A force control method for a subcutaneous soft tissue surgical robot

Abstract

This invention discloses a force control method for a subcutaneous soft tissue surgical robot, relating to the field of medical robot technology. The method includes: acquiring a soft tissue ultrasound echo dataset and performing sound velocity calibration analysis to generate actual sound velocity values; using a tissue viscoelasticity compensation factor to perform nonlinear correction on the actual sound velocity values, outputting real-time corrected sound velocity values; calculating the actual fat thickness value based on soft tissue ultrasound images and the real-time corrected sound velocity values; and performing a difference operation between the actual fat thickness value and a zero-pressure reference value to form subcutaneous fat layering deformation variables. This invention improves the controllability and safety of complex soft tissue manipulations during surgery by performing sound velocity calibration analysis and nonlinear correction on the soft tissue ultrasound echo dataset, and simultaneously constructing a depth attenuation compensation model to accurately model the deep muscle correction deformation variables.

Description

Technical Field

[0001] This invention relates to the field of medical robot technology, and in particular to a force control method for a subcutaneous soft tissue surgical robot. Background Technology

[0002] With the rapid development of robotics technology, its applications have expanded from the industrial sector to the medical and health field. Particularly in surgery, the use of robots offers new possibilities for improving surgical precision, reducing trauma, and shortening recovery time. Subcutaneous soft tissue surgery, as an important branch of surgery, involves complex biomechanical properties and highly personalized anatomical structures, thus placing higher demands on the real-time performance, precision, and adaptability of robot control.

[0003] However, existing robotic technology still faces many challenges when applied to subcutaneous soft tissue surgery. First, due to the viscoelastic and non-uniform characteristics of soft tissue, traditional robotic force control methods struggle to accurately predict tissue deformation, leading to potential over- or under-cutting during surgery and affecting surgical outcomes. Second, most current image-guided robotic force control methods fail to adequately consider the differences in physical properties between internal soft tissue structures (such as the fat and muscle layers), thus limiting force control accuracy. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a force control method for a subcutaneous soft tissue surgical robot to solve the problems of inaccurate tissue deformation prediction and insufficient force control precision in existing robot force control methods.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a force control method for a subcutaneous soft tissue surgical robot, comprising: acquiring a soft tissue ultrasound echo dataset, performing sound velocity calibration analysis, generating an actual sound velocity value, using a tissue viscoelasticity compensation factor to perform nonlinear correction on the actual sound velocity value, and outputting a real-time corrected sound velocity value.

[0008] The thickness is calculated based on soft tissue ultrasound images and real-time corrected sound velocity values ​​to obtain the actual fat thickness value. The actual fat thickness value and the zero-pressure reference value are then compared to form the subcutaneous fat layering deformation.

[0009] The subcutaneous fat layer deformation is input into the depth attenuation compensation model, the deformation transfer layer performs attenuation weight allocation, the biomechanical layer calculates the contact stress distribution, and the deep muscle correction deformation is output.

[0010] Tissue state mapping is performed on the deep muscle correction deformation to generate vertical force correction coefficients. These coefficients are then input into the surgical robot control center for spatial force control conversion and joint torque calculation, outputting the surgical robot joint control torque.

[0011] As a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, the soft tissue ultrasound echo dataset includes soft tissue ultrasound images, bone tissue transit time, and tissue deformation displacement data.

[0012] In a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, the real-time correction sound velocity output specifically includes the following steps.

[0013] The bone tissue interface was located and the sound velocity was calibrated and analyzed in the soft tissue ultrasound echo dataset to generate the actual sound velocity value.

[0014] The actual sound speed value is mapped to biomechanical properties to obtain viscoelastic compensation parameters.

[0015] The viscoelastic compensation factor is applied to perform piecewise interpolation and nonlinear correction on the viscoelastic compensation parameters, and the real-time corrected sound velocity value is output.

[0016] In a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, obtaining the actual fat thickness value specifically includes the following steps.

[0017] Average grayscale segmentation was performed on soft tissue ultrasound images to obtain the coordinates of the fat layer boundary.

[0018] Based on the real-time corrected sound velocity value and the coordinates of the fat layer boundary, physical size conversion and thickness calculation are performed to obtain the actual fat thickness value.

[0019] As a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, the formation of subcutaneous fat layer deformation refers to performing a differential calculation between the actual fat thickness value and the zero pressure reference value to obtain a differential thickness value; and performing confidence weighting on the differential thickness value to form a subcutaneous fat layer deformation.

[0020] In a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, the step of outputting deep muscle correction deformation specifically includes the following steps.

[0021] A deformation transfer layer and a biomechanical layer are built by using a gated recurrent network and a residual attention architecture, and a deep attenuation compensation model is constructed by applying skip connections for weighted stacking.

[0022] Subcutaneous fat layer deformation is input into the depth attenuation compensation model. The deformation transfer layer uses depth-separable convolution to perform attenuation weight allocation and generate deformation transfer coefficients.

[0023] The biomechanical layer uses the viscoelastic constitutive equation to calculate the contact stress distribution and form stress response characteristics;

[0024] A bidirectional gating unit is used to fuse the deformation transfer coefficient and stress response characteristics in a time sequence to generate a deformation-stress coupling vector.

[0025] Multi-scale feature extraction and cross-layer residual aggregation are performed on the deformation-stress coupling vector to obtain the deep muscle correction deformation.

[0026] In a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, the generation of the vertical force correction coefficient specifically includes the following steps.

[0027] The displacement gradient algorithm is used to model the strain distribution of deep muscle correction deformation and generate muscle strain feature vectors.

[0028] The muscle strain feature vector is quantified by the vertical relative movement angle to form a slope-stress correlation matrix; the slope-stress correlation matrix is ​​then mapped to tissue state to generate a vertical force correction coefficient.

[0029] In a preferred embodiment of the force control method for the subcutaneous soft tissue surgical robot of the present invention, the output of the joint control torque of the surgical robot specifically includes the following steps.

[0030] The vertical force correction coefficient is input into the control center of the surgical robot to perform spatial force control conversion and generate an end-effector vertical force command.

[0031] The homogeneous transformation method is applied to project the end vertical force command into the base coordinate system to generate a three-dimensional force vector in the base coordinate system.

[0032] Perform joint torque calculation on the three-dimensional force vector of the base coordinate system and output the joint control torque of the surgical robot.

[0033] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the force control method for the subcutaneous soft tissue surgical robot as described in the first aspect of the present invention.

[0034] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the force control method for the subcutaneous soft tissue surgical robot as described in the first aspect of the present invention.

[0035] The beneficial effects of this invention are as follows: By performing sound velocity calibration analysis and nonlinear correction of the viscoelastic compensation factor on the soft tissue ultrasound echo dataset, the actual state parameters of soft tissue can be obtained more accurately, thereby improving the perception ability of subcutaneous fat layer deformation and solving the inaccuracy of traditional robot force control methods in tissue deformation prediction. Furthermore, a depth attenuation compensation model is constructed, and through the collaborative working mechanism of the deformation transfer layer and the biomechanical layer, accurate modeling of deep muscle correction deformation is achieved, improving the controllability and safety of complex soft tissue operations during surgery. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart of the force control method for a subcutaneous soft tissue surgical robot.

[0038] Figure 2 This is a flowchart of the sound speed calibration analysis process.

[0039] Figure 3 A flowchart for generating the deformation of subcutaneous fat layering.

[0040] Figure 4 A flowchart illustrating the workings of the deep attenuation compensation model. Detailed Implementation

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0043] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0044] Reference Figures 1-4As one embodiment of the present invention, this embodiment provides a force control method for a subcutaneous soft tissue surgical robot, comprising the following steps:

[0045] S1. Acquire soft tissue ultrasound echo dataset, perform sound velocity calibration analysis, generate actual sound velocity values, use tissue viscoelasticity compensation factor to perform nonlinear correction on actual sound velocity values, and output real-time corrected sound velocity values.

[0046] The specific steps are as follows.

[0047] S1.1 Acquire soft tissue ultrasound echo dataset, which includes soft tissue ultrasound images, bone tissue transit time, and tissue deformation and displacement data.

[0048] Soft tissue ultrasound images are acquired using a high-frequency ultrasound probe.

[0049] Bone tissue transit time refers to the time required for ultrasound waves to travel from the transmitting end through bone tissue to the receiving end, which is collected by an ultrasound pulse echo detector.

[0050] Tissue deformation displacement data refers to the deformation displacement information of soft tissue under the action of external force, which is collected using image analysis software (such as ImageJ);

[0051] Soft tissue ultrasound echo datasets can not only improve the real-time analysis of soft tissue structure and mechanical properties during surgery, but also enhance the accuracy and reliability of personalized medical solutions in the biomedical engineering industry.

[0052] S1.2. Preprocessing of the soft tissue ultrasound echo dataset: Specifically, for soft tissue ultrasound images, Gaussian filtering is applied for noise reduction to improve image clarity, and histogram equalization is used for contrast enhancement to optimize the layer resolution of tissue ultrasound images. For bone tissue transit time, linear interpolation is used for outlier identification and removal to ensure the reliability of bone tissue transit time data, while NTP time synchronization protocol is used for timestamp synchronization to ensure consistent acquisition time. For tissue deformation displacement data, spatial transformation is performed to ensure geometric consistency of tissue deformation displacement data, and morphological processing is used for noise point suppression to improve the smoothness and accuracy of tissue deformation displacement data. The preprocessed soft tissue ultrasound echo dataset is then output.

[0053] S1.3. Perform bone tissue interface localization and sound velocity calibration analysis on the preprocessed soft tissue ultrasound echo dataset to generate actual sound velocity values. Specifically, edge enhancement and multi-scale feature fusion are performed on the preprocessed soft tissue ultrasound echo dataset to generate soft tissue multimodal features; contour tracking and parameter extraction are performed on the soft tissue multimodal features to obtain soft tissue contour data; based on the soft tissue contour data, the Hough transform method is used to perform bone tissue interface localization; further, the soft tissue contour data is transformed into coordinates according to affine transformation coefficients to output bone tissue interface coordinates; spatial alignment and temporal calibration are performed on the bone tissue interface coordinates to obtain a calibration coordinate sequence; feature recombination and mean filtering are performed on the calibration coordinate sequence to output a bone tissue calibration feature set.

[0054] It should be noted that the affine transformation coefficients are obtained by least-squares fitting of the soft tissue contour data.

[0055] Next, the sound velocity calibration analysis is performed on the bone tissue calibration feature set. Furthermore, the propagation distance is mapped onto the bone tissue calibration feature set to generate initial sound velocity data. Outliers in the initial sound velocity data are removed to obtain an effective sound velocity set. Then, a weighted average and local smoothing are performed on the effective sound velocity set to generate an optimized sound velocity distribution. The optimized sound velocity distribution is then normalized to output the actual sound velocity value.

[0056] S1.4. The actual sound velocity value is nonlinearly corrected using a tissue viscoelasticity compensation factor, and the real-time corrected sound velocity value is output. In specific operation, the actual sound velocity value is mapped to biomechanical properties. Furthermore, tissue stiffness matching is performed on the actual sound velocity value to obtain the elastic modulus distribution. Simultaneously, frequency shift features are extracted from the actual sound velocity value to generate damping correlation parameters. The elastic modulus distribution and damping correlation parameters are fused to obtain a sound velocity-mechanical response relationship set. The sound velocity-mechanical response relationship set is decomposed into frequency bands to obtain a viscoelastic characteristic matrix. The viscoelastic characteristic matrix is ​​parameterized and sampled to output viscoelastic compensation parameters.

[0057] It should be noted that tissue stiffness matching refers to the process of retrieving the elastic modulus of the actual sound velocity value using a tissue stiffness grading table (based on the definition of a clinical tissue stiffness database).

[0058] Cubic spline interpolation is performed on the viscoelastic compensation parameters. Furthermore, uniform sampling and parameter alignment are performed on the viscoelastic compensation parameters to generate interpolated compensation parameters. Weight allocation and parameter fusion are performed on the interpolated compensation parameters to obtain optimized compensation parameters. Piecewise interpolation is performed on the optimized compensation parameters to output piecewise compensation parameters.

[0059] Next, nonlinear correction is performed on the segmented compensation parameters. Furthermore, tissue characteristic mapping is performed on the segmented compensation parameters to obtain characteristic matching parameters. The tissue viscoelasticity compensation factor is used to perform phase synchronization and amplitude calibration on the characteristic matching parameters to generate calibration compensation parameters. Dynamic adjustment and range limitation are performed on the calibration compensation parameters to obtain real-time corrected sound velocity values.

[0060] It should be noted that the tissue viscoelasticity compensation factor is defined based on the tissue type (fat / muscle) of the characteristic matching parameter, with an exemplary value range of [0.8 (muscle) ~ 1.2 (fat)].

[0061] S2. Based on the soft tissue ultrasound image and the real-time corrected sound velocity value, the thickness is converted to obtain the actual fat thickness value. The actual fat thickness value and the zero pressure reference value are then differentially calculated to form the subcutaneous fat layer deformation.

[0062] The specific steps are as follows.

[0063] S2.1. Perform average grayscale segmentation on soft tissue ultrasound images to obtain the coordinates of the fat layer boundary. Specifically, Gaussian filtering is used to perform noise suppression and edge preservation on the soft tissue ultrasound images to eliminate speckle noise and retain soft tissue boundary information, outputting a smooth ultrasound image. Subsequently, the Otsu thresholding algorithm is used to perform average grayscale segmentation on the smooth ultrasound images. Further, hole filling and edge connection are performed on the smooth ultrasound images to generate a connected soft tissue image. Grayscale equalization and binary segmentation are performed on the connected soft tissue image to form preliminary grayscale segmentation regions. According to the segmentation threshold, the preliminary grayscale segmentation regions are filtered. Further, the average grayscale value of the preliminary grayscale segmentation regions is extracted. When the average grayscale value exceeds the segmentation threshold, it is classified as a muscle tissue region; when the average grayscale value is lower than the segmentation threshold, it is classified as a fat tissue region. The segmented muscle tissue regions and fat tissue regions are integrated at the boundary to obtain the final tissue segmentation result.

[0064] It should be noted that the segmentation threshold is defined based on the bimodal distribution of the gray-level histogram of the initial gray-level segmentation region, with an exemplary value range of [40, 160].

[0065] The Moore-Neighbor tracing method is used to locate the boundaries of the final tissue segmentation results. Further, a neighborhood search is performed on the final tissue segmentation results to generate a preliminary boundary contour. Boundary tracing is performed on the preliminary boundary contour to obtain an initial boundary coordinate set. The initial boundary coordinate set is then simplified into a polygon approximation to form an optimized boundary contour. The optimized boundary contour is then subjected to curvature optimization and geometric parameter extraction to output the fat layer boundary coordinates.

[0066] S2.2. Based on the real-time corrected sound velocity value and the fat layer boundary coordinates, perform physical size conversion and thickness calculation to obtain the actual fat thickness value. In specific operations, sound velocity-geometric coupling is performed on the real-time corrected sound velocity value and the fat layer boundary coordinates. Furthermore, sound velocity gradient adjustment and coordinate space registration are performed on the real-time corrected sound velocity value and the fat layer boundary coordinates to form propagation characteristic features. The propagation characteristic features are then mapped to a spatial path to obtain the sound wave propagation path. The path length is integrated on the sound wave propagation path to form initial thickness data. The initial thickness data is then subjected to median filtering and moving weighted averaging to obtain a smooth thickness sequence. The smooth thickness sequence is then optimized in terms of parameters and its distribution is adjusted to output fat thickness distribution data.

[0067] The fat thickness distribution data is converted to physical dimensions. Further, the coordinates of the fat thickness distribution data are normalized to generate a normalized thickness matrix. An affine transformation is performed on the normalized thickness matrix to output quantized physical thickness values. Cubic spline interpolation is then performed on the quantized physical thickness values ​​to obtain a continuous thickness curve. Finally, the trapezoidal integral formula is used to calculate the thickness of the continuous thickness curve, outputting the actual fat thickness value. The specific mathematical format is as follows.

[0068] ;

[0069] in, This indicates the actual fat thickness value. This represents the total number of data points in a continuous thickness curve. This indicates the index of the data point in a continuous thickness curve. Indicates at data points The weight, Indicates at data points The physical thickness quantification value.

[0070] S2.3. The actual fat thickness value and the zero-pressure baseline value are differentially calculated to obtain the differential thickness value. This differential thickness value is then weighted with confidence levels to form the subcutaneous fat layering deformation. Specifically, baseline alignment and differential calculations are performed on the actual fat thickness value and the zero-pressure baseline value to obtain the original thickness change. A sliding window is then used to perform a weighted moving average and numerical normalization on the original thickness change, outputting the differential thickness value. The specific mathematical formula is as follows.

[0071] ;

[0072] in, Indicates the differential thickness value. This indicates the total time step within the sliding window; Indicates the index of a time point within the sliding window; Indicates the first Weighting coefficients for each time point; Indicates the first The actual fat thickness value at the i-th time point, l represents the i-th time point. Zero pressure baseline values ​​at each time point;

[0073] It should be noted that the zero-pressure baseline value refers to the fat thickness value of soft tissue under zero pressure, which is collected by an ultrasound probe combined with a torque sensor instrument; the weighting coefficient is defined based on the signal-to-noise ratio (SNR) of the original thickness change, and the exemplary value range is 0.8 to 1.2; the differential thickness value characterizes the degree of soft tissue deformation. When the differential thickness value is negative, it indicates that the soft tissue is in a state of compression deformation, and when the differential thickness value is positive, it indicates that the soft tissue is in a state of elastic rebound.

[0074] The differential thickness values ​​are weighted by confidence using a Bayesian weighting method. Further, region filtering and dynamic normalization are performed on the differential thickness values ​​to obtain stable thickness variations. Based on the confidence weighting coefficients, weighted aggregation and probability density fusion are performed on the stable thickness variations to generate probability-enhanced deformation variables. Stress distribution mapping is performed on the probability-enhanced deformation variables to obtain the fat layer deformation field. Deformation gradient quantization is performed on the fat layer deformation field to output subcutaneous fat layer deformation variables. These fat layer deformation variables reflect the dynamic deformation characteristics of the fat layer under pressure and can guide the real-time force control adjustment of the surgical robot.

[0075] It should be noted that the confidence weighting coefficient is based on the stability index (SSI) of stable thickness variation, with an exemplary range of 0.7 to 1.3.

[0076] S3. Input the subcutaneous fat layer deformation variables into the depth attenuation compensation model, perform attenuation weight allocation in the deformation transfer layer, calculate the contact stress distribution in the biomechanical layer, and output the deep muscle correction deformation variables.

[0077] The specific steps are as follows.

[0078] S3.1. Construct and train a depth decay compensation model. Specifically, in the TensorFlow framework, a gated recurrent network (RNN) is invoked using the `nn.GRU` parameter, and a depthwise separable convolution is embedded within it. The number of hidden units in the RRNN is set to 64, the input sequence length to 50, and the activation function to tanh. After the RRNN, a batch normalization layer is used for feature standardization to suppress gradient explosion, and a Dropout layer is used for regularization to complete the deformation propagation layer. The MHA (Multi-Head Attention) function is used to invoke the residual attention architecture, with the number of heads set to 4, the key dimension to 32, and the query dimension to 32. A viscoelastic constitutive equation is embedded within the residual attention architecture to quantify the stress-strain relationship, completing the biomechanical layer.

[0079] Skip connections are used to concatenate features and concatenate channels in the deformation transfer layer and biomechanical layer to obtain a fused feature matrix. The fused feature matrix is ​​then mapped using a fully connected layer to generate attenuation compensation coefficients. Dropout regularization is used to randomly deactivate the attenuation compensation parameters to obtain robust compensation parameters. The sigmoid function is used to nonlinearly activate the robust compensation parameters to generate the final fused weights. Based on the final fused weights, the deformation transfer layer and biomechanical layer are weighted and stacked to complete the construction of the deep attenuation compensation model.

[0080] Next, the depth decay compensation model is trained. Further, the historical subcutaneous fat stratification deformation variables are divided into a sample set, a training set, and a validation set. On the sample set, a data augmentation tool is used for random shifting and noise injection, followed by Z-score normalization using a standardizer to form a preprocessed sample set. On the training set, gradient descent is used to update the parameters of the preprocessed sample set, with an early stopping mechanism applied to monitor overfitting, outputting updated depth decay compensation model parameters. On the validation set, the mean squared error loss function is applied to quantify the loss value of the updated depth decay compensation model parameters, obtaining the validation error. When the validation error exceeds the convergence threshold for several consecutive rounds (e.g., 5 times), training terminates, and the trained depth decay compensation model is output simultaneously.

[0081] It should be noted that the convergence threshold is defined based on the relative rate of change of the verification error.

[0082] S3.2 The deformation transfer layer uses depthwise separable convolutions to perform attenuation weight allocation and generate deformation transfer coefficients. In specific operations, the subcutaneous fat layer deformation variables are input into the depth attenuation compensation model through the Input interface. The deformation transfer layer uses 1×1 convolutions to perform local feature extraction and temporal alignment on the subcutaneous fat layer deformation variables to form initial deformation features. A two-layer depthwise separable convolutional layer is used to perform attenuation weight allocation on the initial deformation features. Further, the first layer uses 3×1 depthwise convolution kernels to perform sliding filtering and channel separation on the initial deformation features along the time axis to generate channel-independent temporal patterns. The channel-independent temporal patterns are gated and weighted to capture the deformation transfer attenuation rate of different muscle layers (e.g., the deformation transfer attenuation rate of superficial muscles is 0.8, and the deformation transfer attenuation rate of deep muscles is 0.3), and the weighted temporal features are output. The second layer uses pointwise convolutions to perform linear combination on the weighted temporal features to generate fused attenuation features, and batch normalizes the fused attenuation features to output standardized attenuation weights.

[0083] Next, time-dependent modeling is performed on the standardized decay weights through a gated recurrent network. Furthermore, hidden state updates and nonlinear activations are performed on the standardized decay weights to obtain time-enhanced weights. The time-enhanced weights and the fused decay features are multiplied element-wise to obtain the dynamic decay coefficient matrix. Finally, the dynamic decay coefficient matrix is ​​residually superimposed through step connections to ensure effective gradient backpropagation and form a stable deformation transfer feature. Linear projection is performed on the stable deformation transfer feature to output the deformation transfer coefficient.

[0084] S3.3. The physical and mechanical layer uses the viscoelastic constitutive equation to calculate the contact stress distribution and form stress response characteristics. In specific operations, the residual attention architecture is used to analyze the energy distribution of the deformation transfer coefficient. Furthermore, the deformation transfer coefficient is weighted by multi-head attention to obtain temporal deformation characteristics. Channel fusion and residual connection are performed on the temporal deformation characteristics to generate composite deformation characteristics. Feature standardization and linear projection are performed on the composite deformation characteristics to obtain the deformation energy distribution vector.

[0085] The instantaneous contact stress is solved using the viscoelastic constitutive equation. Furthermore, stress relaxation iteration is performed on the deformation energy distribution vector to generate the instantaneous stress field. Spatial integration and energy normalization are then performed on the instantaneous stress field to obtain the contact stress distribution values. The specific mathematical formulas are as follows.

[0086] ;

[0087] in, Indicates the contact stress distribution value. Represents the instantaneous elastic modulus. Represents the deformation energy distribution vector. Indicates stress relaxation time;

[0088] It should be noted that the instantaneous elastic modulus is defined based on the step strain response curve of the DMA (Dynamic Mechanical Analysis) experiment; the stress relaxation time is defined based on the exponential fitting curve of the stress relaxation experiment.

[0089] In the spatial dimension, the contact stress distribution values ​​are reconstructed using radial basis functions. Further, node interpolation is performed on the contact stress distribution values ​​to obtain high-density stress sampling points, and local gradient alignment is performed on these high-density stress sampling points to obtain the initial stress distribution field. Then, edge enhancement is applied to the initial stress distribution field to generate a spatially continuous stress distribution. In the temporal dimension, temporal filtering is performed on the spatially continuous stress distribution to obtain a dynamic stress sequence. Gaussian smoothing is applied to the dynamic stress sequence to eliminate numerical oscillations. Simultaneously, dynamic response parameters are extracted and feature fusion is performed to output stress response features.

[0090] S3.4. A bidirectional gating unit is applied to perform temporal feature fusion of the deformation transfer coefficient and stress response features to generate a deformation-stress coupling vector. Specifically, the deformation transfer coefficient and stress response features are adjusted to the same dimension through zero padding to obtain an aligned feature vector. The aligned feature vector is then input into the bidirectional gating unit for temporal feature fusion. Further, the update gate of the forward gating unit performs weighted accumulation on the aligned feature vector to generate a forward hidden state. The reset gate adjusts the weights and performs nonlinear activation on the forward hidden state to output the forward temporal features. The update gate of the backward gating unit performs reverse weighted accumulation on the aligned feature vector to form a backward hidden state. The reset gate dynamically scales the features of the backward hidden state to obtain backward temporal features. The forward and backward temporal features are concatenated, and gradient clipping is applied to limit the gradient magnitude to prevent gradient explosion, outputting the bidirectional fused features. The bidirectional fused features are then dimensionality-reduced and projected through a fully connected layer to obtain the deformation-stress coupling vector.

[0091] S3.5. Perform multi-scale feature extraction and cross-layer residual aggregation on the deformation-stress coupling vector to obtain deep muscle correction deformation. Specifically, three parallel dilated convolution branches (dilation rates of 1, 3, and 5) are used to extract multi-scale features from the deformation-stress coupling vector. Furthermore, the branch with a dilation rate of 1 performs local feature extraction on the deformation-stress coupling vector to obtain high-resolution features, and performs gradient magnitude constraints on the high-resolution features to capture the micro-strain features of the local muscle. The branch with a dilation rate of 3 performs regional feature integration on the deformation-stress coupling vector to obtain mesoscale features, and performs spatial continuity optimization on the mesoscale features to capture the mesoscale motion features of the muscle bundles. The branch with a dilation rate of 5 performs global feature extraction on the deformation-stress coupling vector to obtain low-resolution features, and performs energy density normalization on the low-resolution features to extract the macroscopic deformation features of the entire muscle layer.

[0092] The micro-strain characteristics of local muscles, mesoscale motion characteristics of muscle bundles, and macroscopic deformation characteristics of the entire muscle layer are aggregated across layers using residuals. Furthermore, scale unification and feature smoothing are performed through bilinear interpolation to output multi-scale fusion features. The multi-scale fusion features are then connected across layers using residuals and weighted aggregation to form optimized fusion features. Orthogonal projection transformation is performed on the optimized fusion features to obtain the deep muscle correction deformation.

[0093] S4. Perform tissue state mapping on the deep muscle correction deformation to generate vertical force correction coefficients. Input the vertical force correction coefficients into the surgical robot control center to perform spatial force control conversion and joint torque calculation, and output the surgical robot joint control torque.

[0094] The specific steps are as follows.

[0095] S4.1. The displacement gradient algorithm is used to model the strain distribution of deep muscle correction deformation variables and generate muscle strain feature vectors. Specifically, spatial gradient mapping and tensor reconstruction are performed on the deep muscle correction deformation variables to generate strain component matrices. The strain component matrices are then symmetricized and normalized to obtain the strain distribution gradient field. Subsequently, energy density quantization and local averaging filtering are performed on the strain distribution gradient field to obtain the strain gradient amplitude. Tensor structure transformation is performed on the strain gradient amplitude to obtain the strain gradient tensor. Eigenvalue decomposition is then performed on the strain gradient tensor to extract the first L eigenvalues ​​as principal strain components. At the same time, orthogonal projection transformation is performed on the principal strain components to obtain shear strain components.

[0096] The direction parameters of the principal strain components and shear strain components are extracted and fused to obtain the strain direction vector. At the same time, an arctangent transformation is performed on the strain direction vector to generate the principal strain direction angle. Then, the principal strain components, shear strain components, principal strain direction angle and strain gradient tensor are Z-score normalized to eliminate dimensional differences and integrated to generate muscle strain feature vector.

[0097] S4.2 Quantize the muscle strain feature vector with vertical relative movement angle and map the tissue state to generate a vertical force correction coefficient. In the specific operation, during the angle quantization stage, perform directional gradient decomposition on the muscle strain feature vector to form axial and lateral gradient components. Perform geometric transformation on the axial and lateral gradient components to output the vertical relative movement angle. Use the arctangent function method to perform slope-stress coupling analysis on the vertical relative movement angle. Further, perform stress relationship mapping on the vertical relative movement angle to obtain stress-angle coupling parameters. Perform feature space transformation on the stress-angle coupling parameters to generate slope-stress features. Then, perform matrix filling on the slope-stress features to output the slope-stress correlation matrix.

[0098] In the tissue state mapping stage, a kernel function mapping method is used to perform nonlinear feature transformation on the slope-stress correlation matrix. Furthermore, attention weighting is applied to the slope-stress correlation matrix to enhance biomechanical features, resulting in a weighted slope-stress coupling vector. Feature aggregation and dimensionality compression are performed on the weighted slope-stress coupling vector to generate a tissue state feature vector. Linear projection is performed on the tissue state feature vector to obtain preliminary correction parameters. The preliminary correction parameters are then numerically normalized to output the vertical force correction coefficient.

[0099] S4.3 Input the vertical force correction coefficient into the surgical robot control center for spatial force control transformation and base coordinate system projection to generate a three-dimensional force vector in the base coordinate system. In specific operation, the vertical force correction coefficient is multiplied with the preset reference vertical force to obtain the actual expected vertical force. Then, spatial force control transformation is performed. Further, the actual expected vertical force is executed by instrument coordinate system mapping to obtain the instrument coordinate system force vector. The instrument coordinate system force vector is decomposed into components to obtain standardized force components. The standardized force components are encapsulated with instructions and amplitude constraints to obtain the end vertical force command.

[0100] It should be noted that the preset reference vertical force refers to the nominal vertical force of the surgical robot under standard contact conditions (critical separation of the ultrasound probe and soft tissue), based on the definition of dynamic force control standards (such as surgical robot performance specifications).

[0101] The homogeneous transformation method is applied to project the end vertical force command into the base coordinate system. Furthermore, matrix multiplication is performed on the end vertical force command to obtain the base coordinate system force components. Orthogonal decomposition and dynamic equilibrium are performed on the base coordinate system force components to generate a stable force vector. Base coordinate system alignment is performed on the stable force vector to obtain the three-dimensional force vector of the base coordinate system.

[0102] S4.4. Using the Jacobian matrix transpose formula, the joint torque is calculated on the three-dimensional force vector in the base coordinate system, and the joint control torque of the surgical robot is output. The specific mathematical formula is as follows.

[0103] ;

[0104] in, This indicates the joint control torque of the surgical robot. This indicates the transpose operation. Represents the Jacobian matrix. Represents the three-dimensional force vector in the base coordinate system. This represents the joint friction compensation factor;

[0105] It should be noted that the Jacobian matrix is ​​obtained by performing a joint space velocity mapping operation on the three-dimensional force vector of the base coordinate system. The joint friction compensation factor is defined based on the contact pressure component of the three-dimensional force vector of the base coordinate system, and the exemplary value range is [0.05Nm, 0.5Nm].

[0106] This embodiment also provides a computer device applicable to the force control method of a subcutaneous soft tissue surgical robot, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the force control method of the subcutaneous soft tissue surgical robot as proposed in the above embodiment.

[0107] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0108] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the force control method for a subcutaneous soft tissue surgical robot as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0109] In summary, this invention, through sound velocity calibration analysis and nonlinear correction of viscoelastic compensation factors on soft tissue ultrasound echo datasets, can more accurately obtain the actual state parameters of soft tissues, thereby improving the perception of subcutaneous fat layer deformation and solving the inaccuracy of traditional robot force control methods in predicting tissue deformation. Furthermore, a depth attenuation compensation model is constructed, and through the collaborative working mechanism of the deformation transfer layer and the biomechanical layer, accurate modeling of deep muscle correction deformation is achieved, improving the controllability and safety of complex soft tissue manipulation during surgery.

[0110] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor implements the steps of the force control method of the subcutaneous soft tissue surgical robot when executing the computer program; An ultrasound echo data set of soft tissue is collected, and speed of sound calibration analysis is performed to generate an actual speed of sound value, which is nonlinearly corrected using a tissue viscoelasticity compensation factor to output a real-time corrected speed of sound value; According to the soft tissue ultrasound image and the real-time corrected speed of sound value, thickness conversion is performed to obtain an actual fat thickness value, and the actual fat thickness value and a zero-pressure reference value are differentially operated to form a subcutaneous fat layer deformation variable; The subcutaneous fat layer deformation variable is input into a depth attenuation compensation model, a deformation transmission layer performs attenuation weight distribution, a biomechanics layer performs contact stress distribution calculation, and a deep muscle correction deformation variable is output; The deep muscle correction deformation variable is subjected to tissue state mapping to generate a vertical force correction coefficient, which is input into a surgical robot control center for spatial force control conversion and joint torque calculation to output a surgical robot joint control torque.

2. The computer device of claim 1, wherein: The soft tissue ultrasound echo data set includes a soft tissue ultrasound image, bone tissue transit time, and tissue deformation displacement data.

3. The computer device of claim 2, wherein: The output real-time corrected speed of sound value specifically includes the following steps, The bone tissue interface positioning and speed of sound calibration analysis are performed on the soft tissue ultrasound echo data set to generate an actual speed of sound value; The actual speed of sound value is subjected to biomechanics characteristic mapping to obtain a viscoelasticity compensation parameter; The viscoelasticity compensation parameter is subjected to piecewise interpolation and nonlinear correction using a tissue viscoelasticity compensation factor to output a real-time corrected speed of sound value.

4. The computer device of claim 3, wherein: The actual fat thickness value is obtained by specifically including the following steps, The soft tissue ultrasound image is subjected to average gray scale segmentation to obtain fat layer boundary coordinates; According to the real-time corrected speed of sound value and the fat layer boundary coordinates, physical size conversion and thickness calculation are performed to obtain an actual fat thickness value.

5. The computer device of claim 4, wherein: The formation of the subcutaneous fat layer deformation variable refers to differentially operating the actual fat thickness value and a zero-pressure reference value to obtain a differential thickness value; the differential thickness value is subjected to confidence weighting to form a subcutaneous fat layer deformation variable.

6. The computer device of claim 5, wherein: The output deep muscle correction deformation variable specifically includes the following steps, A deformation transmission layer and a biomechanics layer are built through a gated recurrent network and a residual attention architecture, and a depth attenuation compensation model is constructed by applying a skip connection for weighted stacking; The subcutaneous fat layer deformation variable is input into the depth attenuation compensation model, and the deformation transmission layer uses a depth separable convolution to perform attenuation weight distribution to generate a deformation transmission coefficient; The biomechanics layer performs contact stress distribution calculation using a viscoelasticity constitutive equation to form a stress response feature; The deformation transmission coefficient and the stress response feature are subjected to feature temporal fusion using a bidirectional gate unit to generate a deformation-stress coupling vector; The deformation-stress coupling vector is subjected to multi-scale feature extraction and cross-layer residual aggregation to obtain a deep muscle correction deformation variable.

7. The computer device of claim 6, wherein: The generation of the vertical force correction coefficient specifically includes the following steps, A displacement gradient algorithm is used to model the strain distribution of the deep muscle correction deformation variable to generate a muscle strain feature vector; The muscle strain feature vector is subjected to vertical relative movement angle quantization to form a slope-stress correlation matrix; The slope-stress correlation matrix is mapped to an organization state to generate a normal force correction coefficient.

8. The computer device of claim 7, wherein: The output surgical robot joint control torque specifically comprises the following steps, The normal force correction coefficient is input into a surgical robot control center to perform spatial force control conversion to form an end normal force instruction. A homogeneous transformation method is applied to project the end normal force instruction into a base coordinate system to generate a base coordinate system three-dimensional force vector. Joint torque calculation is performed on the base coordinate system three-dimensional force vector to output a surgical robot joint control torque.

9. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by a processor to implement the steps of the subcutaneous soft tissue surgical robot force control method in claim 1.

Images