A control method of a smart surgical robot

Abstract

The present application relates to the field of robot control, and particularly relates to a control method of an intelligent surgical robot. The content comprises: acquiring preoperative multi-modal medical image data of a patient, and preprocessing to obtain preprocessed medical image data; based on the preprocessed medical image data, generating a three-dimensional anatomical model through a multi-modal medical image registration and segmentation algorithm; based on the three-dimensional anatomical model, generating a real-time optimized surgical path through a multi-scale dynamic disturbance optimization reinforcement learning algorithm; and based on the real-time optimized surgical path, controlling the robot. The technical problems of inaccurate medical image data processing and analysis and poor flexibility of surgical path planning in the control of the intelligent surgical robot by the traditional robot control method are solved.

Description

Technical Field

[0001] This invention relates to the field of robot control, and more particularly to a control method for an intelligent surgical robot. Background Technology

[0002] With the continuous advancement of medical technology, especially the rapid development of robot-assisted surgery, surgical robots are playing an increasingly important role in various types of complex surgeries. Surgical robots can reduce human error, minimize surgical trauma, and improve surgical success rates and patient recovery speed through highly precise operation. Especially in delicate surgeries requiring extremely high precision, such as neurosurgery, ophthalmology, and spinal surgery, surgical robots have demonstrated irreplaceable advantages. However, surgical robots face significant challenges in real-time, precise navigation and path optimization during implementation.

[0003] Currently, surgical robot systems primarily rely on preoperative imaging data to generate 3D models, and then use these models for path planning and surgical navigation. Traditional navigation methods often depend on a single imaging data source, such as CT or MRI. However, due to the limitations of different imaging technologies, the anatomical information provided by a single image source is often insufficient to fully cover the surgical area, leading to inadequate accuracy in path planning. Furthermore, intraoperative dynamic changes and individual patient differences also significantly impact the accuracy of robot navigation.

[0004] The aforementioned technologies, when used in the control of intelligent surgical robots, suffer from technical problems such as inaccurate processing and analysis of medical image data and poor flexibility in surgical path planning. Summary of the Invention

[0005] This invention provides a control method for an intelligent surgical robot to solve the technical problems of inaccurate medical image data processing and analysis and poor flexibility in surgical path planning in traditional robot control methods for intelligent surgical robots.

[0006] The present invention provides a control method for an intelligent surgical robot, specifically comprising the following technical solutions:

[0007] A control method for an intelligent surgical robot includes the following steps:

[0008] S1. Acquire the patient's preoperative multimodal medical image data and preprocess it to obtain preprocessed medical image data; based on the preprocessed medical image data, through multimodal medical image registration and segmentation algorithm, introduce an adaptive transformation model based on morphological gradient consistency to optimize the registration accuracy of multimodal medical images, establish a morphologically consistent mapping relationship, and generate a three-dimensional anatomical model.

[0009] S2. Based on a three-dimensional anatomical model, a real-time optimized surgical path is generated through a multi-scale dynamic perturbation optimization reinforcement learning algorithm; the robot is then controlled based on the real-time optimized surgical path.

[0010] Preferably, S1 specifically includes:

[0011] In the implementation of the multimodal medical image registration and segmentation algorithm, the preprocessed medical image data is initially feature-mapped, and feature analysis is performed jointly in the spatial and frequency domains to obtain the feature data of the preprocessed medical image data.

[0012] Preferably, S1 specifically includes:

[0013] Based on the feature data of the preprocessed medical image data, an adaptive transformation model based on morphological gradient consistency is introduced to establish a morphologically consistent mapping relationship. During the registration process, a regularized loss function is constructed to optimize the registration accuracy of multimodal medical images by minimizing the gradient deviation between modalities, thus obtaining the registered medical image data.

[0014] Preferably, S1 specifically includes:

[0015] Based on the registered medical image data, a weighted feature fusion strategy is introduced; the fusion weight is calculated using the signal-to-noise ratio and entropy value of the registered medical image data; based on the fusion weight, multimodal fusion is performed to obtain fused image data; the fused image data is subjected to high-order morphological processing to obtain morphologically processed fused image data; based on the morphologically processed fused image data, three-dimensional reconstruction is performed using voxel interpolation and morphological repair methods to obtain a three-dimensional anatomical model.

[0016] Preferably, S2 specifically includes:

[0017] In the implementation of the multi-scale dynamic perturbation optimization reinforcement learning algorithm, a multi-scale state description model is constructed to obtain a state vector; the state vector includes the robot's current posture matrix, force feedback matrix, velocity vector, target distance, and path parameters.

[0018] Preferably, S2 specifically includes:

[0019] In the implementation of the multi-scale dynamic perturbation optimization reinforcement learning algorithm, a multi-scale dynamic perturbation optimization mechanism is introduced. Combining global and local perturbation terms, a perturbation function is constructed to obtain the perturbation output. The specific formula for the perturbation function is as follows:

[0020] ,

[0021] in, At a certain point in time The output result of the disturbance; It is the surgical robot at a certain point in time. The state vector; It represents the number of disturbances; It is the amplitude coefficient, representing the first... The intensity of the disturbance; It is the first The perturbation frequency factor of the perturbation; It is the first The phase offset of each disturbance; It is the first Local perturbation weights for each perturbation; It is the first The rate of change of a disturbance; It is a global perturbation term; It is a local disturbance term.

[0022] Preferably, S2 specifically includes:

[0023] The output of the perturbation is used as the input for adjusting the path decision strategy. Based on the state vector and the update step size of the current path decision strategy, a dynamic strategy adjustment process is constructed to update the strategy and determine the new path selection.

[0024] Preferably, S2 specifically includes:

[0025] In the implementation of the multi-scale dynamic perturbation optimization reinforcement learning algorithm, a reward function is constructed based on the attitude matrix, force feedback matrix, velocity vector, target distance, and path parameters in the state vector; the specific formula of the reward function is as follows:

[0026] ,

[0027] in, At a certain point in time The reward function; This is the reward coefficient; This is the robot's current pose matrix; It is the target location; This is the robot's current force feedback matrix; It is the safety force threshold; This is the robot's current speed; It is the spatial distance between the robot and the target; It is the first One path parameter; It is the partial derivative of the target distance with respect to the path parameters; It is the number of path parameters; It is the attenuation factor; It is the time limit; the first item The second term measures the Euclidean distance between the current position and the target position. The third item measures the safety of force feedback during surgery. The fourth term uses a logical decay function to suppress excessive velocity changes. The smoothness of the path is measured by the second derivative of the target distance with respect to the path parameters, specifically the fifth term. This is the integral of time decay.

[0028] Preferably, S2 specifically includes:

[0029] By combining the perturbation function and the reward function, the surgical path is iteratively updated to obtain a real-time optimized surgical path. The real-time optimized surgical path is then converted into motion commands for the robot, and the intelligent surgical robot is controlled by an error-based control algorithm.

[0030] The beneficial effects of the technical solution of the present invention are:

[0031] 1. In the process of fusing preprocessed medical image data using a multimodal medical image registration and segmentation algorithm, this invention introduces an adaptive transformation model based on morphological gradient consistency to optimize the accuracy of multimodal medical image registration. By minimizing gradient deviation, it solves the registration error between multimodal medical images. Based on the signal-to-noise ratio and entropy value of the registered medical image data, the fusion weight is calculated to obtain fused image data. The contrast and boundary sharpness of the fused image data are improved through high-order morphological processing to obtain morphologically processed fused image data. The morphologically processed fused image data is then subjected to three-dimensional reconstruction. Using Bezier interpolation and morphological repair methods, a detailed three-dimensional anatomical model is generated to accurately reflect the morphological characteristics of organs and tissues.

[0032] 2. Based on the three-dimensional anatomical model, combined with the multi-scale dynamic perturbation optimization reinforcement learning algorithm, a perturbation function is introduced. By combining the perturbation patterns of the local and global paths, the path is fine-tuned to ensure the accuracy and safety of the surgical process. A real-time optimized surgical path is generated and precisely controlled by the robot to ensure stable operation of the robot in the dynamically changing surgical environment. Attached Figure Description

[0033] Figure 1 This is a flowchart of a control method for an intelligent surgical robot according to the present invention. Detailed Implementation

[0034] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] The following describes in detail, with reference to the accompanying drawings, a specific scheme for the control method of an intelligent surgical robot provided by the present invention.

[0037] See attached document Figure 1 The diagram illustrates a control method for an intelligent surgical robot according to an embodiment of the present invention, the method comprising the following steps:

[0038] S1. Acquire the patient's preoperative multimodal medical image data and preprocess it to obtain preprocessed medical image data; based on the preprocessed medical image data, through multimodal medical image registration and segmentation algorithm, introduce an adaptive transformation model based on morphological gradient consistency to optimize the registration accuracy of multimodal medical images, establish a morphologically consistent mapping relationship, and generate a three-dimensional anatomical model.

[0039] The preoperative multimodal medical image data (such as CT, MRI, ultrasound, etc.) of the patient is obtained through the hospital PACS (Image Archiving and Communication System) interface, and the preoperative multimodal medical image data of the patient is preprocessed by means such as format conversion, image coordinate system normalization, noise reduction, image enhancement, and artifact removal to obtain preprocessed medical image data. The preprocessing process adopts technical means well known to those skilled in the art, which will not be described in detail here.

[0040] Furthermore, the preprocessed medical image data is fused using a multimodal medical image registration and segmentation algorithm to generate a three-dimensional anatomical model;

[0041] The specific implementation process of the multimodal medical image registration and segmentation algorithm is as follows:

[0042] First, feature extraction is performed on the preprocessed medical image data. A method combining local and global features is used to comprehensively analyze the edge, texture, and density information of the preprocessed medical image data. Specifically, the preprocessed medical image data... Preliminary feature mapping is performed, and feature analysis is conducted jointly in the spatial and frequency domains to obtain the feature data of the preprocessed medical image data. The specific formula is as follows:

[0043] ,

[0044] in, Indicates the first The preprocessed medical image data is located at Feature data at the location; These are image grid coordinates, representing the spatial location of the image, used to determine the local area for feature calculation; It refers to the number of feature channels, using multiple channels to extract rich feature information; It is the first The local feature weight allocation of each feature channel represents the degree of contribution of different features. It is used to weight and combine features in the local space to adjust the influence of features such as edges and textures. It is obtained through experimental methods. , , They represent the positions respectively. place, along The degree of gray-level change in a direction, i.e., the gradient, is used to reflect edge information and enhance the boundary features of the image. The gradient is calculated using the Roberts operator. It is the first The preprocessed medical image data is located at grayscale at the location; It is the first The frequency domain feature weights of each feature channel are used to control the degree of influence of frequency domain features on the overall features, so as to adjust the balance between noise and texture information in the preprocessed medical image data. These are adjustment parameters set according to the image spectral characteristics. The Fourier transform result is used to capture frequency domain information and extract frequency domain features from preprocessed medical image data to describe the periodicity and global structure of the preprocessed medical image data, thus supplementing the deficiencies of spatial domain features. The purpose of feature extraction is to provide a consistent representation for subsequent fusion, enabling data from different modalities to be registered in the same feature space.

[0045] Based on the feature data of preprocessed medical image data, an adaptive transformation model based on morphological gradient consistency is introduced to optimize the registration accuracy of multimodal medical images. The goal is to establish a morphologically consistent mapping relationship by minimizing the gradient deviation between modalities. The regularization loss function in the registration process is defined as follows:

[0046] ,

[0047] in, It is the regularization loss function in the registration process, representing the difference between the registered medical image data and the reference image data. It also includes a smoothing term (gradient term) to ensure the smoothness and continuity of the registration transformation. It is the domain of the entire image, representing the set of positions of all voxels in the preprocessed medical image data; It is the first after registration transformation Medical imaging data in voxels (locations) Feature data at the location, , Reference image data In voxels Feature data at the location, which serves as the target for registration, is used to guide the preprocessed medical image data in location registration. Feature data at the location Perform a transformation to align it with the reference image data; It is a minimization operation, which means finding the solution that minimizes the regularization loss function among all possible registration transformations, that is, finding the optimal medical image data after registration transformation that minimizes the error between medical image feature data. It is determined based on reference image data and expert experience. This is a regularization parameter used to suppress artifacts caused by excessive deformation; It is the first after registration transformation Medical imaging data in voxels (locations) The gradient at the point is determined. The regularization loss function is solved using the existing steepest descent method to obtain the registered medical image data. ;

[0048] Furthermore, after registration, the process proceeds to the multimodal fusion stage, employing a weighted feature fusion strategy. Fusion weights are calculated using the signal-to-noise ratio (SNR) and entropy. Based on these weights, multimodal fusion is performed to obtain the fused image data, as shown in the following formula:

[0049] ,

[0050] ,

[0051] in, In position Fusion image data at the location; It is the first The registered medical image data is located at Feature data at the location; This is the total number of registered medical image data; It is the first The weights of the registered medical image data, i.e., the fusion weights; It is the first Signal-to-noise ratio of the registered medical image data; It is the first Signal-to-noise ratio of the registered medical image data; It is an adjustment factor used to balance the influence of signal-to-noise ratio and entropy on the weights, and is determined based on expert experience. It is the first The entropy value of the registered medical image data; It is the first The entropy value of the registered medical image data. The purpose of multimodal fusion is to integrate the advantages of multimodal images and improve the contrast and detail preservation of anatomical structures. Based on the fused image data, high-order morphological processing is used to remove noise and improve the boundary sharpness of the fused image data, resulting in morphologically processed fused image data;

[0052] Based on the fused image data after image processing, a 3D reconstruction is performed using existing voxel interpolation and morphological restoration methods to obtain a 3D anatomical model. The specific implementation process is as follows: First, interpolation is performed on the continuous voxel space based on interpolation weights, which are obtained based on the Bessel interpolation formula. The purpose of interpolation is to generate a more refined voxel representation to improve the spatial accuracy of the 3D anatomical model. Then, morphological restoration is performed to eliminate artifacts and defects that may occur during the 3D reconstruction process, resulting in the voxel density function of the restored 3D anatomical model. Furthermore, isosurfaces are generated using the MarchingCubes algorithm to form a visualized 3D anatomical model. :

[0053] ,

[0054] in, The tissue density threshold was set based on expert experience; the final result was a three-dimensional anatomical model that reflected the morphological characteristics of organs and tissues.

[0055] S2. Based on a three-dimensional anatomical model, a real-time optimized surgical path is generated through a multi-scale dynamic perturbation optimization reinforcement learning algorithm; the robot is then controlled based on the real-time optimized surgical path.

[0056] Based on a three-dimensional anatomical model, a multi-scale dynamic perturbation optimization reinforcement learning algorithm is introduced to generate a real-time optimized surgical path, and the robot is controlled based on the real-time optimized surgical path. The specific implementation process of the multi-scale dynamic perturbation optimization reinforcement learning algorithm is as follows:

[0057] First, the surgical environment is modeled. The surgical environment consists of multiple complex physiological structures and constraints. To achieve optimal path planning while ensuring surgical safety, a multi-scale state description model needs to be constructed to model the surgical environment. This multi-scale state description model comprehensively considers factors such as the robot's posture, force feedback, velocity, and target distance in space to construct a state vector. The model then represents the state vector of the surgical robot at any given time point. The state is represented as a state vector Its components include the robot's current pose matrix. Force feedback matrix velocity vector Target distance and path parameters The mathematical expression for the state vector is:

[0058] ,

[0059] During path planning, the uncertainties of the surgical environment can affect the robot's motion stability. Therefore, a multi-scale dynamic perturbation optimization mechanism is introduced, combining multiple perturbation modes to construct a perturbation function. This allows for fine-tuning of the path at different scales, ensuring the robot can make joint local and global adjustments when responding to dynamic changes in physiological tissues. The specific formula for the perturbation function is as follows:

[0060] ,

[0061] in, At a certain point in time The output of the perturbation describes the dynamic perturbation of the surgical robot during the path planning process, and is used to represent the effects of minor adjustments during the operation and changes in the surgical environment. It represents the number of disturbances; It is the amplitude coefficient, representing the first... The intensity of each perturbation, used to control the magnitude of the perturbation amplitude, is obtained through experimental fitting and its value range is [value range missing]. ; It is the first The perturbation frequency factor of each perturbation determines the oscillation frequency of the perturbation and affects the sensitivity of path planning. It is obtained using expert experience and its value range is [value missing]. ; It is the first The phase offset of each perturbation is used to adjust the phase offset of the sine function to ensure the alignment between the perturbation and the state of the intelligent surgical robot. This value is obtained based on expert experience and ranges from [value missing]. ; It is the first The local perturbation weights are used to control the contribution of local perturbation terms, thereby adjusting the accuracy of local path refinement. These weights are obtained experimentally and their values ​​range from [value range missing]. ; It is the first The rate of change of each perturbation is used to control the response rate of the perturbation function to changes in the state of the intelligent surgical robot, affecting the adjustment frequency of the local path. It is determined based on expert experience and the response speed requirements under different surgical scenarios, and its value range is [value missing]. ; It is a global perturbation term used to provide path perturbation on a global scale in order to break through local optimum traps; It is a local perturbation term (logistic regression function) used to adjust the refined path of the intelligent surgical robot in the micro-surgical area to ensure that tissue damage is avoided;

[0062] Furthermore, the output of the perturbation is used as input for subsequent path decision strategy adjustments, providing dynamic perturbation input for path optimization. During path search, the path decision strategy needs continuous adjustment to balance exploration and exploitation. Therefore, a dynamic strategy adjustment process is constructed based on the state vector and the update step size of the current path decision strategy. Then, the strategy is updated to determine the new path selection. The expression for the strategy update is:

[0063] ,

[0064] in, This indicates that the intelligent surgical robot is in a certain state. Select action The probability; specifically It is a policy function that gives the policy value in the current state. Next, select different actions. The probability of; Indicates the time point of the intelligent surgical robot The selected action is a control signal or a robot operation instruction, such as how the robot should adjust its position, posture, speed, or perform a certain operation step; Indicates at a point in time The policy update step size is used to control the speed at which the path decision policy is adjusted; a larger step size is preferable. A value of [value] leads to a larger policy update, resulting in a rapid change in the path decision strategy; conversely, a smaller value leads to a smaller update, determined through experimentation; the left side of the arrow [is affected by this]. This indicates the state of the intelligent surgical robot after the strategy is corrected. Select action The probability; the one to the right of the arrow This indicates the state of the intelligent surgical robot under the current path decision-making strategy. Select action The probability of; Indicates based on the current state and strategy parameters The calculated policy correction function, used to adjust the current path decision strategy to make the robot more efficient in performing tasks, is derived through machine learning algorithms (such as deep reinforcement learning). , It represents the total number of dimensions for strategy correction; It is in the Each dimension is related to the force feedback matrix The weight parameters related to the square root are determined based on expert experience. Indicates the first The policy parameters in each dimension can be weights in a neural network or adjustable parameters in other machine learning models, used to control how the path decision policy operates in the current state. Select actions and optimize robot behavior by adjusting these parameters;

[0065] Furthermore, by combining multiple factors such as target distance, force feedback, speed, and path smoothness, a reward function adapted to complex physiological environments is constructed to ensure the safety and optimality of the path, guaranteeing that the intelligent surgical robot achieves path optimization in the shortest possible time. The specific formula for the reward function is as follows:

[0066] ,

[0067] in, It is the reward function, which represents the quality of the robot's current path selection. It is the objective function used to optimize path selection, so as to ensure that the robot can achieve a balance in terms of safety, efficiency and path smoothness during operation. The reward coefficient is used to balance the contributions of different evaluation items to the reward function. The magnitude of different coefficients represents the degree of influence of different objectives on the final path selection, and is determined based on expert experience. The first item... Measure the current position relative to the target position Euclidean distance, second term The third item measures the safety of force feedback during surgery. The fourth term uses a logical decay function to suppress excessive velocity changes. The smoothness of the path is measured by the second derivative of the target distance with respect to the path parameters, specifically the fifth term. This is the time decay integral, used to ensure that the path planning process gradually stabilizes as the operation time increases; It is the robot's current pose matrix, representing the robot's position and orientation (rotation and displacement) in three-dimensional space. The target location refers to the predetermined target location or area that the surgical robot needs to reach, which is set using expert experience. This is the robot's current force feedback matrix, representing the interaction forces between the robot and the surgical target; It is the safety force threshold, which represents the safe force value when the robot comes into contact with the tissue during surgery. It is set according to the safety standards when the robot was designed. This is the robot's current speed, representing the magnitude of the robot's speed during movement; It is a logical decay function of speed, used to control the sensitivity of the reward function to speed through a smooth decay function; It is the position or distance on the target path, representing the spatial distance between the robot and the target; It is a path parameter, representing a specific parameter in the path; It is the partial derivative of the target distance with respect to the path parameters, representing the sensitivity of the target distance to changes in the path parameters; It represents the number of path parameters, indicating the total number of parameters that may be involved in the path; It is the decay factor, used to control the rate of the time decay function, and determines how quickly the reward decays over time; It is the time limit, representing the maximum time range considered during the path planning process; It is the integral of the time decay function, which means that as time goes by, the impact of time on path optimization during path planning gradually weakens.

[0068] Finally, by combining the perturbation function and reward feedback, the surgical path is iteratively updated, and its mathematical expression is as follows:

[0069] ,

[0070] in, These are the new path parameters, i.e., the real-time optimized surgical path. They represent the path parameters after one optimization step and refer to the robot's path parameters, which may include the robot's position, posture, speed, acceleration, etc. This represents the current path parameters, that is, the path that the robot is executing at the current moment; It is the path update step size, which represents the update magnitude of path parameters during the optimization process. It is used to control the speed and stability of path updates and is set according to expert experience. It is the gradient of the reward function, representing the reward function. Relative to the state vector The gradient reflects the direction and magnitude of path optimization, and determines how the robot should adjust its path during the optimization process to increase the reward value during execution. It is the local derivative of the reward function with respect to the state vector, representing the reward function. At the present moment For the state vector The rate of change; It is a perturbation function; It represents the cumulative count, indicating the number of times the calculation needs to be performed during path optimization, which is equivalent to the reward function. Total number of items;

[0071] Through the above process, the intelligent surgical robot can dynamically adjust in a multi-scale environment during path planning. By combining multi-factor perturbation and reward feedback, it can achieve precise planning of the surgical path. During path exploration, it can make full use of global information to conduct a large-scale search and combine real-time feedback to make local corrections, forming an adaptive path optimization scheme to cope with complex surgical environments.

[0072] Finally, the real-time optimized surgical path is converted into motion commands for the robot, and an error-based control algorithm, such as a proportional-integral-derivative (PID) controller or a more complex robust control algorithm, is used to control the intelligent surgical robot.

[0073] In summary, a control method for an intelligent surgical robot has been developed.

[0074] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0075] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0076] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A control method for an intelligent surgical robot, characterized in that, Includes the following steps: S1. Acquire the patient's preoperative multimodal medical image data and preprocess it to obtain preprocessed medical image data; based on the preprocessed medical image data, generate a three-dimensional anatomical model and construct a surgical navigation model through multimodal medical image registration and segmentation algorithms; S2. Based on the surgical navigation model, a multi-scale dynamic perturbation optimization reinforcement learning algorithm is introduced. This algorithm combines global and local perturbation terms to construct a perturbation function, obtaining the perturbation output and generating a real-time optimized surgical path. The specific formula for the perturbation function is as follows: ， in, At a certain point in time The output result of the disturbance; It is the surgical robot at a certain point in time. The state vector; It represents the number of disturbances; It is the amplitude coefficient, representing the first... The intensity of the disturbance; It is the first The perturbation frequency factor of the perturbation; It is the first The phase offset of each disturbance; It is the first Local perturbation weights for each perturbation; It is the first The rate of change of a disturbance; It is a global perturbation term; It is a local disturbance term; The robot is controlled based on the real-time optimized surgical path.

2. The control method for an intelligent surgical robot according to claim 1, characterized in that, S1 specifically includes: In the implementation of the multimodal medical image registration and segmentation algorithm, the preprocessed medical image data is initially feature-mapped, and feature analysis is performed jointly in the spatial and frequency domains to obtain the feature data of the preprocessed medical image data.

3. The control method for an intelligent surgical robot according to claim 2, characterized in that, S1 specifically includes: Based on the feature data of preprocessed medical image data, an adaptive transformation model based on morphological gradient consistency is introduced to establish a morphologically consistent mapping relationship. A regularized loss function is constructed during the registration process to optimize the registration accuracy of multimodal medical images by minimizing the gradient deviation between modalities, resulting in registered medical image data. The regularized loss function is defined as follows: ， in, It is the regularization loss function in the registration process; This represents the set of locations of all voxels in the preprocessed medical image data. It is the first after registration transformation Medical imaging data in voxels Feature data at the location; It is reference image data; For regularization parameters; It is the first after registration transformation Medical imaging data in voxels The gradient at that point.

4. The control method for an intelligent surgical robot according to claim 3, characterized in that, S1 specifically includes: Based on the registered medical image data, a weighted feature fusion strategy is introduced; the fusion weight is calculated using the signal-to-noise ratio and entropy value of the registered medical image data; based on the fusion weight, multimodal fusion is performed to obtain fused image data; the fused image data is then subjected to high-order morphological processing to obtain morphologically processed fused image data; based on the morphologically processed fused image data, three-dimensional reconstruction is performed using voxel interpolation and morphological repair methods to obtain a three-dimensional anatomical model; based on the three-dimensional anatomical model, a surgical navigation model is constructed using image segmentation technology.

5. The control method for an intelligent surgical robot according to claim 1, characterized in that, S2 specifically includes: In the implementation of the multi-scale dynamic perturbation optimization reinforcement learning algorithm, a multi-scale state description model is constructed to obtain a state vector; the state vector includes the robot's current posture matrix, force feedback matrix, velocity vector, target distance, and path parameters.

6. The control method for an intelligent surgical robot according to claim 5, characterized in that, S2 specifically includes: The output of the perturbation is used as the input for adjusting the path decision strategy. Based on the state vector and the update step size of the current path decision strategy, a dynamic strategy adjustment process is constructed to update the strategy and determine the new path selection.

7. The control method for an intelligent surgical robot according to claim 6, characterized in that, S2 specifically includes: In the implementation of the multi-scale dynamic perturbation optimization reinforcement learning algorithm, a reward function is constructed based on the attitude matrix, force feedback matrix, velocity vector, target distance, and path parameters in the state vector; the specific formula of the reward function is as follows: ， in, At a certain point in time The reward function; This is the reward coefficient; This is the robot's current pose matrix; It is the target location; This is the robot's current force feedback matrix; It is the safety force threshold; This is the robot's current speed; It is the spatial distance between the robot and the target; It is the first One path parameter; It is the partial derivative of the target distance with respect to the path parameters; It is the number of path parameters; It is the attenuation factor; It is the time limit; the first term measures the Euclidean distance between the current position and the target position, the second term measures the safety of force feedback during the operation, the third term suppresses excessive speed changes through a logical decay function, the fourth term measures the path smoothness, and the fifth term is the time decay integral.

8. The control method for an intelligent surgical robot according to claim 7, characterized in that, S2 specifically includes: By combining the perturbation function and the reward function, the surgical path is iteratively updated to obtain a real-time optimized surgical path. The real-time optimized surgical path is then converted into motion commands for the robot, and the intelligent surgical robot is controlled by an error-based control algorithm.

Images