Cutting path planning method and system for ultrasonic knife equipment

Abstract

The invention discloses a cutting path planning method and system for ultrasonic knife equipment, and relates to the technical field of medical treatment. The invention discloses a cutting path planning system of ultrasonic knife equipment. The cutting path planning system comprises a cutting path planning module and a cutting path calibration module, according to the method, image recognition is carried out by inputting the cut image annotation model, the cut object feature map is obtained, and the improved space trajectory planning algorithm and the image registration algorithm are utilized, so that the initial cutting path can be efficiently predicted and optimized, and the accuracy of the cutting path is ensured; according to the method, the image of the simulated cutting object is registered, analysis is carried out based on the cutting error judgment model, potential cutting errors can be recognized before an operation, a cutting path is adjusted in advance, and therefore the influence caused by the errors is reduced. And the initial cutting path is updated by using the error judgment result, so that the path deviation caused by the error is effectively prevented.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a cutting path planning method and system for ultrasonic knife equipment. Background Art

[0002] Currently, ultrasonic scalpel devices are mainly used in surgical operations to accurately cut and coagulate soft tissues and reduce bleeding and thermal damage. The importance of ultrasonic scalpel device path planning lies in ensuring the accuracy, safety and efficiency of the operation. Accurate path planning can ensure that the ultrasonic scalpel cuts along the predetermined path, thereby reducing human errors and improving the accuracy of the operation. However, in actual operations, especially when long-term surgery or treatment is involved, the patient may experience uncontrollable displacement. This displacement may be caused by the patient's own physiological activities, such as breathing, heartbeat, changes in body position, muscle relaxation or tension, etc., which may cause the ultrasonic scalpel to deviate from the predetermined cutting path, which will seriously affect the accuracy of the predetermined path.

[0003] Therefore, it is necessary to design a cutting path planning method for ultrasonic knife equipment that takes into account the patient displacement factor. This method simulates the possible displacement of the patient before the operation, predicts and adjusts the planned path, so that it can flexibly respond to the patient's displacement during the operation, ensure the accuracy of the auxiliary cutting path, and thus improve the overall effect of the operation. Summary of the invention

[0004] The present invention aims to provide a cutting path planning method and system for an ultrasonic knife device, which can flexibly respond to patient displacement and ensure the accuracy of the auxiliary cutting path.

[0005] A cutting path planning method for an ultrasonic knife device comprises the following steps: Obtain an image of the object to be cut; the image of the object to be cut contains a marker ball of the object to be cut; input the image of the object to be cut into the cutting image annotation model for image recognition, and obtain a cutting object feature map P 1 ; Cutting object feature map P 1 The cutting feature point and the marking ball feature point O are marked on 1 '; Cut the object feature map P 1 Input into the cutting path planning model for calculation to obtain the initial cutting path; the cutting path planning model is established based on the improved space trajectory planning algorithm; A cutting digital twin model is established based on the object to be cut; based on the cutting digital twin model, M-1 simulated cutting object images and corresponding simulated physiological parameters of the object to be cut are obtained; all simulated cutting object images are input into the cutting image annotation model for image recognition, and a cutting object feature map P is obtained. m and the feature point O of the marker ball m', m=2, ..., M; using the improved image registration algorithm to cut the object feature map P m and the cutting object feature map P m-1 Perform image registration and obtain the image error judgment result T m ; The image error judgment result T m The corresponding simulated physiological parameters of the object to be cut are input into the cutting error judgment model for calculation to obtain the cutting error judgment result J m ; Judgment result based on cutting error J m The initial cutting path is updated to obtain an updated cutting path; and auxiliary path planning is performed on the object to be cut based on the updated cutting path.

[0006] As a preferred technical solution of the present invention, the cut image annotation model includes an image preprocessing layer, an improved convolution layer, an improved attention feature mechanism layer and an improved residual convolution layer; The image preprocessing layer is used to perform image preprocessing operations on the image of the object to be cut, so as to obtain a preprocessed image of the object to be cut; The improved convolution layer is used to perform a convolution operation on the preprocessed image of the object to be cut, so as to obtain a convolution feature map of the object to be cut; In the improved convolution layer, the number of layers of the convolution feature map of the object to be cut is set to N, n=1, 2, ..., N; Using formula R n =R n-1 + (a-1)*(l-1) to calculate the receptive field R of the nth layer n , a is the size of the convolution kernel in the convolution feature map of the object to be cut at the nth layer, l is the expansion rate set in the convolution feature map of the object to be cut at the nth layer; all receptive fields R n Perform convolution calculation to obtain the convolution feature map of the object to be cut; The improved attention feature mechanism layer is used to extract features from the convolution feature map of the object to be cut using the improved attention feature mechanism to obtain the initial feature map of the object to be cut; In the improved attention feature mechanism layer, the output initial feature map of the object to be cut is set to include B layers, b=1, 2, ..., B; Using the formula H=conv 1*1*1 (concat(V d，b *F b , V s，b *F b )+concat(Z d，b *F b , Z s，b *F b )+F b ) Calculate the convolution feature map of the object to be cut to obtain the feature map of the object to be cut; Among them, V d，b Represents the channel attention vector of the initial feature map of the object to be cut at layer b, V d，b =Φ(conv 1*1*1 (conv 1*1*1 (Q(F b-1 ))));V s，b Represents the estimated value of the spatial attention vector of the initial feature map of the object to be cut at layer b, V s，b =Φ(conv 1*1*1 (conv 2*2*2 (conv 1*1*1 (F b-1 ))));Z d，b Represents the channel attention vector of the high-level feature channel after filtering the initial feature map of the object to be cut in the bth layer, Z d，b =Φ(conv 1*1*1 (conv 1*1*1 (Q(F b+1 ))));Z s，b represents the attention mechanism vector contained in the feature space after segmentation of the initial feature map of the object to be cut at layer b, Z s，b =Φ(conv 1*1*1 (conv 2*2*2 (conv 1*1*1 (F b+1 ))));conv( ) represents the convolution operation, concat ( ) represents the feature channel fusion operation, F b represents the low-dimensional feature map of layer b, Q( ) represents the global average pooling operation; Φ( ) represents the sigmoid activation operation; The improved residual convolution layer is used to further extract features from the initial feature map of the object to be cut, and obtain the feature map of the cut object P 1 ; In the improved residual convolution layer, using formula P 1 =σ(W 1*1 (W 1*1 (K(U)))) performs fusion convolution on the initial feature map of the object to be cut to obtain the feature map of the cut object P 1 ; where σ represents the activation function, K(U) represents the average pooling value of the initial feature map of the object to be cut, W represents the convolution kernel, and W 1*1 Represents a 1*1 convolution operation, and U represents the initial feature map of the object to be cut.

[0007] As a preferred technical solution of the present invention, the cutting path planning model includes a feature point recognition layer and a path planning layer; The feature point recognition layer is used to identify the cutting object according to the feature map P 1 Perform feature point recognition to obtain a cutting feature point set; The path planning layer is used to perform path planning based on the cutting feature point set and the improved spatial trajectory planning algorithm to obtain the initial cutting path.

[0008] As a preferred technical solution of the present invention, the specific steps of performing path planning in the path planning layer include: According to the cutting feature point set and the spatial trajectory planning algorithm, I groups of cutting path individuals Y to be optimized are randomly generated i , i=1,2,…,I; I groups of individual cutting paths to be optimized Y i Combine to obtain the optimized cutting path iterative population; set the inertia weight X 1 , inertia weight X 2 and the current number of iterations g, g=1, 2, …, G, G is the maximum number of iterations; For the individual Y of the cutting path to be optimized in the iterative population of the optimized cutting path i Perform simulated ultrasonic knife path cutting calculation to obtain simulated cutting time , the simulated cutting time The reciprocal of is taken as the individual Y of the cutting path to be optimized i Fitness ; When performing population iteration, use the formula: and Calculate and get the inertia weight X 1 and inertia weight X 2 , inertia weight X 1 Represents the individual Y of the cutting path to be optimized in the iterative population of optimized cutting paths i The weight of approaching the current individual extreme point, inertia weight X 2 represents the individual Y of the cutting path to be optimized in the iterative population of optimized cutting paths i The weight of approaching the global individual extreme point; where α is the preset calculation parameter, β is the standard deviation of the Gaussian distribution, g is the current number of iterations; e is a mathematical constant; according to the inertia weight X 1 and inertia weight X 2 Perform crossover and mutation operations on the iterative population of optimized cutting paths; When the current iteration number g=G, the output is the cutting path individual Y to be optimized corresponding to the maximum fitness i , which is the optimal cutting path individual to be optimized; the path in the optimal cutting path individual to be optimized is used as the initial cutting path.

[0009] As a preferred technical solution of the present invention, an improved image registration algorithm is used to register the cutting object feature map P m and the cutting object feature map P m-1 Perform image registration, the specific steps include: Calculate the cutting object feature map P m and the cutting object feature map P m-1 The similarity error of the image registration is obtained. Mark the ball feature point O m ' and mark ball feature point O m-1 'Construct three-dimensional coordinate calibration and obtain the three-dimensional calibration coordinate ZB (P m ') and three-dimensional calibration coordinates ZB (P m-1 '); Calculate the three-dimensional calibration coordinates ZB (P m ') and three-dimensional calibration coordinates ZB (P m-1 '), and obtain the image registration reprojection error and image registration benchmark error; The image error judgment result T is obtained by subtracting the image registration reprojection error and the image registration benchmark error from the overall similarity error of the image registration. m .

[0010] As a preferred technical solution of the present invention, the cutting error judgment model includes a physiological parameter evaluation layer, an image error feature extraction layer and an error factor fusion layer; The physiological parameter evaluation layer is used to extract the characteristics of the simulated physiological parameters of the object to be cut, and obtain the characteristics of the simulated physiological parameters of the object to be cut; The image error feature extraction layer is used to judge the image error result T m Perform feature extraction to obtain the image error judgment result feature T m '; The error factor fusion layer is used to simulate the physiological parameter characteristics of the object to be cut and the image error judgment result characteristics T m 'Carry out error factor fusion calculation to obtain the cutting error judgment result J m .

[0011] A cutting path planning system for an ultrasonic knife device, comprising: The cutting path planning module includes a planning preparation unit and a path planning unit; the planning preparation unit is used to obtain an image of the object to be cut; the image of the object to be cut includes a marking ball of the object to be cut; the image of the object to be cut is input into the cutting image annotation model for image recognition, and a cutting object feature map P is obtained. 1 ; Cutting object feature map P 1 The cutting feature point and the marking ball feature point O are marked on 1'; The path planning unit is used to cut the object feature map P 1 Input into the cutting path planning model for calculation to obtain the initial cutting path; the cutting path planning model is established based on the improved space trajectory planning algorithm; The cutting path calibration module includes an error judgment unit and a path recalibration unit; the error judgment unit is used to establish a cutting digital twin model based on the object to be cut; based on the cutting digital twin model, M-1 simulated cutting object images and corresponding simulated physiological parameters of the object to be cut are obtained; all simulated cutting object images are input into the cutting image annotation model for image recognition to obtain a cutting object feature map P m and the feature point O of the marker ball m ', m=2, ..., M; using the improved image registration algorithm to cut the object feature map P m and the cutting object feature map P m-1 Perform image registration and obtain the image error judgment result T m ; The image error judgment result T m The corresponding simulated physiological parameters of the object to be cut are input into the cutting error judgment model for calculation to obtain the cutting error judgment result J m ; The path recalibration unit is used to judge the result based on the cutting error J m The initial cutting path is updated to obtain an updated cutting path; and auxiliary path planning is performed on the object to be cut based on the updated cutting path.

[0012] The present invention has the following advantages: 1. The present invention performs image recognition by inputting the cutting image annotation model to obtain the cutting object feature map, and uses the improved space trajectory planning algorithm and image registration algorithm to efficiently predict and optimize the initial cutting path to ensure the accuracy of the cutting path; by registering the simulated cutting object image and analyzing it based on the cutting error judgment model, it can identify potential cutting errors before surgery and adjust the cutting path in advance, thereby reducing the impact of the error. The initial cutting path is updated using the error judgment result, effectively preventing the path deviation caused by the error.

[0013] 2. The present invention ensures image quality through the image preprocessing layer, laying a good foundation for subsequent processing. The improved convolution layer dynamically calculates the receptive field to enhance the model's ability to capture image features. The improved attention feature mechanism layer strengthens the recognition of key features and improves the accuracy of feature extraction through the channel and spatial attention mechanism. The improved residual convolution layer further refines features to ensure high-quality output of feature maps. The training process of the entire cut image annotation model is continuously optimized until the preset annotation accuracy is reached, ensuring the practicability and reliability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of the structure of a cutting path planning system for an ultrasonic knife device used in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0016] Embodiment 1, a cutting path planning method of an ultrasonic knife device, comprising the following steps: Obtain an image of the object to be cut; the image of the object to be cut contains a marker ball of the object to be cut; input the image of the object to be cut into the cutting image annotation model for image recognition, and obtain a cutting object feature map P 1 ; Cutting object feature map P 1 The cutting feature point and the marking ball feature point O are marked on 1 '; The object to be cut is the cutting part of the patient who needs surgery, and the marking ball of the object to be cut is a fixed small ball used for position calibration during path planning, which is fixed and consistent with the position of the camera; The cut image annotation model includes an image preprocessing layer, an improved convolution layer, an improved attention feature mechanism layer, and an improved residual convolution layer; The image preprocessing layer is used to perform image preprocessing operations on the image of the object to be cut, so as to obtain a preprocessed image of the object to be cut; The improved convolution layer is used to perform a convolution operation on the preprocessed image of the object to be cut, so as to obtain a convolution feature map of the object to be cut; In the improved convolution layer, the number of layers of the convolution feature map of the object to be cut is set to N, n=1, 2, ..., N; Using formula R n =R n-1 + (a-1)*(l-1) to calculate the receptive field R of the nth layer n , a is the size of the convolution kernel in the convolution feature map of the nth layer of the object to be cut, l is the expansion rate set in the convolution feature map of the nth layer of the object to be cut; all receptive fields R n Perform convolution calculation to obtain the convolution feature map of the object to be cut; set the expansion rate to change according to actual needs, use a smaller expansion rate to extract local information in the image, and use a larger expansion rate to extract long-distance information in the overall image; The improved attention feature mechanism layer is used to extract features from the convolution feature map of the object to be cut using the improved attention feature mechanism to obtain the initial feature map of the object to be cut; In the improved attention feature mechanism layer, the output initial feature map of the object to be cut is set to include B layers, b=1, 2, ..., B; Using the formula H=conv 1*1*1 (concat(V d，b *F b , V s，b *F b )+concat(Z d，b *F b , Z s，b *F b )+F b ) Calculate the convolution feature map of the object to be cut to obtain the feature map of the object to be cut; Among them, V d，b Represents the channel attention vector of the initial feature map of the object to be cut at layer b, V d，b =Φ(conv 1*1*1 (conv 1*1*1 (Q(F b-1 ))));V s，b Represents the estimated value of the spatial attention vector of the initial feature map of the object to be cut at layer b, V s，b =Φ(conv 1*1*1 (conv 2*2*2 (conv 1*1*1 (F b-1 ))));Z d，b Represents the channel attention vector of the high-level feature channel after filtering the initial feature map of the object to be cut in the bth layer, Z d，b =Φ(conv 1*1*1 (conv 1*1*1 (Q(F b+1 ))));Z s，b represents the attention mechanism vector contained in the feature space after segmentation of the initial feature map of the object to be cut at layer b, Z s，b =Φ(conv 1*1*1 (conv 2*2*2 (conv 1*1*1 (F b+1 ))));conv( ) represents the convolution operation, concat ( ) represents the feature channel fusion operation, F b represents the low-dimensional feature map of layer b, Q( ) represents the global average pooling operation; Φ( ) represents the sigmoid activation operation; The improved residual convolution layer is used to further extract features from the initial feature map of the object to be cut, and obtain the feature map of the cut object P 1 ; In the improved residual convolution layer, using formula P 1 =σ(W1*1 (W 1*1 (K(U)))) performs fusion convolution on the initial feature map of the object to be cut to obtain the feature map of the cut object P 1 ; where σ represents the activation function, K(U) represents the average pooling value of the initial feature map of the object to be cut, W represents the convolution kernel, and W 1*1 represents a 1*1 convolution operation, and U represents the initial feature map of the object to be cut; The cut image annotation model is trained using the cut image annotation training set, and the specific steps are as follows: collecting a number of images with feature point annotations and original images, combining the images with feature point annotations and the original images to obtain the cut image annotation training set; using the cut image annotation training set to perform model training on the cut image annotation model to obtain an initial cut image annotation model; if the initial cut image annotation model meets the preset annotation accuracy, the initial cut image annotation model is used as the cut image annotation model; otherwise, the cut image annotation training set is used to continue model training; the preset annotation accuracy is set manually; The improved convolution layer improves the distortion problem in image processing by adjusting the size of the receptive field according to the characteristics of the visual receptive field, thereby improving the overall image quality. The original image is passed through a multi-scale semantic feature focusing module, namely the improved convolution layer, which can extract highly representative features; by implementing the dilated convolution operation, the receptive field of the model is effectively expanded, its ability to capture image details is enhanced, and the efficiency and quality of feature extraction are improved; the feature selection and aggregation process is further optimized through the improved attention feature mechanism module, significantly improving the accuracy of segmentation; the model also finely filters and processes the features through a combination of deconvolution and pooling layers, namely the improved residual convolution layer, to ensure that the final output image has excellent feature extraction effect; Cut the object feature map P 1 Input into the cutting path planning model for calculation to obtain the initial cutting path; the cutting path planning model is established based on the improved space trajectory planning algorithm; The cutting path planning model includes a feature point recognition layer and a path planning layer; The feature point recognition layer is used to identify the cutting object according to the feature map P 1 Perform feature point recognition to obtain a cutting feature point set; The path planning layer is used to perform path planning based on the cutting feature point set and the improved spatial trajectory planning algorithm to obtain the initial cutting path; The specific steps of path planning in the path planning layer include: According to the cutting feature point set and the spatial trajectory planning algorithm, I groups of cutting path individuals Y to be optimized are randomly generated i , i=1,2,…,I; I groups of individual cutting paths to be optimized Yi Combine to obtain the optimized cutting path iterative population; set the inertia weight X 1 , inertia weight X 2 and the current number of iterations g, g=1, 2, …, G, G is the maximum number of iterations, and the maximum number of iterations is set manually; For the individual Y of the cutting path to be optimized in the iterative population of the optimized cutting path i Perform simulated ultrasonic knife path cutting calculation to obtain simulated cutting time , the simulated cutting time The reciprocal of is taken as the individual Y of the cutting path to be optimized i Fitness ; When performing population iteration, use the formula: and Calculate and get the inertia weight X 1 and inertia weight X 2 , inertia weight X 1 represents the individual Y of the cutting path to be optimized in the iterative population of optimized cutting paths i The weight of approaching the current individual extreme point, inertia weight X 2 represents the individual Y of the cutting path to be optimized in the iterative population of optimized cutting paths i The weight of approaching the global individual extreme point; where α is the preset calculation parameter, β is the standard deviation of the Gaussian distribution, g is the current number of iterations; e is a mathematical constant; according to the inertia weight X 1 and inertia weight X 2 Perform crossover and mutation operations on the optimized cutting path iterative population; the crossover operation is based on the inertia weight X 2 Make adjustments based on the inertia weight X 2 The value of , when it is too high, will lead to frequent crossover during global convergence to explore the global optimal solution; the compilation operation is based on the inertia weight X 1 Make adjustments based on the inertia weight X 1 The value of the inertia weight X 1 When the inertia weight X is larger, the frequency of variation is lower, indicating that the individual is mainly moving towards the current optimal path; 1 When it is smaller, the frequency of mutation is higher, increasing the breadth of exploration; Inertia Weight X 1 It is used to control the influence of the current optimal path, which determines how the individual continues to explore based on the current solution and avoids over-reliance on historical solutions; inertia weight X 2 Control the degree of convergence of individuals to the global optimal solution, so that individuals in the population can continuously adjust toward the global optimal solution to explore a wider area of ​​the solution space; When the current iteration number g=G, the output is the cutting path individual Y to be optimized corresponding to the maximum fitness i , which is the optimal cutting path individual to be optimized; the path in the optimal cutting path individual to be optimized is used as the initial cutting path; Through the feature point recognition layer, key feature points can be accurately extracted to provide reliable input data for path planning and improve the accuracy of path planning. The improved spatial trajectory planning algorithm, combined with randomly generated path individuals and optimized iterative processes, can quickly find the optimal cutting path. The feasibility and efficiency of the path are ensured by simulating ultrasonic knife path cutting calculations. The path generation is combined with feature points and spatial trajectory planning algorithms to meet the actual surgical needs of the object to be cut, while minimizing the wound surface and the shortest time. The feasibility and efficiency of the path are ensured by simulating ultrasonic knife path cutting calculations. The inertia weight X is dynamically adjusted. 1 and X 2 , making the path optimization process more flexible and efficient, and being able to balance local and global searches at different iteration stages; through multiple iterations and optimizations, ensuring that the final output path is globally optimal, improving the accuracy and efficiency of cutting; through efficient path planning and optimization algorithms, being able to generate high-quality cutting paths in a relatively short time, which is suitable for real-time application scenarios; A cutting digital twin model is established based on the object to be cut; based on the cutting digital twin model, M-1 simulated cutting object images and corresponding simulated physiological parameters of the object to be cut are obtained; all simulated cutting object images are input into the cutting image annotation model for image recognition, and a cutting object feature map P is obtained. m and the feature point O of the marker ball m ', m=2, ..., M; using the improved image registration algorithm to cut the object feature map P m and the cutting object feature map P m-1 Perform image registration and obtain the image error judgment result T m ; The image error judgment result T m The corresponding simulated physiological parameters of the object to be cut are input into the cutting error judgment model for calculation to obtain the cutting error judgment result J m ; Judgment result based on cutting error J m The initial cutting path is updated to obtain an updated cutting path; and auxiliary path planning is performed on the object to be cut based on the updated cutting path; Digital twin model refers to the construction of virtual models through real-time mapping of data and physical objects, so as to conduct analysis and prediction in a virtual environment; Steps to establish a cutting digital twin model: First, it is necessary to obtain detailed information of the object to be cut through high-precision sensors and scanning equipment. This information includes the object's geometric shape, position, density, hardness and other parameters; Based on the collected data, use computer-aided design software or 3D modeling tools to build a geometric model of the physical object. The patient's anatomical structure can be converted into a virtual three-dimensional model to ensure the accuracy of the model; Based on the geometric model, physical properties need to be added to each part, such as the elastic modulus, hardness, density, etc. of the tissue. These properties usually come from medical image processing, physical simulation and experimental data; Simulate the constraints that may be encountered during the cutting process to ensure that the digital twin model can not only reflect the morphology of the target object, but also reflect the physical processes related to the surgery;

[0017] The purpose of acquiring the simulated cutting object image and the corresponding physiological parameters is to predict and optimize the surgical process and simulate the displacement movements that may occur during the cutting process; applying the cutting algorithm in the digital twin model, such as cutting simulation based on the finite element method or other physical simulation technologies to simulate cutting, this process requires simulating the interaction between the tool and the object, including mechanical behavior, thermal effects, etc.; whenever the simulated object is offset, the digital twin model updates and generates a new object image, namely the simulated cutting object image, which represents the simulated effect of the movement of the patient's cutting part during the cutting process; The improved image registration algorithm is used to identify the cutting object feature map P m and the cutting object feature map P m-1 Perform image registration, the specific steps include: Calculate the cutting object feature map P m and the cutting object feature map P m-1 The similarity error of the image registration is obtained. Mark the ball feature point O m ' and mark ball feature point O m-1 'Construct three-dimensional coordinate calibration and obtain the three-dimensional calibration coordinate ZB (P m ') and three-dimensional calibration coordinates ZB (P m-1 '); Calculate the three-dimensional calibration coordinates ZB (P m ') and three-dimensional calibration coordinates ZB (P m-1 '), and obtain the image registration reprojection error and image registration benchmark error; The image error judgment result T is obtained by subtracting the image registration reprojection error and the image registration benchmark error from the overall similarity error of the image registration. m ; By calculating the overall similarity error of image registration, the matching degree between two images can be accurately evaluated to ensure high precision of registration; the use of three-dimensional coordinate calibration further improves the accuracy of registration, especially in complex scenes of simulated cutting, by calculating the reprojection error and benchmark error, it can effectively reduce the registration error caused by factors such as noise and occlusion, and improve the robustness of registration; by comprehensively considering the overall similarity error, reprojection error and benchmark error, it can comprehensively evaluate the accuracy of the registration result and ensure the quality of registration; through efficient similarity calculation and error evaluation methods, image registration can be completed in a relatively short time; The cutting error judgment model includes a physiological parameter evaluation layer, an image error feature extraction layer and an error factor fusion layer; The physiological parameter evaluation layer is used to extract the characteristics of the simulated physiological parameters of the object to be cut, and obtain the characteristics of the simulated physiological parameters of the object to be cut; The image error feature extraction layer is used to judge the image error result T m Perform feature extraction to obtain the image error judgment result feature T m '; The error factor fusion layer is used to simulate the physiological parameter characteristics of the object to be cut and the image error judgment result characteristics T m 'Carry out error factor fusion calculation to obtain the cutting error judgment result J m ; The specific steps of training the error factor fusion layer in the cutting error judgment model include: Collecting several groups of error factor fusion training samples; each group of error factor fusion training samples contains physiological parameter features, image error features and annotated fusion cutting error results; combining several groups of error factor fusion training samples to obtain an error factor fusion training set; The error factor fusion training set is input into the cutting error judgment model to train the error factor fusion layer to obtain the initial error factor fusion layer; the initial error factor fusion layer is evaluated to obtain the initial error factor fusion layer model evaluation result; if the initial error factor fusion layer model evaluation result is passed, the initial error factor fusion layer is used as the error factor fusion layer in the cutting error judgment model; otherwise, the error factor fusion training set is used to continue model training; Through the physiological parameter evaluation layer and the image error feature extraction layer, the key features that affect the cutting error can be fully extracted to improve the accuracy of error evaluation; the combination of physiological parameter features and image error features enables the model to predict the cutting error more accurately; the model can adapt to different types of objects to be cut and different cutting environments, and has strong adaptability; through multi-step feature extraction and error evaluation, the final cutting error judgment result is ensured to be reliable, the possibility of misjudgment is reduced, and the time for re-path planning is saved; If the cutting error judgment result J m If it does not meet the cutting path planning threshold, then according to all cutting object feature maps P m The initial cutting path is updated to obtain an updated cutting path; otherwise, the initial cutting path is maintained and used as the updated cutting path; the cutting path planning threshold is set manually; Based on the cutting object feature map P m A new path planning is performed with the cutting path planning model to obtain M-1 process cutting paths; the M-1 process cutting paths and the initial cutting path are weighted averaged to perform feature fusion to obtain an updated cutting path.

[0018] In this embodiment, for example, before an ultrasonic scalpel surgery for tumor resection, the above-mentioned ultrasonic scalpel device cutting path planning method is used to assist path planning; before the surgery, firstly, an image of the patient's object to be cut is obtained through imaging examination, and the patient is scanned by CT / MRI or ultrasonic equipment to obtain a high-resolution image of the object to be cut (such as a tumor or other tissue area); the image of the object to be cut is input into the cutting image annotation model, and the feature points of the image are extracted through the improved image preprocessing layer, convolution layer, attention feature mechanism layer and residual convolution layer; this step will generate a cutting object feature map, and mark the cutting feature points and marker ball feature points; the cutting object feature map will be input into the cutting path planning model for path calculation; in this process, an improved spatial trajectory planning algorithm is used to generate an initial cutting path; based on the object to be cut, Image and physiological parameters, establish a cutting digital twin model, which can accurately reflect the geometric shape, physical characteristics and physiological state of the object to be cut; obtain several simulated cutting object images in the cutting digital twin model, simulate the displacement action of the object to be cut during the cutting process, and obtain the corresponding simulated physiological parameters; use the improved image registration algorithm to align all simulated cutting object images to obtain the image error judgment result. This step ensures that the path planning model can accurately reflect the changes in different cutting scenarios by calculating the image similarity error, reprojection error and reference error; input the image error judgment result and the simulated physiological parameters of the object to be cut into the cutting error judgment model for calculation to obtain the cutting error judgment result; based on this result, the initial cutting path is updated to ensure that the path planning has high accuracy and stability.

[0019] Example 2, a cutting path planning system for an ultrasonic knife device, see Figure 1 As shown, including: The cutting path planning module includes a planning preparation unit and a path planning unit; the planning preparation unit is used to obtain an image of the object to be cut; the image of the object to be cut includes a marking ball of the object to be cut; the image of the object to be cut is input into the cutting image annotation model for image recognition, and a cutting object feature map P is obtained. 1 ; Cutting object feature map P 1 The cutting feature point and the marking ball feature point O are marked on 1 '; The path planning unit is used to cut the object feature map P 1 Input into the cutting path planning model for calculation to obtain the initial cutting path; the cutting path planning model is established based on the improved space trajectory planning algorithm; The cutting path calibration module includes an error judgment unit and a path recalibration unit; the error judgment unit is used to establish a cutting digital twin model based on the object to be cut; based on the cutting digital twin model, M-1 simulated cutting object images and corresponding simulated physiological parameters of the object to be cut are obtained; all simulated cutting object images are input into the cutting image annotation model for image recognition to obtain a cutting object feature map P m and the feature point O of the marker ball m ', m=2, ..., M; using the improved image registration algorithm to cut the object feature map P m and the cutting object feature map P m-1 Perform image registration and obtain the image error judgment result T m ; The image error judgment result T m The corresponding simulated physiological parameters of the object to be cut are input into the cutting error judgment model for calculation to obtain the cutting error judgment result J m ; The path recalibration unit is used to judge the result based on the cutting error J m The initial cutting path is updated to obtain an updated cutting path; and auxiliary path planning is performed on the object to be cut based on the updated cutting path.

[0020] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A cutting path planning method for an ultrasonic knife device, characterized in that: The following steps are involved: Acquire an image of the object to be cut; the image of the object to be cut includes a marker ball of the object to be cut; input the image of the object to be cut into a cutting image annotation model for image recognition to obtain a cutting object feature map P1; the cutting object feature map P1 is annotated with cutting feature points and marker ball feature points O1'; Input the cutting object feature map P1 into the cutting path planning model for calculation to obtain the initial cutting path; The cutting path planning model is established based on the improved spatial trajectory planning algorithm; A cutting digital twin model is established based on the object to be cut; based on the cutting digital twin model, M-1 simulated cutting object images and corresponding simulated physiological parameters of the object to be cut are obtained; all simulated cutting object images are input into the cutting image annotation model for image recognition, and a cutting object feature map P is obtained. m and the feature point O of the marker ball m ', m = 2, ..., M; The improved image registration algorithm is used to identify the cutting object feature map P m and the cutting object feature map P m-1 Perform image registration and obtain the image error judgment result T m ; The image error judgment result T m The corresponding simulated physiological parameters of the object to be cut are input into the cutting error judgment model for calculation to obtain the cutting error judgment result J m ; Based on the cutting error judgment result J m The initial cutting path is updated to obtain an updated cutting path; Auxiliary path planning is performed on the object to be cut based on the updated cutting path.

2. The cutting path planning method of an ultrasonic knife device according to claim 1, characterized in that: The cut image annotation model includes an image preprocessing layer, an improved convolution layer, an improved attention feature mechanism layer, and an improved residual convolution layer; The image preprocessing layer is used to perform image preprocessing operations on the image of the object to be cut, so as to obtain a preprocessed image of the object to be cut; The improved convolution layer is used to perform a convolution operation on the preprocessed image of the object to be cut, so as to obtain a convolution feature map of the object to be cut; In the improved convolution layer, the number of layers of the convolution feature map of the object to be cut is set to N, n=1, 2, ..., N; Using formula R n =R n-1 + (a-1)*(l-1) to calculate the receptive field R of the nth layer n , a is the size of the convolution kernel in the convolution feature map of the nth layer of the object to be cut, l is the expansion rate set in the convolution feature map of the nth layer of the object to be cut; all receptive fields R n Perform convolution calculation to obtain the convolution feature map of the object to be cut; The improved attention feature mechanism layer is used to extract features from the convolution feature map of the object to be cut using the improved attention feature mechanism to obtain the initial feature map of the object to be cut; In the improved attention feature mechanism layer, the output initial feature map of the object to be cut is set to include B layers, b=1, 2, ..., B; Using the formula H=conv 1*1*1 (concat(V d，b *F b , V s，b *F b )+concat(Z d，b *F b , Z s，b *F b )+F b ) Calculate the convolution feature map of the object to be cut to obtain the feature map of the object to be cut; Among them, V d，b Represents the channel attention vector of the initial feature map of the object to be cut at layer b, V d，b =Φ(conv 1*1*1 (conv 1*1*1 (Q(F b-1 ))));V s，b Represents the estimated value of the spatial attention vector of the initial feature map of the object to be cut at layer b, V s，b =Φ(conv 1*1*1 (conv 2*2*2 (conv 1*1*1 (F b-1 ))));Z d，b Represents the channel attention vector of the high-level feature channel after filtering the initial feature map of the object to be cut in the bth layer, Z d，b =Φ(conv 1*1*1 (conv 1*1*1 (Q(F b+1 ))));Z s，b represents the attention mechanism vector contained in the feature space after segmentation of the initial feature map of the object to be cut at layer b, Z s，b =Φ(conv 1*1*1 (conv 2*2*2 (conv 1*1*1 (F b+1 ))));conv( ) represents the convolution operation, concat ( ) represents the feature channel fusion operation, F b represents the low-dimensional feature map of layer b, Q( ) represents the global average pooling operation; Φ( ) represents the sigmoid activation operation; The improved residual convolution layer is used to further extract features from the initial feature map of the object to be cut, and obtain the feature map P1 of the cut object; In the improved residual convolution layer, the formula P1=σ(W 1*1 (W 1*1 (K(U)))) performs fusion convolution on the initial feature map of the object to be cut to obtain the feature map of the cut object P1; where σ represents the activation function, K(U) represents the value of the average pooling of the initial feature map of the object to be cut, W represents the convolution kernel, and W 1*1 Represents a 1*1 convolution operation, and U represents the initial feature map of the object to be cut.

3. The cutting path planning method of an ultrasonic knife device according to claim 2, characterized in that: The cutting path planning model includes a feature point recognition layer and a path planning layer; The feature point recognition layer is used to recognize feature points according to the cutting object feature map P1 to obtain a cutting feature point set; The path planning layer is used to perform path planning based on the cutting feature point set and the improved spatial trajectory planning algorithm to obtain the initial cutting path.

4. The cutting path planning method of an ultrasonic knife device according to claim 3, characterized in that: The specific steps of path planning in the path planning layer include: According to the cutting feature point set and the spatial trajectory planning algorithm, I groups of cutting path individuals Y to be optimized are randomly generated i , i=1,2,…,I; I groups of individual cutting paths to be optimized Y i Combine to obtain the optimized cutting path iteration population; set the inertia weight X1, inertia weight X2 and the current iteration number g, g = 1, 2, ..., G, G is the maximum iteration number; For the individual Y of the cutting path to be optimized in the iterative population of the optimized cutting path i Perform simulated ultrasonic knife path cutting calculation to obtain simulated cutting time , the simulated cutting time The reciprocal of is taken as the individual Y of the cutting path to be optimized i Fitness ; When performing population iteration, use the formula: and Calculate and obtain the inertia weight X1 and inertia weight X2. The inertia weight X1 represents the individual Y of the cutting path to be optimized in the iterative population of the optimized cutting path. i The weight of approaching the current individual extreme point, the inertia weight X2 represents the individual Y of the cutting path to be optimized in the iterative population of the optimized cutting path i The weight of approaching the global individual extreme point; where α is the preset calculation parameter, β is the standard deviation that conforms to the Gaussian distribution, g is the current number of iterations; e is a mathematical constant; according to the inertia weight X1 and the inertia weight X2, the crossover and mutation operations are performed on the optimized cutting path iteration population; When the current iteration number g=G, the output is the cutting path individual Y to be optimized corresponding to the maximum fitness i , which is the optimal cutting path individual to be optimized; the path in the optimal cutting path individual to be optimized is used as the initial cutting path.

5. A cutting path planning method for ultrasonic knife equipment according to claim 4, characterized in that: The improved image registration algorithm is used to identify the cutting object feature map P m and the cutting object feature map P m-1 Perform image registration, the specific steps include: Calculate the cutting object feature map P m and the cutting object feature map P m-1 The similarity error of the image registration is obtained. Mark the ball feature point O m ' and mark ball feature point O m-1 'Construct three-dimensional coordinate calibration and obtain the three-dimensional calibration coordinate ZB (P m ') and three-dimensional calibration coordinates ZB (P m-1 '); Calculate the three-dimensional calibration coordinates ZB (P m ') and three-dimensional calibration coordinates ZB (P m-1 '), and obtain the image registration reprojection error and image registration benchmark error; The image error judgment result T is obtained by subtracting the image registration reprojection error and the image registration benchmark error from the overall similarity error of the image registration. m .

6. A cutting path planning method for ultrasonic knife equipment according to claim 5, characterized in that: The cutting error judgment model includes a physiological parameter evaluation layer, an image error feature extraction layer and an error factor fusion layer; The physiological parameter evaluation layer is used to extract the characteristics of the simulated physiological parameters of the object to be cut, and obtain the characteristics of the simulated physiological parameters of the object to be cut; The image error feature extraction layer is used to judge the image error result T m Perform feature extraction to obtain the image error judgment result feature T m '; The error factor fusion layer is used to simulate the physiological parameter characteristics of the object to be cut and the image error judgment result characteristics T m 'Carry out error factor fusion calculation to obtain the cutting error judgment result J m .

7. A cutting path planning system for an ultrasonic knife device, characterized in that: The system applies a cutting path planning method for ultrasonic knife equipment according to any one of claims 1 to 6, comprising: The cutting path planning module includes a planning preparation unit and a path planning unit; the planning preparation unit is used to obtain an image of the object to be cut; the image of the object to be cut includes a marking ball of the object to be cut; the image of the object to be cut is input into the cutting image annotation model for image recognition to obtain a cutting object feature map P1; the cutting object feature map P1 is annotated with cutting feature points and marking ball feature points O1'; the path planning unit is used to input the cutting object feature map P1 into the cutting path planning model for calculation to obtain an initial cutting path; the cutting path planning model is established based on the improved spatial trajectory planning algorithm; The cutting path calibration module includes an error judgment unit and a path recalibration unit; the error judgment unit is used to establish a cutting digital twin model based on the object to be cut; based on the cutting digital twin model, M-1 simulated cutting object images and corresponding simulated physiological parameters of the object to be cut are obtained; all simulated cutting object images are input into the cutting image annotation model for image recognition to obtain a cutting object feature map P m and the feature point O of the marker ball m ', m=2, ..., M; using the improved image registration algorithm to cut the object feature map P m and the cutting object feature map P m-1 Perform image registration and obtain the image error judgment result T m ; The image error judgment result T m The corresponding simulated physiological parameters of the object to be cut are input into the cutting error judgment model for calculation to obtain the cutting error judgment result J m ; The path recalibration unit is used to judge the result based on the cutting error J m The initial cutting path is updated to obtain an updated cutting path; and auxiliary path planning is performed on the object to be cut based on the updated cutting path.

Images