A non-rigid registration method and system based on bending mode of puncture needle

Abstract

This invention relates to a non-rigid registration method and system based on the bending morphology of a puncture needle. The method includes acquiring the three-dimensional bending morphology of the puncture needle within the tissue in real time using multimodal sensing technology to generate needle morphology data; performing initial rigid registration based on preoperative CT and intraoperative space based on the data; performing non-rigid fine registration on this basis to compensate for errors caused by tissue deformation and physiological movement; fusing multi-source sensor data for state estimation and error compensation to generate dynamic registration results; and using the results to drive a robot to achieve closed-loop control of the puncture path. The system includes multimodal sensors, a registration calculation unit, a multi-source data fusion unit, and a control unit, which collaboratively complete morphology reconstruction, registration calculation, and dynamic guidance. This invention can significantly improve the registration accuracy and stability of puncture surgery, reduce the risks caused by target displacement, and is applicable to various robot-assisted puncture scenarios.

Description

Technical Field

[0001] This invention relates to the field of medical robot navigation and registration technology, specifically to a non-rigid registration method and system based on the bending shape of a puncture needle, which is suitable for the alignment and dynamic guidance of preoperative images and intraoperative anatomical structures in robot-assisted puncture surgery. Background Technology

[0002] With the increasing application of medical robots in puncture, ablation, and targeted injection, accurate registration between preoperative images and intraoperative anatomical structures has become a key technology. Existing techniques generally employ rigid or weakly rigid registration methods between preoperative CT images and intraoperative spatial coordinate systems to guide the robot in completing the needle insertion path. However, human soft tissue has non-rigid characteristics, and factors such as forces, tissue deformation, and physiological movements during puncture often lead to registration errors, target deviations, or path deviations when relying solely on rigid registration.

[0003] CN107808698B discloses a method for modeling the mechanics and bending deformation of flexible needles puncturing soft tissue, including the following steps: establishing a clamping friction force model; establishing a cutting force model; establishing a tissue resistance force model; based on the established mechanical models, deriving and summing the lateral components of each force, and substituting the resultant lateral force into the large deformation theory to solve for the bending deformation of the flexible needle. This invention's method for modeling the mechanics and bending deformation of flexible needles considers the influence of factors such as supporting friction and tissue resistance friction, and selects a large deformation theory that better reflects the actual bending deformation of flexible needles to study their bending deformation.

[0004] CN110202022A discloses a method for modeling the mechanics and bending deformation of flexible needles puncturing soft tissues. This method predicts needle deflection and bending by establishing a mechanical model of the interaction between the needle and the tissue.

[0005] In summary, the existing technology still has the following shortcomings: (1) The offline model is the main model and lacks real-time adaptation during the operation, which cannot accurately reflect the dynamic changes in the state of tissue and needle during the operation; (2) Without the fusion of multimodal sensing data, the prediction relies solely on a single mechanical model, which cannot compensate for model errors and sensing errors; (3) The lack of a closed-loop control guidance mechanism makes it impossible to directly use the registration results for real-time path control of the robot; (4) The path compensation capability is limited, and the ability to actively compensate for target deviation and tissue movement is insufficient in complex surgical environments. Summary of the Invention

[0006] This invention aims to overcome the above-mentioned shortcomings and provide a non-rigid registration method and system based on the bending shape of the puncture needle. Through multimodal real-time sensing and non-rigid registration, it realizes dynamic and precise guidance of the puncture needle during surgery, significantly improving the accuracy and safety of surgical registration.

[0007] This invention provides a dynamic non-rigid registration method based on the bending morphology of a puncture needle, comprising the following steps: S1. The multimodal sensor uses multimodal sensing technology to acquire the three-dimensional bending shape of the puncture needle in the tissue in real time and generate needle body morphology data. S2. The registration calculation unit performs initial rigid registration between the preoperative CT coordinate system and the intraoperative surgical space coordinate system based on the needle morphology data to obtain the initial alignment result. S3. Based on the initial alignment results and the needle body morphology data, the registration calculation unit uses the finite element method to establish a puncture needle-tissue interaction model and a decoupled two-stage registration strategy, performs non-rigid fine registration, and obtains non-rigid registration results. S4. The multi-source data fusion unit, based on the non-rigid registration result, fuses multi-source sensor data to perform state estimation and error compensation, and generates dynamic registration result. S5. The control unit uses the dynamic registration result to drive the robot to perform closed-loop control.

[0008] Preferably, the multimodal sensing technology in step S1 includes at least one of ultrasonic imaging, FBG sensing (fiber Bragg grating sensing), and EM sensing (electromagnetic sensing).

[0009] Preferably, the non-rigid fine registration in step S3 includes the following steps: S31. The error modeling module calculates the residual error between the preoperative CT model and the intraoperative image based on the initial alignment results; and generates the initial estimation field of tissue deformation by combining the needle morphology data with the residual field estimation and deformation propagation model. S32. The tissue deformation modeling module uses the finite element method to establish an interaction model between the puncture needle and the tissue, and uses the initial estimated field of tissue deformation as the boundary condition input to solve for the complete tissue deformation field. S33. The registration module receives the tissue deformation field and the initial alignment result, and performs non-rigid optimization based on the cross-modal registration algorithm of image features or image intensity to obtain the non-rigid optimization result. S34, the decoupled registration module is based on a decoupled two-stage registration strategy, which gradually aligns real-time two-dimensional ultrasound slices to preoperative three-dimensional CT images and integrates them with non-rigid optimization results to output non-rigid registration results.

[0010] Preferably, the calculation model for the initial estimation field of tissue deformation is: ∆ u ( x )=f ( k ( x ) ， ∆ r ), where ∆ u ( x ) represents the initial estimated field of tissue deformation, which is the predicted displacement distribution calculated based on needle morphology data and rigid registration error, ∆ r This represents the pose error of rigid registration. k ( x ) represents the needle body morphology data; f ( k ( x ) ， ∆ r The function relation defined by the deformation propagation model is used to generate the initial estimation field of the microstructure deformation; this estimation field is input as the boundary condition into the finite element model, and the microstructure deformation field is obtained by solving. u ( x The values ​​at each point are deformation field displacement vectors, representing the actual displacement of the tissue under the action of the needle.

[0011] Preferably, the multi-source sensor data includes ultrasound image data, FBG sensor data, EM sensor data, and force or torque sensor data; the fusion process takes the non-rigid registration result as input, and uses an unscented Kalman filter to fuse and predict the state of the multi-source sensor data to generate a dynamic registration result.

[0012] Preferably, the closed-loop control in step S5 is performed based on the dynamic registration result and includes the following steps: S51. A path planning strategy based on deep reinforcement learning is adopted to generate predictive compensation paths. S52. Based on the predictive compensation path, drive the robot to perform a puncture operation to achieve active compensation for target displacement and tissue deformation.

[0013] The present invention also provides a dynamic non-rigid registration system based on the bending morphology of a puncture needle, comprising: Multimodal sensor: used to acquire the three-dimensional bending morphology of the puncture needle in the tissue in real time and send the needle morphology data to the registration calculation unit; Registration calculation unit: used to receive needle morphology data output by multimodal sensors, perform initial rigid registration between preoperative CT and intraoperative space; perform non-rigid fine registration based on needle morphology data, and output non-rigid registration results to multi-source data fusion unit; Multi-source data fusion unit: used to receive non-rigid registration results as well as ultrasound image data, FBG sensor data, EM sensor data and force or torque sensor data, perform state estimation and error compensation, and output dynamic registration results to the control unit; Control unit: Used to receive dynamic registration results and drive the robot arm to perform closed-loop control.

[0014] Preferably, the multimodal sensor includes: FBG sensor: installed inside the puncture needle, used to collect strain along the needle axis and output curvature distribution information; EM sensor: Located at the tip and base of the puncture needle, used to output the position and attitude information of the needle body in the global coordinate system; Ultrasound imaging probe: placed on the patient's surgical area or mounted on the robotic arm to acquire cross-sectional images of the puncture needle.

[0015] Preferably, the registration calculation unit includes: Error modeling module: used for needle morphology data k ( x ) and rigid registration ∆r Functional relationship ∆ u ( x ) =f ( k ( x ) ， ∆ r ), calculate the initial estimated field ∆ of tissue deformation. u ( x ), and the initial estimated field ∆ of tissue deformation. u ( x Send to the tissue deformation modeling module; Tissue Deformation Modeling Module: Used to establish an interaction model between the puncture needle and tissue based on the finite element method, and to estimate the initial tissue deformation field ∆. u ( x As a boundary condition input, the tissue deformation field is output to the image registration module; Image registration module: Used to receive the tissue deformation field and the initial alignment result, perform cross-modal non-rigid registration optimization based on image features or image intensity, generate non-rigid optimization results, and output them to the decoupled registration module; Decoupling Registration Module: Based on a decoupling two-stage registration strategy, it gradually aligns real-time two-dimensional ultrasound slices with preoperative three-dimensional CT images, fuses non-rigid optimization results, and outputs non-rigid registration results to the multi-source data fusion unit.

[0016] Preferably, the control unit includes: Path planning module: Used to receive dynamic registration results and generate predictive compensation paths using deep reinforcement learning methods; Predictive control module (42): used to perform active compensation of target displacement and tissue deformation based on predictive compensation path, and output control commands; Actuator interface (43): Used to receive control commands and drive the robot manipulator and end effector to perform puncture operations.

[0017] During the procedure, multimodal sensors first acquire real-time data on the bending morphology of the puncture needle within the tissue. Ultrasonic imaging obtains cross-sectional images of the needle, an FBG sensor measures needle strain, and an EM sensor acquires spatial pose information. Subsequently, the system performs initial rigid registration based on preoperative CT images and intraoperative spatial data, followed by non-rigid fine registration to dynamically compensate for errors caused by tissue deformation and physiological movement. The registration results are fused with multi-source sensor data using an unscented Kalman filter to obtain the real-time dynamic relationship between the needle and the target position. Finally, the control unit invokes a path planning model trained with deep reinforcement learning to generate a predictive compensation path and drives the robotic arm to perform the puncture operation in a closed loop, thereby achieving precise arrival and stable guidance of the target.

[0018] The key innovation of this invention lies in: 1. For the first time, ultrasound, FBG and EM sensing technologies are integrated to obtain the three-dimensional bending morphology of the puncture needle in the tissue in real time, which breaks through the limitations of existing technologies that rely on a single sensing method and have insufficient accuracy and robustness.

[0019] 2. A two-step registration mode with rigid initial alignment and non-rigid fine compensation is adopted, and combined with finite element modeling, image feature matching and decoupled 2D ultrasound-3D CT registration, to achieve dynamic alignment across modalities and dimensions, which significantly improves the registration accuracy under tissue deformation and physiological motion.

[0020] 3. Based on the registration results, unscented Kalman filtering is introduced for state estimation, and combined with deep reinforcement learning to achieve predictive path planning and active compensation. For the first time, non-rigid registration results are directly applied to robot closed-loop control to achieve dynamic guidance and precise control of the surgical process. Attached Figure Description

[0021] Figure 1 This is a functional block diagram of a non-rigid registration method and system based on the bending shape of a puncture needle. Figure 2 This is a schematic diagram illustrating a non-rigid registration method based on the bending shape of a puncture needle and the function of a multimodal sensor in the system. Figure 3 This is a schematic diagram of the registration calculation process for a non-rigid registration method based on the bending shape of a puncture needle. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0023] The following embodiments are only used to illustrate the technical principles of the present invention and a feasible implementation method, and do not constitute a limitation on specific algorithms, model parameters or implementation details. Those skilled in the art can make various equivalent modifications without departing from the concept of the present invention. Example 1

[0024] like Figure 1 As shown, a dynamic non-rigid registration method and system based on the bending morphology of a puncture needle mainly includes: The system includes a multimodal sensor 10, a registration calculation unit 20, a multi-source data fusion unit 30, and a control unit 40.

[0025] The modules are connected via data links, forming a closed-loop structure of "sensing-registration-fusion-control".

[0026] The entire system workflow is as follows: Multimodal sensors acquire the three-dimensional bending morphology of the needle body within the tissue in real time; The registration calculation unit performs rigid and non-rigid registration between preoperative CT and intraoperative space; The multi-source data fusion unit 30 performs time-series fusion and state estimation of the registration results and sensor information; The control unit 40 generates predictive control commands based on the fusion results to achieve closed-loop control of the puncture process.

[0027] like Figure 2 As shown, the multimodal sensor functionally includes: FBG sensor 11: Located inside the puncture needle, it is used to measure the strain distribution along the needle axis and calculate the local curvature; it outputs needle curvature information in the form of a function.

[0028] EM sensor 12: Installed at the needle tip and needle base, used to measure the attitude and displacement of the needle body in the global coordinate system; outputs pose quaternion and displacement vector.

[0029] Ultrasound probe 13: placed on the patient's surgical area or mounted on the robot arm to acquire cross-sectional images of the puncture needle and the echo characteristics of the surrounding tissue.

[0030] The strain / curvature distribution output by FBG, the pose data output by EM, and the ultrasound images are aligned and spatially interpolated using a registration calculation unit to form complete needle morphology data. k(x) This describes the spatial bending shape of the needle body along the axial direction x.

[0031] like Figure 3 As shown, the non-rigid registration process of the registration calculation unit includes the following functional steps: The registration calculation unit first performs rigid registration between the preoperative CT three-dimensional reconstruction model and the intraoperative spatial image.

[0032] The process involves obtaining the rigid transformation matrix through marker points or spatial coordinate matching. R,t Output the initial alignment result.

[0033] Non-rigid registration includes four sub-steps: The purpose of step S31 is to calculate the residual error between the preoperative CT model and the intraoperative images, and to estimate the initial estimation field of tissue deformation by combining the needle morphology data (such as curvature distribution, pose information, etc.).

[0034] Based on the initial alignment results, the rigid registration pose error ∆ is calculated by inputting the spatial coordinates of the marker points corresponding to the preoperative CT and intraoperative images through the error modeling module. r : ∆ r=M CT −T CT→US M US It reflects the accuracy error caused by the flexibility of the needle body, among which... T CT→US It is a rigid registration matrix. M CT and M US These are the coordinates of corresponding points in the CT and ultrasound images, respectively.

[0035] Based on needle morphology data k ( x ), combined with rigid registration pose error ∆ r The initial deformation estimation field is obtained through the deformation propagation model: ∆ u ( x ) =f ( k ( x ),∆ r ) in, k ( x The needle morphology data is typically provided by FBG and EM sensors and includes the needle's geometry and orientation information.

[0036] Deformation propagation model f ( ∙ The expected deformation of the needle in the tissue can be calculated using machine learning or physical model-based methods.

[0037] Needle morphology data obtained from multimodal sensors are used to estimate deformation and ensure that appropriate sensor calibration is used.

[0038] In terms of model selection, the deformation propagation model... f ( ∙ You can choose to train based on an elasticity model or using an existing regression model.

[0039] The function of step S32 is to establish an interaction model between the puncture needle and the tissue using the finite element method (FEM) to generate the tissue deformation field.

[0040] Initial estimation field ∆ using needle deformation u ( x Using ) as boundary conditions, a finite element model of the tissue is established, which is used to calculate the deformation of the tissue under the action of the puncture needle.

[0041] The tissue is divided into multiple small finite element elements, and the displacement field of each element is solved by the finite element method.

[0042] Based on the elastic modulus, Poisson's ratio, and the force applied by the external puncture needle, an equilibrium equation is constructed: (In one feasible implementation, tissue deformation can be described through mathematical modeling. The following formula is only used to illustrate a mathematical expression of the tissue deformation relationship and does not constitute a limitation on specific calculation methods or implementation steps.) ∇∙ ( C: ∇ u ) +F ext = 0 Where C is the stiffness matrix of the organization. F ext It is the force applied by the puncture needle. In practical applications, those skilled in the art can use other equivalent mathematical models or calculation methods to describe tissue deformation.

[0043] By solving the above equations, the displacement field of the tissue can be obtained. u ( x ), which is the amount of deformation at each point after the force is applied.

[0044] The process calculates the actual tissue deformation field, reflecting the specific displacement of the tissue under the action of the needle.

[0045] The elastic modulus and Poisson's ratio of an organization are selected based on the actual organizational characteristics and are usually obtained through mechanical testing in the laboratory.

[0046] The finite element solver can be either ANSYS or Abaqus, or a custom solver can be created according to actual needs.

[0047] The function of step S33 is to perform non-rigid optimization of the tissue deformation field and the initial alignment result using a cross-modal registration algorithm based on image features or image intensity.

[0048] According to the organizational deformation field u ( x Cross-modal registration with preoperative CT images can be performed using image feature methods such as SIFT and SURF to extract feature points from the images for matching; alternatively, mutual information (MI) methods based on image intensity can be used. ; in, P ( i,j ) represents the joint probability distribution of CT and ultrasound images.

[0049] Feature points or texture information are extracted from CT and ultrasound images, and spatial transformations between images are obtained through feature point matching.

[0050] The initial alignment result is optimized using mutual information from image intensity or feature point matching to minimize the objective function: ; in, I CT ( p i )and I US ( T ( p i )) represent the intensity values ​​of CT and ultrasound images, respectively. T ( p i () represents the transformed point.

[0051] The optimized transformation matrix and registration results can be used in subsequent decoupling registration steps.

[0052] Mutual information method is applicable to CT and ultrasound images with overlapping areas and can effectively avoid the effects of illumination changes under different modalities.

[0053] Step S34 functions based on a decoupled two-stage registration strategy, progressively aligning real-time two-dimensional ultrasound slices with preoperative three-dimensional CT images. This involves a stepwise optimization approach, first addressing intramodal registration and then performing cross-modal registration to reduce computational complexity and improve registration accuracy. First, register the real-time two-dimensional ultrasound slices and match their intramodal alignment with the three-dimensional ultrasound volume data: ; in, I US ( pi () represents the intensity of a two-dimensional ultrasound slice. I 3D ( T ( p i The intensity is the transformed 3D data.

[0054] Then, the real-time ultrasound slices were registered with the preoperative 3D CT images, and the spatial transformation was optimized by minimizing mutual information or feature point matching. ; in, I CT ( pi )and I US ( T ( pi )) represent the intensity values ​​of CT and ultrasound images, respectively. T ( pi () represents the transformed points. This result represents the spatial transformation between the two images and is the final result after optimization for registration errors.

[0055] The decoupled registration result is fused with the initial alignment result to output the final non-rigid registration result, ensuring accurate alignment between the two.

[0056] The key to the decoupling registration step is to progressively align the CT and ultrasound images, register them into a unified spatial frame, and generate a registration matrix. T decoupled This matrix is ​​used to describe the relationship between ultrasound images and CT images, ensuring that they can be aligned in the same coordinate system.

[0057] In this step, the two registration results (the non-rigid optimization result and the result of stepwise alignment of 2D ultrasound slices with preoperative 3D CT images) need to be combined. To obtain the final accurately aligned image, the two registration results need to be weighted and fused.

[0058] There are generally two options for fusion methods: Weighted average method: By assigning weights to each result, a weighted average is calculated.

[0059] T opt It is a non-rigid optimization registration matrix, while T decoupled The result is the stepwise alignment of two-dimensional ultrasound slices with preoperative three-dimensional CT images, then the fusion matrix... T fused Weighted calculation can be used: ; Among them, among them,w 1 and w 2 represents the weighting coefficient, used to characterize the relative influence of different registration results in the fusion process. In practical applications, the determination of weights and the fusion strategy can be adjusted according to specific needs, or other equivalent fusion methods can be used. The above formula is only used to illustrate a mathematical representation of multi-source registration result fusion, to help understand the combination relationship between different registration results, and does not constitute a limitation on the fusion algorithm or parameter setting method.

[0060] Cascaded Transformation Method: Non-rigid Optimization Transformation Matrix T opt and decoupling registration matrix T decoupled All of these are valid transformation matrices and can be directly concatenated: T fused = T opt ·T decoupled The above formula indicates that by first applying the decoupling registration matrix and then applying the non-rigid optimization matrix, the final non-rigid registration result is obtained.

[0061] The multi-source data fusion unit 30 receives non-rigid registration results from the registration calculation unit and ultrasonic, FBG, EM, and force / torque sensing data from the multimodal sensor 10.

[0062] The fusion process is based on the unscented Kalman filter (UKF) algorithm, which generates dynamic registration results by predicting, updating, and iteratively calculating the tissue state estimation vector.

[0063] The core process of the unscented Kalman filter is state transition and observation update: ; in, The estimated state at the current moment. z t For the observation data at the current moment, f ( ∙ )and h ( ∙ These are the state transition function and the observation function, respectively. K t The above formula is only used to illustrate the relationship between state variables and to help understand the state update mechanism. It does not constitute a limitation on specific filtering algorithms or calculation processes. In practical applications, those skilled in the art can use Kalman filtering methods or other equivalent state estimation methods to achieve dynamic updating and correction of the system state.

[0064] The UKF algorithm can suppress errors caused by sensor noise, data delay, etc., and generate accurate dynamic registration results for use in control unit decision-making.

[0065] This process enables the fusion and temporal filtering of multi-source sensor information, outputting dynamic registration results that reflect the temporal continuity and spatial consistency of tissue deformation and needle state in real time.

[0066] The control unit 40 includes a path planning module 41, a predictive control module 42, and an actuator interface 43.

[0067] The path planning module is based on the deep reinforcement learning algorithm PPO to generate predictive compensation paths. This path optimization algorithm continuously adjusts its strategy through a reinforcement learning framework, enabling the system to predict future paths based on the dynamic changes of the target organ and plan compensation paths in advance.

[0068] The predictive control module receives the dynamic registration results and adjusts the robot arm's control strategy based on the generated predictive compensation path. By controlling target displacement and tissue deformation, the robot can actively compensate for the movement of the target organ, ensuring accurate tracking of the puncture path.

[0069] The actuator interface receives predictive control commands and drives the robotic arm and end effector to perform the puncture operation. The entire process is carried out under closed-loop control, which not only monitors the needle position in real time, but also adjusts the puncture path based on real-time feedback to ensure precise alignment of the target position.

[0070] In robotic puncture applications, the robotic arm carries the puncture needle, aligning it with the pre-planned target point, and monitors the needle's position using a real-time ultrasound probe. Utilizing data from multimodal sensors within the system, the system calculates the relative displacement between the needle and the target point in real time, achieving precise puncture through dynamic registration and closed-loop control. In practical applications, the system effectively overcomes factors such as tissue movement and needle bending, ensuring the accuracy and safety of the puncture path.

[0071] The entire process can maintain spatial consistency between the preoperative CT model and the intraoperative images under dynamic conditions, thereby significantly improving puncture accuracy and safety.

[0072] In practical applications, the method described in this invention can maintain high registration stability and path guidance accuracy even in the presence of tissue deformation and physiological movement, thereby improving the safety and reliability of the puncture process.

[0073] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method of dynamic non-rigid registration based on the bending behavior of a puncture needle, characterized by, Includes the following steps: S1, Multimodal sensor (10) acquires the three-dimensional bending shape of the puncture needle in the tissue in real time through multimodal sensing technology and generates needle body shape data; S2, the registration calculation unit (20) performs initial rigid registration between the preoperative CT coordinate system and the intraoperative surgical space coordinate system based on the needle body morphology data to obtain the initial alignment result; S3. The registration calculation unit (20) uses the finite element method to establish a puncture needle-tissue interaction model and a decoupled two-stage registration strategy based on the initial alignment result and the needle body morphology data, performs non-rigid fine registration, and obtains non-rigid registration results. S4. The multi-source data fusion unit (30) performs state estimation and error compensation by fusing multi-source sensor data based on the non-rigid registration result, and generates dynamic registration result. S5. The control unit (40) uses the dynamic registration result to drive the robot to perform closed-loop control.

2. The dynamic non-rigid registration method based on the bending shape of the puncture needle according to claim 1, characterized in that, The needle morphology data is obtained by at least two of the following technologies: ultrasound imaging, FBG sensing, and EM sensing.

3. The dynamic non-rigid registration method based on the bending shape of the puncture needle according to claim 1, wherein, The non-rigid fine registration in step S3 includes the following steps: S31, Error modeling module (21) calculates the residual error between the preoperative CT model and the intraoperative image based on the initial alignment result; combined with the needle morphology data, it generates the initial estimation field of tissue deformation using residual field estimation and deformation propagation model. S32, Tissue Deformation Modeling Module (22) uses the finite element method to establish the interaction model between the puncture needle and the tissue, and takes the initial estimated field of tissue deformation as the boundary condition input to solve the complete tissue deformation field. S33, the registration module (23) receives the tissue deformation field and the initial alignment result, performs non-rigid optimization based on the cross-modal registration algorithm of image features or image intensity, and obtains the non-rigid optimization result; S34, Decoupling Registration Module (24) Based on the decoupling two-stage registration strategy, it gradually aligns the real-time two-dimensional ultrasound slices to the preoperative three-dimensional CT image and merges them with the non-rigid optimization results to output the non-rigid registration results.

4. The dynamic non-rigid registration method based on the bending shape of the puncture needle according to claim 3, characterized in that, The calculation model of the initial estimation field of the tissue deformation is: u ( x ) =f ( k ( x ) ， ∆ r ), wherein ∆ u ( x ) represents the initial estimation field of the tissue deformation, ∆ r represents the rigid registration pose error, k ( x ) represents the needle body shape data.

5. The dynamic non-rigid registration method based on the bending shape of the puncture needle according to claim 1, wherein, The multi-source sensor data includes ultrasound image data, FBG sensor data, EM sensor data, and force or torque sensor data. The fusion process takes the non-rigid registration result as input and uses a Kalman filter-type state estimation method to fuse and predict the state of the multi-source sensor data, generating a dynamic registration result.

6. The dynamic non-rigid registration method based on the bending shape of the puncture needle according to claim 1, wherein, The closed-loop control in step S5 is executed based on the dynamic registration result and includes the following steps: S51. A path planning strategy based on deep reinforcement learning is adopted to generate predictive compensation paths. S52. Based on the predictive compensation path, drive the robot to perform a puncture operation to achieve active compensation for target displacement and tissue deformation.

7. A dynamic non-rigid registration system based on the bending modality of a puncture needle, characterized in that, include: Multimodal sensor (10): used to acquire the three-dimensional bending shape of the puncture needle in the tissue in real time, and send the needle shape data to the registration calculation unit (20). Registration calculation unit (20): used to receive the needle body morphology data output by the multimodal sensor (10), perform initial rigid registration between preoperative CT and intraoperative space; perform non-rigid fine registration based on the needle body morphology data, and output the non-rigid registration result to the multi-source data fusion unit (30). Multi-source data fusion unit (30): used to receive the non-rigid registration result as well as ultrasound image data, FBG sensor data, EM sensor data and force or torque sensor data, perform state estimation and error compensation, and output dynamic registration result to control unit (40). Control unit (40): Used to receive the dynamic registration result and drive the robot arm to perform closed-loop control.

8. The non-rigid registration system based on the bending configuration of a puncture needle of claim 7, wherein, The multimodal sensor (10) includes: FBG sensor (11): installed inside the puncture needle, used to collect strain along the needle axis and output curvature distribution information; EM sensor (12): set at the tip and base of the puncture needle, used to output the position and attitude information of the needle body in the global coordinate system; Ultrasonic imaging probe (13): placed on the patient's surgical area or mounted on the robot arm to acquire cross-sectional images of the puncture needle.

9. The dynamic non-rigid registration system based on the bending mode of a puncture needle of claim 7, wherein, The registration calculation unit (20) includes: an error modeling module (21) configured to compute an initial estimate of the tissue deformation field based on the needle morphology data k ( x ) and a function of the rigid registration r u ( x ) =f ( k ( x ), and r ),​ ∆ u x ), and sending the initial estimate field of tissue deformation u x ) to the tissue deformation modeling module (22);​​ a tissue deformation modeling module (22) for establishing a puncture needle and tissue interaction model based on a finite element method, inputting the initial estimate of the tissue deformation field Δ u ( x ) as a boundary condition, outputting the tissue deformation field to the image registration module (23); Image registration module (23): used to receive the tissue deformation field and the initial alignment result, perform cross-modal non-rigid registration optimization based on image features or image intensity, generate the non-rigid optimization result, and output it to the decoupled registration module (24). Decoupling registration module (24): Based on the decoupling two-stage registration strategy, it gradually aligns the real-time two-dimensional ultrasound slices with the preoperative three-dimensional CT images, and fuses the non-rigid optimization results to output the non-rigid registration results to the multi-source data fusion unit (30).

10. The non-rigid registration system based on the bending modality of a puncture needle of claim 7, wherein, The control unit (40) includes: Path planning module (41): used to receive the dynamic registration result and generate a predictive compensation path using a deep reinforcement learning method; Predictive control module (42): used to perform active compensation of target displacement and tissue deformation based on the predictive compensation path, and output control commands; Actuator interface (43): used to receive the control command and drive the robot manipulator and end effector to perform puncture operation.

Images