures, the patient's voluntary respiratory movement, muscle relaxation or tension under anesthesia, and the pulling and extrusion of surgical instruments on the surrounding soft tissue will all cause slight but critical changes in position and posture of the target vertebra and its surrounding anatomical structures. These changes are dynamic and nonlinear, making the path planned based on preoperative static data no longer accurate at the moment.

[0059] 2. Lack of real-time feedback and closed-loop control capability: This navigation method is essentially an open-loop or semi-open-loop system. It performs path planning once before executing the puncture, and even with the correction function, it is only limited to one-time compensation for the initial positioning error. During the puncture process, the system cannot sense and respond to real-time changes, such as changes in bone density encountered by the instrument tip, accidental tissue sliding or slight movement of the patient. This "blind execution" mode makes it less robust and unable to ensure continuous safety during the process.

[0060] 3. Single dimension of path planning: Its planning is purely based on "accessibility" and "safety" in geometry, without integrating deeper information such as biomechanics. For example, it cannot assess the quality of the microstructure of the bone on different puncture paths, nor can it choose a path that can obtain the maximum holding force, nor can it predict and avoid high-risk areas that temporarily appear on the path due to tissue deformation.

[0061] 4. Lack of decision-making ability and non-evolutionary: the system is a sophisticated slide rule, not an intelligent agent with decision-making ability. Its performance upper limit is completely locked at the time of design completion, and it cannot learn and evolve from successful or failed surgery cases, and it does not have the ability to optimize its strategy from experience.

[0062] Therefore, the present application aims to solve the above-mentioned core technical problems, i.e. how to break through the fundamental limitation of traditional navigation technology relying on static and rigid anatomical models, and construct a new technical system capable of intelligent and adaptive puncture path planning and high-robustness closed-loop control in a dynamic and uncertain real surgery environment by fusing multi-source heterogeneous information. For details, see the following embodiments.

[0063] Referring to Figure 1 , Figure 1 is a structural schematic diagram of an embodiment of the intelligent navigation system provided by the present application. The system includes three very core modules:

[0064] Module 1: perception and modeling module 10. The core task of this module is to completely overturn the idealized assumption of "static" and "rigid" of the patient's anatomical structure in the background technology. It fuses multi-source and heterogeneous intraoperative information to construct and dynamically refresh a high-fidelity surgery "digital twin" model in real time. This model not only accurately reproduces the current anatomical morphology of the patient, but also predicts the dynamic change trend during the surgery, providing unprecedented accurate environmental perception capability for subsequent intelligent decision-making and path planning.

[0065] Referring to Figure 2 , Figure 2 is a structural schematic diagram of an embodiment of the perception and modeling module provided by the present application.

[0066] In some embodiments, the perception and modeling module 10 includes a multi-modal data fusion and non-rigid three-dimensional reconstruction unit 11 for accurately driving a high-precision preoperative three-dimensional model to deform to match the real dynamics of the patient in surgery using limited, low-dose real-time images; and a hybrid dynamic modeling unit 12 combining physical laws and data-driven methods for predicting change trends during surgery.

[0067] Regarding the multi-modal data fusion and non-rigid three-dimensional reconstruction unit 11, the problem of how to accurately drive a high-precision preoperative three-dimensional model to deform to match the real dynamics of the patient in surgery using limited, low-dose real-time images is solved.

[0068] We introduce an efficient reconstruction idea for handling incomplete data here, which can be popularly compared to a high-dimensional data completion problem. Specifically:

[0069] Complete but "outdated" high-dimensional dataset: High-resolution 3D CT or MRI scans taken preoperatively for a patient can be considered as a high-dimensional data matrix containing complete anatomical information, but its timestamp stays at pre-operation.

[0070] Sparse but "fresh" accurate samples: During the operation, in order to reduce the radiation dose, we only take a small amount of low-dimensional two-dimensional C-arm X-ray fluoroscopy images when necessary. These images, although sparse in information (only two-dimensional projection of three-dimensional structure at a certain angle), accurately reflect the true position of the patient's anatomy at a certain moment, and are "fresh" and highly reliable observation samples.

[0071] Referring to Figure 3 , Figure 3 is the workflow schematic diagram of an embodiment of the multi-modal data fusion and non-rigid three-dimensional reconstruction unit provided by the present application.

[0072] The core algorithm of the multi-modal data fusion and non-rigid three-dimensional reconstruction unit 11 is to solve the problem of calibrating "complete but outdated data set" with "sparse fresh samples" through an iterative alternating projection mechanism. The working process is as follows:

[0073] 11A. Data constraint projection, project the current three-dimensional digital model according to the simulated X-ray imaging physical process to the same observation angle as the real-time C-arm image, forming a simulated two-dimensional image. Then, calculate the difference between the simulated image and the real C-arm image, and use this difference as the driving force to adjust the morphology of the three-dimensional model, so that the projection result and the real observation achieve the best match;

[0074] 11B. Physical / anatomical constraint projection: In the adjustment of the last step, the three-dimensional model may appear distorted or deformed that does not conform to the anatomical rules. In this step, the system projects the deformed model into a "conforming to the anatomical and biological rules" constraint space. This means that the system will smooth and correct the unreasonable deformation according to the built-in anatomical prior knowledge (such as the incompressibility of bones and the elasticity range of ligaments), to ensure the authenticity of the model.

[0075] Through rapid iterative alternating projection between the two constraints, the system can find a high-fidelity three-dimensional dynamic model that can perfectly match the intraoperative sparse observation data and completely conform to the anatomical logic within a few iterations. This technique creatively solves the industry problem of achieving high-precision, real-time three-dimensional dynamic reconstruction while minimizing intraoperative radiation exposure.

[0076] Regarding the hybrid dynamic modeling unit 12 of physical laws and data-driven:

[0077] It is not enough to be able to reconstruct the "current" dynamic model, a truly intelligent system also needs to be able to predict the model changes in the "next instant". To this end, we introduce the idea of modeling based on physical laws. We consider the dynamic changes in the surgical area as a complex system of differential algebraic equations. The differential equation part: describes the physical processes in the system that change continuously over time, for example, the changes in the thoracic and abdominal cavity pressure driven by the patient's autonomous breathing, which in turn leads to periodic micro-displacement of the spine. The algebraic equation part: describes the physical constraints that must be met at all times in the system, for example, the incompressibility of soft tissue, the rigid constraints when surgical instruments come into contact with bones, etc. Directly solving such a complex system of differential algebraic equations is extremely computationally expensive. Therefore, we draw on the advanced idea of approximating the energy function of the system. Instead of directly solving the equation set, the system approximates the "total energy function" of the entire physical system through efficient numerical methods. This energy function can comprehensively reflect the potential energy and kinetic energy of the system in the current state. According to the principles of physics, any system tends to evolve in the direction of decreasing total energy. Therefore, by calculating the gradient of this energy function with respect to various variables (such as the force applied by the instrument, the phase of the breathing cycle), the system can quickly predict how the entire anatomical model will deform and displace in the next infinitesimal time step. This method upgrades the evolution of the model from a purely "data-driven" mode that relies on intermittent images to a hybrid-driven new mode that is "primarily predicted by physical laws and supplemented by real-time data calibration". It greatly improves the temporal continuity and prediction accuracy of the digital twin model between two image acquisitions.

[0078] The core goal of the perception and modeling module 10 proposed in this application is no longer to calculate a simple scalar "adjustment distance", but to solve an extremely complex function "optimal deformation field (space coordinate, time)". This function describes the spatial displacement that each point in the preoperative model should undergo at each moment during the operation, so that it dynamically evolves into a digital twin that is completely consistent with the patient's real situation. The solution of this optimal deformation field can be achieved by minimizing a total energy functional:

[0079] Optimal deformation field (space coordinate, time) = argmin [data fidelity term + physical constraint term + spatial smoothing regularization term]

[0080] Where,

[0081] 1. Data fidelity term = integrate (||projector (preoperative 3D model + deformation field) - intraoperative 2D image||2) di image over all intraoperative images;

[0082] Description: This term is the core driving force of the entire optimization problem. The "projection operator" is a mathematical transformation that simulates the imaging process of X-rays. The physical meaning of this term is: find a "deformation field" such that the three-dimensional model after the action of the "deformation field" has the smallest difference between its two-dimensional projection and the real-time acquired C-arm X-ray image. This directly reflects the idea of multi-modal data fusion.

[0083] 2. Physical constraint term = physical constraint weight x integral (differential algebraic system energy (deformation field, rate of change of deformation field with respect to time)) dtime

[0084] Description: This term ensures the physical reality of the deformation process. It is derived from the modeling of the physical laws of the system, and punishes deformations that can match the image but do not conform to the laws of biomechanics (such as bones being abnormally stretched or soft tissues being excessively compressed). The "physical constraint weight" is a hyperparameter that balances the importance between data matching and physical laws.

[0085] 3. Spatial smoothing regularization term = regularization weight x integral (||gradient operator (deformation field)||2) dspace

[0086] Description: This term ensures the smoothness and continuity of the deformation, avoiding unrealistic and cliff-like mutations on the model. It is a mathematical constraint that ensures the stability and uniqueness of the solution.

[0087] By solving this highly creative functional minimization problem, this module elevates a simple geometric positioning problem to a complex four-dimensional variational problem that integrates multi-modal data, physical laws, and prior knowledge. The "optimal deformation field" output by it provides an unprecedented, dynamic, accurate, and predictive "live" battlefield map for the subsequent decision and control modules of the entire intelligent navigation system, fundamentally forming a solid foundation for the creativity and non-obviousness of this scheme.

[0088] Module 2: Decision and Planning Module 20, this module is the "decision center" of the entire intelligent navigation system, which completely abandons the simple mode of path calculation based on single geometric rules in the background technology. Its core task is to consider safety, accuracy, stability, and long-term efficacy and other multiple complex goals on the high-fidelity dynamic digital twin model constructed by the perception and modeling module, and plan an "intelligent" puncture strategy that is globally optimal and can adapt to environmental changes in real time.

[0089] Reference Figure 4 , Figure 4 is a structural schematic diagram of an embodiment of the decision and planning module provided by the present application.

[0090] In some embodiments, the decision-making and planning module 20 includes a global path corridor planning unit 21 based on optimal control theory, which plans the optimal path of surgery by dynamically calculating the probabilistic distribution during the surgery process, and a sequential decision-making unit 22 based on hierarchical deep reinforcement learning, which constructs an intelligent decision-making system by combining optimal control theory with hierarchical deep reinforcement learning.

[0091] Regarding the global path corridor planning unit 21 based on optimal control theory, the core innovation lies in:

[0092] 1. From finding a "deterministic path" to solving a "probabilistic distribution": Instead of trying to calculate the illusory and unique "optimal line", we transform the problem into finding an optimal probability measure distribution among all possible puncture paths. This distribution describes the probability of the optimal path appearing in a certain location in three-dimensional space.

[0093] 2. Convert non-convex problem to convex optimization problem: Directly finding the optimal path is a highly complex non-convex optimization problem, which is extremely difficult to calculate and prone to local optimal solutions. The measure relaxation technique cleverly relaxes and converts this non-convex path finding problem into a convex optimization problem that can be efficiently and stably solved in polynomial time.

[0094] The solution to this convex optimization problem is no longer a line, but a "optimal safe corridor" that appears as a tube or strip distribution in three-dimensional space. The central region of this "corridor" corresponds to the highest probability of the optimal path existing; while the closer to the boundary of the "corridor", the lower the probability, meaning the higher the risk.

[0095] Regarding the sequential decision-making unit 22 based on hierarchical deep reinforcement learning:

[0096] Within the "optimal safe corridor" defined by optimal control theory, we need a more detailed mechanism to make specific actions at each step. The entire puncture process is essentially a sequential decision-making problem: the current small operation (e.g., moving forward 0.1 mm or adjusting the angle by 0.1 degrees) will directly affect the state and available operations in the next step, and ultimately determine the success or failure of the entire surgery.

[0097] Therefore, we introduce a "hierarchical deep reinforcement learning framework" that includes two intelligent agents working together and having different decision-making granularity and time scales:

[0098] 1. Strategic planning layer

[0099] Responsibility: Responsible for macroscopic, global strategic planning. Its input is the entire "safety corridor" geometry, the overall bone density distribution map of the pedicle in the digital twin model, and the relative positions of key nerves and blood vessels.

[0100] Decision output: It does not output specific movement instructions, but a series of strategic target points, such as the optimal puncture window center point on the cortical bone, the target point that needs to be precisely passed through the narrowest part of the pedicle isthmus, and the safety target point that needs to stop at the anterior edge of the vertebral body.

[0101] Learning goal: The goal of its learning and optimization is to maximize the long-term cumulative reward of the entire puncture task, which includes surgical success rate, long-term biomechanical stability of the screw, and avoidance of surrounding key tissues.

[0102] 2. Tactical execution layer

[0103] Responsibility: Responsible for fine, step-by-step tactical trajectory planning from one strategic point to the next under the guidance of high-level strategy. Its input is the real-time position and attitude of the current surgical instrument tip, as well as the real-time tissue resistance sensed through torque sensors.

[0104] Decision output: It outputs specific surgical puncture device movement instructions, such as "in the next 0.1 seconds, advance 0.5 mm along the current axis while adjusting the pitch angle by 0.2 degrees".

[0105] Learning goal: Its learning goal is to maximize the immediate reward of safely, accurately, and efficiently reaching the next strategic point, while minimizing energy consumption and tissue disturbance during the process.

[0106] In some embodiments, in order for the above two reinforcement learning agents to learn the correct behavior, we need to design a reward function that can accurately quantify the "goodness" of each step of operation. This function is the "soul" of the entire learning process, and it must accurately reflect the complex decision-making logic of the surgeon. This reward function is likely to be non-smooth, i.e. at some key points or boundaries, the reward value will change abruptly, and its derivative does not exist. For example: when the instrument tip is more than a certain safety threshold from the nerve boundary, the risk penalty may be 0; once it is less than the threshold, the penalty term will immediately become a huge complex number. When the puncture trajectory passes through an osteoporotic area, the reward value will decrease; when it passes through a dense bone area, the reward value will increase, and this change may also be stepwise.

[0107] It is very difficult and unstable to directly perform gradient optimization on such a non-smooth reward function to train a neural network. For this purpose, we draw on the ideas in optimization algorithms that deal with non-smooth problems such as piecewise smoothness. Instead of directly using this non-smooth reward function, we construct a smooth, differentiable proxy function to approximate it, or at the optimization algorithm level, use a special optimizer that can stably handle non-smooth gradients. This ensures that the reinforcement learning agent can efficiently and stably learn and converge in complex reward topography, and ultimately internalize a robust decision-making ability that can handle various boundaries and mutations.

[0108] The decision and planning module 20 proposed in the present application is mathematically the solution of an optimal strategy function "π(next action | current state)". This function describes that in any given current state, what action can bring the maximum long-term return. This optimal strategy is defined by a hierarchical, nested Bellman optimality equation:

[0109] Optimal high-level strategy (strategic target point | global state) = argmax [high-level immediate reward + γE (optimal high-level value function (next global state))]

[0110] Where the optimal low-level strategy (tactical fine-tuning action | local state, strategic target point) = argmax [low-level immediate reward + γE (optimal low-level value function (next local state))] The final solution "π" of this nested equation set is obtained by minimizing a complex policy loss function with non-smooth terms:

[0111] Optimal policy network parameters = argmin [E_data (log (policy network output probability) x (advantage function)) + β x entropy regularization where,

[0112] 1. Advantage function = Q-value network output (state, action) - V-value network output (state)

[0113] Note: This is a core concept in reinforcement learning, used to evaluate how much better an action is than the average action in the current state. "Q-value network" and "V-value network" are two other neural networks used to assist policy network training.

[0114] 2. Entropy regularization term = -integral (policy network output probability x log (policy network output probability)) daction

[0115] Note: This term encourages the policy network to maintain a certain degree of exploration and avoid converging too early to a local optimal deterministic policy, which is crucial for handling complex multi-modal decision spaces. "β" is a hyperparameter that controls the degree of exploration.

[0116] 3. The embodiment of non-smooth reward: The training target of the Q-value network and V-value network in the above formula is to fit the cumulative return defined by the reward function. Since the reward function itself is non-smooth, we draw on the idea of non-smooth optimization, and use special loss functions or optimization algorithms when training these two networks, for example, using proximal gradient descent method that can handle points where gradients do not exist, to ensure the stability and convergence of training.

[0117] Therefore, the decision and planning module 20, by combining optimal control theory with hierarchical deep reinforcement learning, creatively builds an intelligent decision-making system that "first strategy and then tactics". It decomposes a high-dimensional and complex path planning problem into a series of interrelated but more easily handled sub-problems, and uses advanced optimization techniques that can handle non-smooth rewards for training. The final output is no longer a rigid path, but an intelligent surgical strategy that can adapt to dynamic environments, balance multiple goals, and has self-optimization capabilities.

[0118] Module three: control and execution module 30.

[0119] This module is the "neural reflex system" of the entire intelligent navigation system, and it plays a key role in controlling the precise, stable, and safe physical movements of the surgical puncture device based on the intelligent surgical strategy output by the upper "decision and planning module". The execution part in the background technology only relies on the rigid constraints of mechanical structure, and once it encounters unexpected physical interactions, it is helpless. The core task of this module is to make up for this fatal flaw and ensure that the system can still achieve sub-millimeter level trajectory tracking accuracy and high operation robustness in complex and uncertain real surgical physical environments.

[0120] Technical core: disturbance rejection control based on dead zone adaptation.

[0121] Specifically, in an ideal situation, the servo motor system of the surgical puncture device will perfectly execute every micro-instruction (i.e., the ideal reference trajectory) issued by the decision module. However, in real surgical scenarios, there are a large number of dynamic disturbances that are difficult to accurately model or even completely unknown, which will continuously interfere with the movement of the puncture device and make it deviate from the predetermined track.

[0122] These disturbances mainly originate from: 1. Patient physiological disturbances: Even under general anesthesia, patients still experience physiological activities such as heartbeat, weak spontaneous breathing, or muscle tremors, all of which cause minute, high-frequency vibrations in the surgical area. 2. Tissue mechanical disturbances: During puncture, surgical instruments encounter uneven bone tissue density. For example, when moving from loose cancellous bone to dense cortical bone, the tissue reaction force increases instantaneously, generating an impactful disturbance torque on the surgical instrument. 3. Internal system disturbances: Joint friction of the puncture device itself, minute fluctuations in motor torque, and sensor measurement noise also affect the final execution accuracy. Traditional controllers (such as PID controllers) often require fine parameter tuning and have limited robustness when dealing with these complex, time-varying disturbances. To fundamentally solve this problem, this module introduces and deeply applies an advanced modern control theory—dead-zone adaptive disturbance suppression control. It possesses characteristics such as independence from precise disturbance models, adaptive gain adjustment, and a dead-zone mechanism to prevent overreaction.

[0123] In some embodiments, the transpedicle puncture intelligent navigation system simultaneously employs a surgical strategy function π to minimize risk:

[0124] The optimal policy π = argmin_{π∈policy space}J(π);

[0125] J(π)=E_{all possible surgical procedures} }[integral_{from the start to the end of the surgery}(L(state(t), action(t)))dt+Φ(final state)];

[0126] The input to the strategy function is all the sensing information of the system at any time t, and the output is the action instruction that the surgical puncture device should execute at that time.

[0127] Explanation of the formula:

[0128] 1. Expectation operator E_{all possible surgical procedures} [...]

[0129] Meaning: This expectation operator E is the top-level structure of the entire formula, reflecting the system's profound understanding of uncertainty. It indicates that our optimization objective is not to achieve the best result in a single "perfect" surgical scenario, but rather to minimize the overall risk by averaging the probabilities of all possible surgical procedures (including different patient anatomy, different intraoperative perturbations, and different sensor noise).

[0130] Implementation: This expectation is approximated through extensive simulation training and learning from historical data. The policy π itself is stochastic (e.g., in the exploration phase of reinforcement learning), thus guiding the surgical process. It is also a stochastic process.

[0131] 2. The instantaneous risk function L(state(t), action(t))

[0132] Meaning: This is the core integral function of the formula, representing the instantaneous risk that the system takes at any time t during the surgery. It is a composite function that internally couples the performance of the perception, decision, and control modules.

[0133] Creative expansion: L(state(t), action(t)) = α[data fidelity term + physical constraint term] + β[norm of trajectory tracking error] + δ[energy consumption of action] + penalty function (safety margin)

[0134] α[...]: Risk of perception modeling

[0135] The data fidelity term and the physical constraint term are directly derived from the functional formula of the perception and modeling modules. Including them in the instantaneous risk means that if the output of the perception and modeling modules (i.e., the digital twin model) does not match the real observation or does not conform to the physical laws, the system will immediately generate a risk signal. This forces the strategy π to generate actions that allow the perception system to work in the "most comfortable" (i.e., most accurate) state. For example, avoid those instrument poses that will produce serious artifacts or occlusions.

[0136] α is the weight coefficient of the perception risk.

[0137] β[...]: Risk of control execution

[0138] The norm of the trajectory tracking error is directly derived from the core performance indicator of the control and execution modules. It measures the deviation between the actual action of the surgical puncture device and the instructions issued by the strategy π. Including it as a risk term means that the strategy π must consider the limitations of physical execution when learning, and it will naturally avoid those "tricks" that are theoretically optimal but physically difficult to accurately implement, thus choosing a smoother and more robust path.

[0139] β is the weight coefficient of the execution risk;

[0140] δ[...]: Economic risk of decision planning;

[0141] The energy consumption of the action represents the energy or resources required to execute an action (such as motor power, surgery time), which penalizes unnecessary and intense operations, making the final generated strategy smoother and more efficient;

[0142] δ is the weight coefficient of the economic risk;

[0143] Penalty function (safety margin): safety risk of decision planning. This penalty function is a non-smooth term, which is directly related to the "safety corridor" defined in the decision planning module and the boundary of critical anatomical structures (e.g. nerves). When the state (t) (including instrument position and anatomical structures) approaches or touches these safety margins, the value of this function will jump from 0 to a huge positive number. This imposes the most stringent hard safety constraints on the whole optimization problem.

[0144] 3. Terminal risk function Φ (final state)

[0145] Meaning: This function Φ is used to evaluate the final outcome at the end of the whole surgery (t = end).

[0146] Expansion: Φ (final state) = ω_1∥final puncture position - ideal target position∥2+ ω_2(1 - screw holding force prediction score) +...

[0147] It contains a comprehensive evaluation of multiple dimensions, such as final accuracy (distance from ideal target), effectiveness (screw holding force predicted according to final position and surrounding bone density), etc.

[0148] ω_1, ω_2 are the weights of each terminal risk.

[0149] Embodiment of logical closed loop: This overall comprehensive formula is no longer a simple splicing of three independent module formulas, but an organic and mutually coupled whole. It deeply embodies the logical closed loop of the whole system:

[0150] In order to minimize the total expected risk J (π), the optimal strategy π must make trade-offs at every step.

[0151] It must generate an action that not only minimizes the expected value of future terminal risk Φ (decision planning goal), but also keeps the risk of perception modeling α [...], and the risk of control execution β [...], at a low level during execution.

[0152] This means that a good decision must be a "perceptible, executable, effective and safe" decision. If a path, although theoretically the shortest, will cause artifacts in C-arm images (high perception risk), or require the puncture device to make extreme actions at the edge of its capability range (high execution risk), then the total risk L of this path will be large, and ultimately will be abandoned by the strategy π.

[0153] Finally, by solving this robust optimal policy functional, we obtain a π that is no longer just a simple navigation instruction generator, but a truly intelligent surgical specialist that internalizes a deep understanding of the physical world uncertainty, perception limitations, and control errors. It can stably, safely, and precisely complete the challenging task of transforaminal lumbar puncture in complex reality.

[0154] The application also provides a transforaminal lumbar puncture navigation device using the transforaminal lumbar puncture navigation system, the device further comprising: a puncture device electrically connected with the intelligent navigation system, and the intelligent navigation system controls precise operation of the puncture device.

[0155] Referring to Figure 5 , Figure 5 is a working flow diagram of an embodiment of the transforaminal lumbar puncture navigation device provided by the application.

[0156] The working steps are:

[0157] 102. Determine the target position of the vertebral body lesion site according to imaging, and determine the number of target points according to the specific puncture needs of the clinic, and input and mark in the transforaminal lumbar puncture navigation device;

[0158] 104. Use the transforaminal lumbar puncture intelligent navigation system to calculate the maximum range of bilateral transforaminal puncture of the vertebral body, that is, to determine the maximum projection range of bilateral transforaminal puncture in the vertebral body, and determine whether the target point site of the vertebral body lesion in the previous step is covered by the maximum projection range. If not, give up this puncture method; if yes, proceed to the next process;

[0159] 106. Dynamic programming of needle operation path;

[0160] 108. The puncture device performs a puncture action.

[0161] In some embodiments, the puncture device comprises: a guide needle for transforaminal puncture; a compass placed on the surface of the patient's back skin during surgery for guiding the puncture direction and position of the guide needle, the guide needle passing through the center of the compass and forming an initial positioning; and a stabilizing component for fixing the compass at a predetermined position and supporting the movement of the compass.

[0162] Wherein, the positioning step of the compass is:

[0163] First, determine the puncture range (the length of the transforaminal puncture is L and the width is d):

[0164] When the length / width = 2 (L / d = 2): At this time, the puncture intersection is located at the midpoint of the transforaminal puncture, and the puncture range on the compass is exactly the inner circle range, without the need for scaling;

[0165] When length / width > 2 (L / d > 2): the pedicle is relatively "thin and long", the puncture range on the compass needs to be scaled down, and the calculation formula of the scaled radius is r = (d / L) * G, G represents the length of the guide needle;

[0166] When length / width < 2 (L / d < 2): the pedicle is relatively "short and thick", the puncture range on the compass needs to be scaled up, and the calculation formula of the scaled radius is also: r = (d / L) * G, G represents the length of the guide needle;

[0167] Second step, handle up and down offset:

[0168] If the intersection point of the first puncture is not in the center of the pedicle midline, but has an upward or downward offset (distance a), then the target puncture point needs to be "translated" to compensate;

[0169] Translation distance calculation formula: c = (L / d) * a,

[0170] Translation direction rule: if the initial intersection point is above the midline, if you need to hit the lower target point, you need to translate to the right; if you need to hit the upper target point, you need to translate to the left;

[0171] If the initial intersection point is below the midline, the rule is completely opposite;

[0172] Third step, if there is an angle deviation (the central guide needle forms an angle with the pedicle midline, the degree is A), then the angle deviation needs to be corrected, and the compass needs to be translated so that the guide needle direction is parallel to the pedicle midline:

[0173] The distance calculation formula for translation is:

[0174]

[0175] T represents the distance the compass needs to be translated, G represents the length of the guide needle, and A represents the angle between the central guide needle and the ideal pedicle midline;

[0176] The inner diameter of the compass is equal to G.

[0177] Specific application example:

[0178] The pedicle structure is a kind of incomplete regular elliptical flat cylindrical bone structure, and the vertebral body is an incomplete regular cylindrical bone structure, one horizontal and one vertical. Through one vertical (pedicle) to reach the accurate position in one horizontal (vertebral body).

[0179] First, according to the theoretical angle of the pedicle puncture theory, the theoretical maximum range of the pedicle puncture and the maximum puncture angle can be calculated. That is, the maximum projection range of the pedicle from the distal end to the proximal end, that is, the entrance to the exit in the vertebral body. The positioning compass of the puncture device is designed to achieve this theoretical maximum range and maximum projection range.

[0180] We imagine the midpoint of the center line of the incomplete regular elliptical and cylindrical structure of the pedicle as the center of a sphere. Starting from a point on the outer surface of the sphere and passing through the center, we can reach a specific point in the vertebral body. These ranges are specific and are directly related to the inner diameter and length of the pedicle. According to the application of the theoretical angle of the pedicle puncture theory, the final calculation is performed. When the diameter of the compass is equal to the distance to the midpoint of the center line of the pedicle, by setting different conditions, even if the first puncture position deviates from the center line, the method described in this application can completely achieve the conclusion of precise puncture target. This greatly improves the practicality of the compass, and it is not necessary to completely insert the center guide needle along the center line.

[0181] Set a compass (two concentric circles) with an inner diameter of 10 cm (an outer diameter of 20 cm), a center guide needle length of 10 cm, and all guide needle tips intersecting at the center guide needle tip, i.e., at a depth of 10 cm. There is a guide hole on the upper surface of the compass every 1 cm (numbered in order from the near center to the far center). The guide channel (diameter 3 mm) has a thickness (i.e., the shortest guide channel length) of 2 cm. All other guide needles are longer than 10 cm when they reach the intersection point from the compass surface. The farther away from the center guide needle, the longer the distance. All guide needles intersect at the center guide needle tip, i.e., at a depth of 10 cm.

[0182] Combined with the pedicle puncture theory angle theory, according to preoperative imaging determination, according to the inclination angle and caudal inclination angle of the pedicle, the center puncture needle of the compass is completely inserted into the middle segment of the pedicle along the long axis of the pedicle. If it can enter and reach the midpoint along the center line, then it is handled according to the general situation; if it deviates from the center line, then it is handled according to the special situation.

[0183] I. General situation, if the center guide needle of the compass is located on the center line of the pedicle when puncturing for the first time, then the puncture needle tip is stopped at the midpoint of the center line. There are three cases:

[0184] 1. The length of the pedicle is always the same, and the ratio of its transverse diameter to the upper and lower diameter, i.e., the widest diameter (width) (length / width) is equal to 2, then the intersection point is located at the midpoint of the center line of the pedicle, and all puncture ranges are exactly within the inner circle of the compass. Calculate the target position and puncture directly.

[0185] 2, the ratio of pedicle length to transverse diameter (width) is greater than 2; the intersection point is located at the midpoint of the pedicle midline, and all puncture ranges are exactly within the compass inner circle, which is scaled down, and the final radius range after scaling down is calculated: the long width value divided by the diameter length value, multiplied by 10 cm. Similarly, the pedicle flat diameter belongs to the case where the ratio is greater than 2, and the flat diameter direction is correspondingly scaled down to the puncture radius range: the short width value divided by the diameter length value, multiplied by 10 cm.

[0186] 3, the ratio of pedicle length to transverse diameter (width) is less than 2; the intersection point is located at the midpoint of the pedicle midline, and all puncture ranges are exactly outside the compass inner circle, which is scaled up, and the final radius range after scaling up is calculated: the long width value divided by the diameter length value, multiplied by 10 cm. Similarly, the pedicle flat diameter is shorter than the long diameter, and the ratio of the pedicle length to the flat diameter may be greater than 2, less than 2, or equal to 2 (the three cases above may occur, but the puncture range of the three cases above may be enlarged, scaled down, or exactly relative to the compass inner circle). The total formula for the puncture radius range of the flat diameter direction is still: the short width value divided by the diameter length value, multiplied by 10 cm.

[0187] According to the above three cases, we deduce that no matter which of the three cases above is, as long as our puncture intersection point intersects the midpoint of the pedicle midline, our compass puncture radius range is: (width / length)*10. At the same time, we conclude that the puncture range of the compass is the reverse projection of the pedicle on the compass. We can conclude that the puncture range on the compass is completely opposite and consistent with the range in the vertebral body, which is the complete diagonal theory, and the reverse projection of the pedicle on the compass, that is, the maximum projection range of the pedicle on the compass from the nearest end to the farthest end, that is, the exit to the entrance.

[0188] II. Special cases, if the compass center needle deviates from the pedicle midline but does not form an angle with the midline, it is in a parallel state, then first measure the distance from the midline (a cm), then also in three cases:

[0189] 1. The ratio of the pedicle length to the widest diameter (length / width) of the transverse section is equal to 2, and the intersection point is located either a cm above or below the midpoint of the pedicle midline. If the puncture is performed directly as before, the puncture range will be deviated. According to the mathematical theory of geometric translation, calculations show that we only need to shift the puncture intersection point from the midpoint towards or away from the vertebral body by a specific distance (2a cm) before puncturing along the compass to achieve a complete overlap with the original puncture side. The projection of this side on the compass corresponds diagonally to the projection within the vertebral body, allowing for precise puncture on that side using the previous method. However, this depends on whether the target point within the vertebral body is above or below the puncture line (i.e., puncturing upwards or downwards): If it is below (i.e., puncturing downwards), the intersection point needs to be shifted to the right, away from the vertebral body, so that the puncture range above the compass reaches the inner circle edge, allowing for precise downward puncture; if it is above (i.e., puncturing upwards), the intersection point needs to be shifted to the left, closer to the vertebral body, so that the puncture range below the compass reaches the inner circle edge, allowing for precise upward puncture. Similarly, if the pedicle flat diameter (anteroposterior diameter, narrowest diameter) is greater than 2, then the puncture range is reduced proportionally in the flat diameter direction, and the puncture intersection is shifted, which is the same as the second case below (see the second case).

[0190] Similarly, if the puncture intersection point is below the pedicle midline, the translation process is completely reversed. If it is below (i.e., downward puncture), the intersection point needs to be translated to the left, away from the vertebral body. The puncture range above the compass will then reach the inner circle edge, allowing for precise downward puncture. If it is above (i.e., upward puncture), the intersection point needs to be translated to the right, closer to the vertebral body. The puncture range below the compass will then reach the inner circle edge, allowing for precise upward puncture. Similarly, if the pedicle flat diameter (anteroposterior diameter, narrowest diameter) has a ratio greater than 2, the puncture range in the flat diameter direction should be proportionally reduced, combined with the method of translating the puncture intersection point, similar to the second case below (see case 2).

[0191] 2. If the ratio of the pedicle length to the widest diameter (length / width) of the transverse section is greater than 2, and the intersection point is located 'a' cm above or below the midpoint of the pedicle midline, the puncture range will deviate if the puncture is performed directly as before. In this case, similar to the second situation in general, the puncture range needs to be reduced proportionally. The final radius after reduction, calculated using the formula, is: length / width divided by diameter, multiplied by 10 cm. Similarly, if the ratio of the pedicle's flat diameter is greater than 2, the puncture range in the flat diameter direction should be reduced proportionally by: short width divided by diameter, multiplied by 10 cm.

[0192] After the calculation, we only need to move the intersection point from the midpoint to the vertebral body or away from the translation of a certain distance (length / width) * a cm, and then along the compass to achieve complete overlap with the original puncture side range, which is diagonally opposite to the projection in the vertebral body, so that this side can be accurately punctured according to the previous puncture method; but this needs to be divided into whether the target point in the vertebral body is above or below the puncture line (i.e. up or down puncture): if it is below (i.e. down puncture), the intersection point needs to be translated to the right, i.e. away from the vertebral body, then the compass puncture range above reaches the inner circle edge, and the downward accurate puncture can be achieved; if it is above (i.e. up puncture), the intersection point needs to be translated to the left, i.e. close to the vertebral body, then the compass puncture range below reaches the inner circle edge, and the upward accurate puncture can be achieved. Similarly, the pedicle diameter (anteroposterior diameter, narrowest diameter) belongs to the case where the ratio is greater than 2, and the puncture range in the flat direction is reduced in proportion and combined with the method of translating the intersection point of puncture.

[0193] Similarly, if the intersection point of puncture is below the midline of the pedicle, the translation is completely opposite, if it is below (i.e. down puncture), the intersection point needs to be translated to the left, i.e. away from the vertebral body, then the compass puncture range above reaches the inner circle edge, and the downward accurate puncture can be achieved; if it is above (i.e. up puncture), the intersection point needs to be translated to the right, i.e. close to the vertebral body, then the compass puncture range below reaches the inner circle edge, and the upward accurate puncture can be achieved. Similarly, the pedicle diameter (anteroposterior diameter, narrowest diameter) belongs to the case where the ratio is greater than 2, and the puncture range in the flat direction is reduced in proportion and combined with the method of translating the intersection point of puncture.

[0194] 3. The ratio of the pedicle length to the transverse diameter (width) is less than 2; and the intersection point is above or below the midpoint of the pedicle midline by a cm. If it is still punctured directly as before, the puncture range will deviate. First, when this situation occurs, it needs to be enlarged in proportion to the third case in the general case, and the final radius range after enlargement is obtained according to the corresponding formula: the length width value divided by the diameter length value, multiplied by 10 cm. Similarly, the pedicle diameter is shorter than the longer diameter, and the ratio of the pedicle length to the diameter may be greater than 2, less than 2, or equal to 2 (the three cases above may occur, but the puncture range of the three cases above relative to the compass inner circle may be enlarged, reduced, or unchanged). The total formula for the puncture range in the flat direction is still: the short width value divided by the diameter length value, multiplied by 10 cm; greater than 10 cm is enlarged, less than 10 cm is reduced, and equal to 10 cm is unchanged.

[0195] After calculation, we only need to move the intersection point of the puncture from the midpoint to the vertebral body or away from the vertebral body by a certain distance (length / width)*a cm, and then along the compass puncture to completely overlap the original puncture range on one side, which is diagonally opposite to the projection in the vertebral body, so that this side can be accurately punctured according to the previous puncture method; but this needs to be divided into whether the target point in the vertebral body is above or below the puncture line (i.e. puncture up or down): if it is below (i.e. puncture down), the intersection point needs to be translated to the right, i.e. away from the vertebral body, then the puncture range above the compass reaches the inner circle edge, and the accurate puncture downward can be achieved; if it is above (i.e. puncture up), the intersection point needs to be translated to the left, i.e. close to the vertebral body, then the puncture range below the compass reaches the inner circle edge, and the accurate puncture upward can be achieved. Similarly, the pedicle flat diameter (anteroposterior diameter, narrowest diameter) belongs to the case where the ratio is greater than 2, less than 2 or equal to 2, then the flat diameter direction is translated left or right or not according to the above three methods.

[0196] Similarly, if the intersection point of the puncture is below the midline of the pedicle, the translation situation is completely opposite, if it is below (i.e. puncture down), the intersection point needs to be translated to the left, i.e. away from the vertebral body, then the puncture range above the compass reaches the inner circle edge, and the accurate puncture downward can be achieved; if it is above (i.e. puncture up), the intersection point needs to be translated to the right, i.e. close to the vertebral body, then the puncture range below the compass reaches the inner circle edge, and the accurate puncture upward can be achieved. Similarly, the pedicle flat diameter (anteroposterior diameter, narrowest diameter) belongs to the case where the ratio is greater than 2, less than 2 or equal to 2, then the flat diameter direction is proportionally reduced, enlarged or unchanged in puncture range and the intersection point of the puncture is translated according to the corresponding method.

[0197] If the first puncture does not enable the center guide needle of the compass to be parallel to the midline, there is an angle between them, which is set as A degrees. The distance of the intersection point from the midpoint of the midline is still set as a cm, then if the center guide needle of the compass is adjusted to be parallel to the midline, and the intersection point falls at a cm from the midline. The distance of the compass needs to be adjusted to T cm, according to the sine theorem of right triangle, it is obtained that where G is the length of the guide needle. If A = 0 degrees, it is the case mentioned earlier. When A is not equal to 0 degrees, i.e. the first puncture is not parallel to the midline of the pedicle, then the compass is adjusted according to the above formula once, and then the puncture is positioned according to the new position of the compass, and the previous case is handled.

[0198] Therefore, in combination with the above cases and the case where the first puncture has an angle with the midline of the pedicle, we draw the following unified conclusion:

[0199] The length of the pedicle is L cm, the diameter of the pedicle is d cm, and the distance between the intersection point and the midline is a cm (positive above, negative below, and 0 on the midline). All three data can be measured preoperatively and intraoperatively. We set the inner circle diameter of the compass to 10 cm, and the center guide needle length to 10 cm.

[0200] The angle between the first compass center guide needle puncture and the pedicle midline is A degrees. Then, the second puncture adjusts the compass center guide needle parallel to the midline, and the distance T cm needs to be adjusted. Then After adjusting the compass, follow the previous procedure.

[0201] The diameter of our compass puncture range scaling is (2d / L)*10 cm. If 2d=L, it is completely within the inner circle compass and does not need to be scaled. If 2d>L, the puncture range is outside the inner circle compass and needs to be enlarged. If 2d<L, the puncture range is within the inner circle compass and needs to be scaled down.

[0202] The intersection point distance from the pedicle midline is a cm (positive above, negative below, and 0 on the midline). The intersection point needs to be translated by c=(L / d)*a cm. When a=0, no translation is needed, which is the first three general cases. When 2d=L, move 2a cm; when 2d>L, move less than 2a cm; when 2d<L, move more than 2a cm.

[0203] When a is positive, the puncture intersection point is above the midline, and the intravertebral target point is below the puncture line (i.e., downward puncture), so the intersection point needs to be translated to the right, i.e., towards the vertebral body. If it is above, the intersection point needs to be translated to the left, i.e., away from the vertebral body.

[0204] When a is negative, the puncture intersection point is below the pedicle midline, and the translation situation is completely opposite. The intravertebral target point is below the puncture line (i.e., downward puncture), so the intersection point needs to be translated to the left. If it is above, the intersection point needs to be translated to the right.

[0205] The complete and specific steps for precise vertebral puncture using the guide compass puncture device are as follows:

[0206] 1. Determine the vertebral lesion target position according to imaging (X-ray, CT, MRI), and determine the precise target point, which can be one or n, according to the clinical specific puncture needs, and input and mark in the system.

[0207] 2、According to preoperative imaging (X-ray and CT), the computer system is used to measure the maximum range of bilateral pedicle puncture of the vertebral body, that is, to determine whether the target site of the lesion in the vertebral body in the first step is covered by the maximum projection range. If not, this puncture method is abandoned. If it is covered, the next process is entered.

[0208] 3、Determine which side of the pedicle covers the target point, or both sides cover it, or both sides need to be combined to cover multiple targets.

[0209] 4、According to the third step, choose which side of the pedicle or both sides of the pedicle are needed. Only one side of the pedicle can be selected if only one side of the pedicle covers it. If both sides cover it, choose the side with a wider, more regular structure that is easier to puncture. Both sides must be selected if both sides are needed.

[0210] 5、According to the pedicle selected in step 4, use the compass center guide needle to puncture along the central axis of the pedicle for the first time. When the puncture angle is A degrees from the midline (A is 0, no adjustment is needed, A is not equal to 0, adjustment is needed), the compass center guide needle is adjusted to be parallel to the midline, and the adjustment distance is T cm. Then After adjustment, remove the first puncture guide needle.

[0211] According to the preoperative imaging, the in-clination angle and the caudal inclination angle of the pedicle are determined according to the pedicle theory angle. The compass center puncture needle is completely along the long axis of the pedicle and enters the middle segment of the pedicle. If it can enter the midpoint along the midline, it is handled as usual. If it deviates from the midline, it is handled as a special case.

[0212] 6、Measure and input specific parameters into the record analysis system: the length of the pedicle is L cm, the diameter of the pedicle is d cm, the distance from the intersection to the midline is a cm (positive above, negative below, and 0 on the midline). All three data can be measured preoperatively and intraoperatively. We set the diameter of the compass center circle to 10 cm, and the length of the center guide needle to 10 cm.

[0213] The radius of the compass puncture range is (d / L)*10 cm.

[0214] The distance from the intersection to the midline of the pedicle is a cm, and the distance of the intersection translation is (L / d)*a cm.

[0215] When a is positive, i.e. the intersection point is above the midline, the target point in the vertebral body is below the puncture line (i.e. puncture downward), the intersection point needs to be translated to the right, i.e. in the direction of the vertebral body; if it is above (i.e. puncture upward), the intersection point needs to be translated to the left, i.e. away from the vertebral body.

[0216] When a is negative, the intersection point is below the pedicle midline, and the translation is completely opposite, the target point in the vertebral body is below the puncture line (i.e. puncture downward), the intersection point needs to be translated to the left; if it is above (i.e. puncture upward), the intersection point needs to be translated to the right.

[0217] Finally, display the specific guide hole (number) of the puncture and the length value of the puncture guide needle on the display system.

[0218] 7. Maintain the compass position unchanged, and withdraw the central guide needle about 4mm backward to leave the intersection point.

[0219] 8. Insert a puncture needle along the selected guide hole to the bone surface of the pedicle surface, and confirm its safety and whether the extension line is on the target point by C-arm fluoroscopy. After confirming the accuracy.

[0220] 9. Open the operation control system, and the system automatically calculates the distance from the compass surface to the puncture target point, the puncture angle and direction, and virtually displays them on the display system. Open the execution key to automatically puncture and drill the guide needle, or manually limit the depth to drill the guide needle.

[0221] 10. Finally, confirm. Remove the first and / or second guide needle, move away the compass and stabilizing assembly, and complete the puncture operation.

[0222] The technical effects formed are as follows:

[0223] Through the systematic deep integration of the above-mentioned modules, the present scheme not only overcomes all the limitations of the background technology, but also forms a new system which is logically closed-loop and has qualitative changes in performance. The technical effects brought by the present scheme cannot be achieved by simple addition of single technical improvement, and have significant creativity and non-obviousness.

[0224] 1. Paradigm innovation from "static blueprint navigation" to "dynamic real-time guidance": Background technology is like taking a printed static building drawing to build a building, which cannot cope with the settlement of the foundation and the deformation of the material. This solution is completely different. It first builds a "real-time digital twin model" of the construction site through the perception modeling module; then the decision planning module adjusts the construction plan dynamically according to the real-time model like an experienced chief engineer; finally, the control execution module is like a precise construction team that accurately executes the plan and resists various disturbances on site. This realizes the fundamental paradigm shift from "following the drawing" to "on-site command", making surgical navigation have the ability to respond to real-world dynamics for the first time.

[0225] 2. Perfect fusion of "global foresight" and "local instantaneous response": Traditionally, algorithms that can make global optimal planning are usually time-consuming in calculation and difficult to apply in real time; while real-time control algorithms that can respond quickly often lack a global perspective and are easily trapped in the trap of local optimization. This solution skillfully unifies this contradiction through its systematic architecture. The optimal control and high-level reinforcement learning module ensures the "strategic foresight" of the decision, ensuring the global optimality and safety of the entire puncture path. The bottom layer of robust adaptive controller provides "tactical instantaneous" ability to resist physical disturbances in milliseconds. This combination of "thinking" and "instant response" makes the system both "strategically planned" and "tactically responsive", reaching an unprecedented level of intelligence.

[0226] 3. Essence transition from "fixed one-time tool" to "evolvable surgical expert system": The compass in the background technology has completely fixed performance when it leaves the factory. The core of this solution, the deep reinforcement learning framework, gives the system the ability to learn and evolve. The system can continuously iterate and optimize its decision strategy network through massive simulated surgeries and learning from past real surgery data. This means that the success experience and failure lessons of each surgery can be absorbed by the system and transformed into the improvement of its "skills". The system is no longer a cold tool, but a "surgical expert" that can grow together with top surgeons, and its performance will continue to approach and even surpass the theoretical limit as the application data increases.

[0227] The above only describes the embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A transpedicular, intelligent navigation system, characterized in that, include: The perception and modeling module is used to build a dynamic digital twin model of the surgical area in real time and predict the dynamic changes during the surgical process by integrating multi-source and heterogeneous intraoperative information. The decision-making and planning module is used to plan a globally optimal "intelligent" puncture strategy for the surgical puncture device that can adapt to environmental changes in real time, by comprehensively considering multiple complex objectives such as safety, accuracy, stability and long-term efficacy, based on the digital twin model constructed by the perception and modeling module. The control and execution module is used to ensure that the system achieves sub-millimeter-level trajectory tracking accuracy and highly robust precision operation in the complex, variable, and uncertain physical environment of a real surgical operation.

2. The transpedicular intelligent navigation system of claim 1, wherein, The intelligent navigation system for transpedicle puncture also employs a surgical strategy function π to minimize risk. π = argmin_{π∈policyspace}J(π); The input to the strategy function is all the sensing information of the system at any time t, and the output is the action instruction that the surgical puncture device should execute at that time.

3. The transpedicular smart navigation system of claim 1, wherein, The perception and modeling module includes: Multimodal data fusion and non-rigid 3D reconstruction units are used to precisely drive the deformation of high-precision preoperative 3D models during surgery using limited, low-dose real-time images to match the patient's actual dynamics. A hybrid dynamic modeling unit combining physical laws and data-driven approaches is used to predict trends during surgery.

4. The transpedicular smart navigation system of claim 1, wherein, The decision-making and planning module includes: The global path corridor planning unit based on optimal control theory plans the optimal surgical path by dynamically calculating the probabilistic distribution during the surgical process. A sequential decision-making unit based on hierarchical deep reinforcement learning is used to construct an intelligent decision-making system by combining optimal control theory with hierarchical deep reinforcement learning.

5. The transpedicular smart navigation system of claim 3, wherein, The workflow of the multimodal data fusion and non-rigid 3D reconstruction unit is as follows: Data-constrained projection projects the current 3D digital model onto the same viewing angle as the real-time C-arm image, based on a simulated X-ray imaging physics process, forming a simulated 2D image. Then, the difference between the simulated image and the real C-arm image is calculated, and this difference is used as a driving force to backpropagate and adjust the shape of the 3D model, ensuring that its projection best matches the actual observation. In the previous adjustment step, the 3D model may exhibit distortions or deformations that do not conform to anatomical principles. The system projects the deformed model into a constrained space that conforms to anatomical and biomechanical principles, and smooths and corrects unreasonable deformations based on built-in anatomical prior knowledge to ensure the model's realism.

6. The transpedicular smart navigation system of claim 3, wherein, The perception and modeling module dynamically calculates the spatial displacement that should occur during the surgery using an optimal deformation field function, thus dynamically evolving into a digital twin model consistent with the real situation. The optimal deformation field function is: Optimal deformation field (spatial coordinates, time) = argmin[data fidelity term + physical constraint term + spatial smoothing regularization term]; Data fidelity term = integral over all intra-operative images (|| projection operator (pre-operative 3D model + deformation field) - intra-operative 2D image || 2 ) d image; Physical constraint term = Physical constraint weight × Integral (Differential algebraic system energy (deformation field, rate of change of deformation field with respect to time)) d time; Spatial smoothing regularizer = regularization weight x integral (||gradient operator (deformation field) || 2 )d space.

7. The transpedicular smart navigation system of claim 1, wherein, The control and execution module comprises a disturbance suppression control unit based on dead zone self-adaption, which realizes fast, accurate and self-adaptive online suppression of various unknown disturbances without increasing the complexity of the system, and ensures high fidelity execution of the surgical strategy in the physical layer.

8. A percutaneous pedicle access navigation device, comprising: The navigation device comprises the intelligent navigation system for transpedicular puncture as claimed in any one of claims 1-7, and further comprises: A puncture device electrically connected with the intelligent navigation system, and the intelligent navigation system controls accurate operation of the puncture device.

9. The transpedicular navigation device of claim 8, wherein, The working steps are as follows: According to the image, the target position of the vertebral body lesion site is determined, and the number of target points is determined according to the specific puncture requirement of the clinic, and the target points are input and marked in the navigation device for transpedicular puncture; The intelligent navigation system for transpedicular puncture is used to calculate the maximum range of bilateral transpedicular puncture of the vertebral body, that is, to determine the maximum projection range of bilateral transpedicular in the vertebral body, and to determine whether the target point part of the vertebral body lesion in the previous step is covered by the maximum projection range, if not, the puncture method is abandoned, if covered, the next process is entered; Dynamic planning of needle guiding operation path; The puncture device performs puncture action.

10. The transpedicular navigation device of claim 8, wherein, The puncture device comprises: A needle for transpedicular puncture; A compass placed on the surface of the patient's back skin during surgery for guiding the puncture direction and position of the needle, the needle passes through the center of the compass and forms an initial positioning; A stabilizing component for fixing the compass at a predetermined position and supporting the movement of the compass.

11. The transpedicular navigation device of claim 10, wherein, The positioning steps of the compass are as follows: Firstly, the puncture range (the length of the transpedicular is L and the width is d) is determined: When the length / width = 2 (L / d = 2): at this time, the intersection point of puncture is just located at the midpoint of the transpedicular center line, and the puncture range on the compass is just the inner circle range, without scaling; When the length / width > 2 (L / d > 2): the transpedicular is relatively "long and thin", and the puncture range on the compass needs to be scaled down, and the calculation formula of the scaled radius is r = (d / L)*G, G represents the length of the needle; When the length / width < 2 (L / d < 2): the transpedicular is relatively "thick and short", and the puncture range on the compass needs to be scaled up, and the calculation formula of the scaled radius is also r = (d / L)*G, G represents the length of the needle; Secondly, the up and down offset is processed: If the intersection point of the first puncture is not in the center of the transpedicular center line, but has an upward or downward offset (distance a), then the target puncture point needs to be "translated" to compensate; The translation distance calculation formula is c = (L / d)*a, The translation direction rule is: if the initial intersection point is above the center line, if the target point needs to be punched below, it needs to be translated to the right; if the target point needs to be punched above, it needs to be translated to the left; If the initial intersection point is below the center line, the rule is completely opposite; Thirdly, if there is an angle deviation (the center needle forms an angle with the transpedicular center line, and the angle is A), the angle deviation needs to be corrected, and the compass needs to be translated to make the needle direction parallel to the transpedicular center line: The translation distance calculation formula is: T represents the distance that the compass needs to be translated, G represents the length of the guide pin, and A represents the angle between the center guide pin and the ideal pedicle center line; The inner diameter of the compass is equal to G.

Images