Mechanical arm thoracocentesis path planning system and method based on dynamic adaptive ant colony optimization algorithm

Abstract

The invention relates to a mechanical arm thoracocentesis path planning system and method based on a dynamic adaptive ant colony optimization algorithm, and belongs to the technical field of path planning. The system comprises a data preprocessing module, an image segmentation module, a path planning module and a visualization module; the data preprocessing module is used for performing denoising and contrast enhancement on the chest CT image so as to improve the image quality and facilitate subsequent image segmentation and analysis; the image segmentation module is used for automatically segmenting tumor, blood vessel, trachea and bone areas through DenseNet and providing data support for path planning; the path planning module is used for selecting an optimal path or an approximate global optimal path by using a dynamic adaptive ant colony optimization algorithm; and the visualization module is used for presenting the optimal path or the approximate global optimal path on the original chest CT image. The system uses an image segmentation technology and an improved ant colony algorithm to provide an optimal puncture path for puncture.

Description

Technical Field

[0001] The present invention relates to a robotic arm thoracentesis path planning system and method based on a dynamic adaptive ant colony optimization algorithm, belonging to the technical field of path planning. Background Art

[0002] Thoracentesis is a common clinical procedure, often used for the diagnosis and treatment of tumors. Traditionally, thoracentesis requires physicians to manually determine the puncture path based on imaging findings. This is time-consuming and carries a high risk of interference with blood vessels or other vital structures, leading to surgical failure or serious harm to the patient. With the development of computer-assisted technology, more and more auxiliary tools are being used in puncture path planning to improve puncture accuracy and safety.

[0003] Existing path planning methods mostly focus on the operational control of surgical robots or robotic arms. However, in practice, how to effectively integrate medical imaging to accurately identify tumors and blood vessels, while ensuring that the puncture path avoids vital structures such as blood vessels, remains an urgent challenge. Furthermore, existing methods fail to fully consider the complex anatomy of the chest cavity during path planning, resulting in inaccurate path selection and impacting treatment outcomes.

[0004] Therefore, how to accurately segment tumors and blood vessels based on chest CT images and plan a safe path that avoids blood vessels has become a key technology to improve the accuracy of thoracentesis. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a robotic arm thoracotomy path planning system and method based on a dynamic adaptive ant colony optimization algorithm. The system uses image segmentation technology and an improved ant colony algorithm to provide the optimal puncture path for clinical puncture.

[0006] The technical solutions of the present invention are as follows: A robotic arm thoracentesis path planning system based on a dynamic adaptive ant colony optimization algorithm. Based on image segmentation and an improved ant colony algorithm, the system includes a data preprocessing module, an image segmentation module, a path planning module, and a visualization module. The data preprocessing module is used to denoise and enhance the contrast of chest CT images to improve image quality and facilitate subsequent image segmentation and analysis; the image segmentation module is used to automatically segment tumors, blood vessels, trachea, and bone areas through DenseNet to provide data support for path planning; the path planning module is used to select the optimal path or an approximate global optimal path using a dynamic adaptive ant colony optimization algorithm; and the visualization module is used to present the optimal path or an approximate global optimal path on the original chest CT image.

[0007] Preferably, the data preprocessing module removes noise and enhances the contrast of the image by performing the following steps: Image denoising: Assume the original chest CT image is I raw , using median filtering I raw Perform denoising to obtain a denoised image; Contrast enhancement: Logarithmic transformation is performed on the grayscale values ​​of the denoised image to expand the grayscale values ​​of the darker areas of the image and compress the grayscale values ​​of the brighter areas, thereby enhancing the contrast of the image. The chest CT image after denoising and contrast enhancement is , H is I The height of W is I The width of C is I The number of channels.

[0008] Preferably, the image segmentation module performs the following steps: Input: , process: Use DenseNet network for image segmentation; output: The category to which each pixel belongs includes tumor, blood vessel, trachea, bone, and background.

[0009] Further preferably, this module uses a DenseNet network to automatically segment tumors, blood vessels, trachea, and bone areas, providing avoidance areas and target areas for subsequent path planning.

[0010] The DenseNet network consists of an initial convolutional layer, multiple dense blocks, and a fully connected layer. There is a transition layer between the two dense blocks. The specific steps are as follows: 1) Initial convolutional layer use α The size is k 1× k 1 convolution kernel for feature extraction with a step size of s 1. Get the local feature map F 1, and use ReLU The activation function introduces nonlinear relationships; 2) Dense blocks use β dense blocks, each of which consists of δ The convolutional layers are composed of, and the output of each convolutional layer is directly connected to all subsequent layers; in each dense block, the convolutional layer uses k 2× k The convolution kernel of 2 is padded with 1 to keep the feature map size unchanged; β Feature maps ; 3) Transition layer The transition layer is located between two dense blocks The main function is to reduce the resolution and number of feature maps and reduce the amount of calculation. The transition layer contains a 1×1 convolution layer to reduce the number of channels, followed by a p 1× p 1 average pooling layer with a step size of s 2. Compress the feature map size to obtain the feature map F 3; 4) Fully connected layer and classifier Using a fully connected layer for chest CT images The segmentation tasks of tumors, blood vessels, trachea, and bones are performed using softmax The function calculates the category probability of each pixel: (1) in It's a pixel i The probability of being classified into the γth class, is the number of categories, The value is {0 , 1 , 2 , 3 , 4}, representing background, tumor, blood vessel, trachea and bone respectively, s γ is of category γ logit value, s i For category i of logit Value; select the category with the highest probability as the pixel i The final category : (2) in Represents pixel points i Belongs to background, tumors, blood vessels, trachea and bones; 5) Loss Function Due to the uneven distribution of background, tumor, blood vessel, trachea and bone categories, the module uses cross entropy loss during training, which is defined as follows: (3) in It's a pixel i The true label of the corresponding category γ, 1 means that the pixel belongs to this category, and 0 means that it does not belong to this category; N is the number of pixels.

[0011] Preferably, the path planning module performs the following steps: Input: image segmentation result ; Process: Use dynamic adaptive ant colony optimization algorithm to select the optimal path or approximate global optimal path; Output: optimal path or approximate global optimal path.

[0012] Furthermore, the present invention proposes a dynamic adaptive ant colony optimization algorithm, which dynamically adjusts the number of ants and the pheromone update strategy according to the path exploration progress to improve the global search efficiency. The robotic arm thoracotomy uses a straight path. In order to ensure that the path selected by the ants is a straight line, the dynamic adaptive ant colony optimization algorithm limits the path selection to a straight path from the starting point to the end point. This means that during the path selection process, the ants will only choose paths that move in a straight line and avoid any curved paths. The equation of the straight line from the starting point to the end point: For each ant, from the starting point to the end point P start ( x 0 ,y 0 ,z 0) To the end P end ( x n , y n ,z n ) can be expressed by the following formula: (4) The parameters t Indicates different positions on the straight line from the starting point to the end point; when t =0, ( x ( t ) ,y ( t ) ,z ( t ))= P start ( x 0 ,y 0 ,z 0); when t =1, ( x ( t ) ,y ( t ) ,z ( t ))= P end ( x n ,y n ,z n ).

[0013] Further preferably, the steps of the dynamic adaptive ant colony optimization algorithm are: 1) Initialization Set the start and end points: Set the start point P start ( x 0 ,y 0 ,z 0) and the end point P end ( x n ,y n ,z n ), and the end point P end ( x n , y n ,z n ) may only belong to the category of tumours; Initialize the number of ant colonies to m ; Initialize pheromone: the pheromone of all paths is initialized to be equal; 2) Ant path selection strategy In each iteration, the ant selects a path based on the straight line equation between the current position and the target point, ensuring that the path is along a straight line in three-dimensional space; the path selection is based on the pheromone concentration. and heuristic information To decide, ants choose from nodes i To Node j The probability of the path is: (5) in It is a slave node i To Node j The pheromone intensity reflects the path ij the importance of Is the path ij The heuristic information of , the calculation process is shown in step 4), and when ants choose a path, they tend to choose a path with larger heuristic information; λ and μ are the weight coefficients of pheromone and heuristic information, respectively, which are used to balance the influence of pheromone on path selection and the role of heuristic information; k is the set of next-step nodes that the ant can choose; 3) Pheromone Update The rules for pheromone updates are as follows: (6) in ρ It is the pheromone evaporation factor, which is used to control the decay rate of pheromones; Represents the pheromone update amount, , d ij Is the path ij length; Q is a constant used to control the pheromone increment; 4) Calculation of heuristic information: (7) in d ij Representation node i To Node j The Euclidean distance of f ( ij ) is the obstacle function; (8) (9); 5) Termination Generally speaking, the ant colony will find an optimal path that has both high pheromone concentration and conforms to the heuristic information. After multiple iterations or reaching the set maximum number of iterations, it will eventually converge to the global optimal or near-optimal path, that is, adopt P ij The largest path ij .

[0014] Preferably, the visualization module displays the searched optimal path or the approximate global optimal path on the original chest CT image. I raw Mark on.

[0015] A method for planning a path for thoracic puncture by a robotic arm based on a dynamic adaptive ant colony optimization algorithm is disclosed. The method executes the steps in a module of the system for planning a path for thoracic puncture by a robotic arm based on a dynamic adaptive ant colony optimization algorithm based on chest CT images.

[0016] The beneficial effects of the present invention are: This invention is used in clinical puncture, particularly in the diagnosis and treatment of thoracic tumors. By combining chest CT images, image segmentation technology, and a path planning algorithm, it enables accurate and safe thoracic puncture path planning. This method can avoid critical structures such as blood vessels and the trachea, providing an optimal puncture path and possessing significant clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the technical route of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described below with reference to embodiments and accompanying drawings, but is not limited thereto.

[0019] Example 1: A robotic arm thoracentesis path planning system based on a dynamic adaptive ant colony optimization algorithm is proposed. The system is based on image segmentation and an improved ant colony algorithm. The system includes a data preprocessing module, an image segmentation module, a path planning module, and a visualization module.

[0020] The data preprocessing module is used to denoise and enhance the contrast of chest CT images to improve image quality and facilitate subsequent image segmentation and analysis; the image segmentation module is used to automatically segment tumors, blood vessels, trachea, and bone areas through DenseNet to provide data support for path planning; the path planning module is used to select the optimal path or an approximate global optimal path using a dynamic adaptive ant colony optimization algorithm; and the visualization module is used to present the optimal path or an approximate global optimal path on the original chest CT image.

[0021] The data preprocessing module removes noise and enhances the contrast of the image using the following steps: Image denoising: Assume the original chest CT image is I raw , using median filtering I raw Perform denoising to obtain the denoised image.

[0022] Contrast enhancement: Logarithmic transformation is performed on the grayscale values ​​of the denoised image to expand the grayscale values ​​of the darker areas of the image and compress the grayscale values ​​of the brighter areas, thereby enhancing the contrast of the image. The chest CT image after denoising and contrast enhancement is , H is I The height of W is I The width of C is I The number of channels.

[0023] The image segmentation module performs the following steps: Input: , process: Use DenseNet network for image segmentation; output: The category to which each pixel belongs includes tumor, blood vessel, trachea, bone, and background.

[0024] This module uses the DenseNet network to automatically segment tumors, blood vessels, trachea, and bone areas, providing avoidance areas and target areas for subsequent path planning.

[0025] The DenseNet network consists of an initial convolutional layer, multiple dense blocks, and a fully connected layer. There is a transition layer between the two dense blocks. The specific steps are as follows: 1) Initial convolutional layer use α The size is k 1× k1 convolution kernel for feature extraction with a step size of s 1. Get the local feature map F 1, and use ReLU The activation function introduces nonlinear relationships.

[0026] 2) Dense blocks use β dense blocks, each of which consists of δ The convolutional layers are composed of, and the output of each convolutional layer is directly connected to all subsequent layers; in each dense block, the convolutional layer uses k 2× k The convolution kernel of 2 is padded with 1 to keep the feature map size unchanged; β Feature maps .

[0027] 3) Transition layer The transition layer is located between two dense blocks The main function is to reduce the resolution and number of feature maps and reduce the amount of calculation. The transition layer contains a 1×1 convolution layer to reduce the number of channels, followed by a p 1× p 1 average pooling layer with a step size of s 2. Compress the feature map size to obtain the feature map F 3.

[0028] 4) Fully connected layer and classifier Using a fully connected layer for chest CT images The segmentation tasks of tumors, blood vessels, trachea, and bones are performed using softmax The function calculates the category probability of each pixel: (1) in It's a pixel i The probability of being classified into the γth class, is the number of categories, The value is {0 , 1 , 2 , 3 , 4}, representing background, tumor, blood vessel, trachea and bone respectively, s γ is of category γ logit value, s i For category i of logit Value; select the category with the highest probability as the pixel i The final category : (2) in Represents pixel points i Belongs to background, tumor, blood vessels, trachea and bones.

[0029] 5) Loss Function Due to the uneven distribution of background, tumor, blood vessel, trachea and bone categories, the module uses cross entropy loss during training, which is defined as follows: (3) in It's a pixel i The true label of the corresponding category γ, 1 means that the pixel belongs to this category, and 0 means that it does not belong to this category; N is the number of pixels.

[0030] The path planning module performs the following steps: Input: Image segmentation result ; Process: Use dynamic adaptive ant colony optimization algorithm to select the optimal path or approximate global optimal path; Output: optimal path or approximate global optimal path.

[0031] The present invention proposes a dynamic adaptive ant colony optimization algorithm that dynamically adjusts the number of ants and the pheromone update strategy according to the path exploration progress to improve the global search efficiency. The robotic arm chest puncture uses a straight path. To ensure that the path selected by the ants is a straight line, the dynamic adaptive ant colony optimization algorithm limits the path selection to a straight line path from the starting point to the end point. This means that during the path selection process, the ants will only choose paths that move in a straight line and avoid any curved paths. The equation of the straight line from the starting point to the end point: For each ant, from the starting point to the end point P start ( x 0 ,y 0 ,z 0) To the end P end ( x n ,y n ,z n ) can be expressed by the following formula: (4) The parameters t Indicates different positions on the straight line from the starting point to the end point; when t =0, ( x ( t ) ,y ( t ) ,z ( t ))= P start ( x 0,y 0 ,z 0); when t =1, ( x ( t ) ,y ( t ) ,z ( t ))= P end ( x n ,y n ,z n ).

[0032] The steps of dynamic adaptive ant colony optimization algorithm are: 1) Initialization Set the start and end points: Set the start point P start ( x 0 ,y 0 ,z 0) and the end point P end ( x n ,y n ,z n ), and the end point P end ( x n , y n ,z n ) can only belong to the tumor category.

[0033] Initialize the number of ant colonies to m .

[0034] Initialize pheromone: The pheromone of all paths is initialized to be equal.

[0035] 2) Ant path selection strategy In each iteration, the ant selects a path based on the straight line equation between the current position and the target point, ensuring that the path is along a straight line in three-dimensional space; the path selection is based on the pheromone concentration. and heuristic information To decide, ants choose from nodes i To Node j The probability of the path is: (5) in It is a slave node i To Nodej The pheromone intensity reflects the path ij the importance of Is the path ij The heuristic information of , the calculation process is shown in step 4), and when ants choose a path, they tend to choose a path with larger heuristic information; λ and μ are the weight coefficients of pheromone and heuristic information, respectively, which are used to balance the influence of pheromone on path selection and the role of heuristic information; k It is the set of next-step nodes that the ant can choose.

[0036] 3) Pheromone Update The rules for pheromone updates are as follows: (6) in ρ It is the pheromone evaporation factor, which is used to control the decay rate of pheromones; Represents the pheromone update amount, , d ij Is the path ij length; Q is a constant used to control the pheromone increment.

[0037] 4) Calculation of heuristic information: (7) in d ij Representation node i To Node j The Euclidean distance of f ( ij ) is the obstacle function; (8) (9).

[0038] 5) Termination Generally speaking, the ant colony will find an optimal path that has both high pheromone concentration and conforms to the heuristic information. After multiple iterations or reaching the set maximum number of iterations, it will eventually converge to the global optimal or near-optimal path, that is, adopt P ij The largest path ij .

[0039] The visualization module displays the searched optimal path or approximate global optimal path on the original chest CT image. I raw Mark on.

[0040] Assume that the chest CT image is I (x, y, z ), the important structural areas are R i The tumor area is R t , the set of pixels contained in the puncture path is R p The starting point of the puncture path is ( x s ,y s ,z s ), the end point of the puncture path is ( x t ,y t ,z t ), the path planned by this system is verified and evaluated as follows.

[0041] The puncture path must first ensure that important structures are avoided to ensure the patient's life safety. , indicating that important structures such as blood vessels were successfully avoided.

[0042] Under the premise of avoiding important structures, check whether the end point of the puncture path belongs to the tumor area. , indicating that the puncture path planning is successful and the puncture is accurate.

[0043] Under the premise of avoiding important structures such as blood vessels and that the end point of the puncture path is in the tumor area, the degree of trauma to healthy tissue is evaluated. The puncture path length is The shorter the puncture path length, the better. For different parts, the paths planned by this system can achieve shorter path lengths.

[0044] Example 2: A method for planning a robotic arm thoracic puncture path based on a dynamic adaptive ant colony optimization algorithm is provided. The method executes the steps in the module of the robotic arm thoracic puncture path planning system based on a dynamic adaptive ant colony optimization algorithm described in Example 1 based on chest CT images.

Claims

1. A robotic arm thoracentesis path planning system based on a dynamic adaptive ant colony optimization algorithm, characterized in that: The system includes data preprocessing module, image segmentation module, path planning module, and visualization module; The data preprocessing module is used to denoise and enhance the contrast of chest CT images; the image segmentation module is used to automatically segment tumors, blood vessels, trachea, and bone areas using DenseNet; the path planning module is used to select the optimal path or an approximate global optimal path using a dynamic adaptive ant colony optimization algorithm; and the visualization module is used to present the optimal path or an approximate global optimal path on the original chest CT image.

2. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 1 is characterized in that: The data preprocessing module removes noise and enhances the contrast of the image by performing the following steps: Image denoising: Assume the original chest CT image is I raw , using median filtering I raw Perform denoising to obtain a denoised image; Contrast enhancement: Logarithmic transformation is performed on the grayscale values ​​of the denoised image to expand the grayscale values ​​of the darker areas of the image and compress the grayscale values ​​of the brighter areas. The chest CT image after denoising and contrast enhancement is , H is I The height of W is I The width of C is I The number of channels.

3. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 1 is characterized in that: The image segmentation module performs the following steps: Input: ,Process: Use DenseNet network for image segmentation; Output: The category to which each pixel belongs includes tumor, blood vessel, trachea, bone, and background.

4. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 3 is characterized in that: The DenseNet network consists of an initial convolutional layer, multiple dense blocks, and a fully connected layer. There is a transition layer between the two dense blocks. The specific steps are as follows: 1) Initial convolutional layer use α The size is k 1× k 1 convolution kernel for feature extraction with a step size of s 1. Get the local feature map F 1, and use ReLU The activation function introduces nonlinear relationships; 2) Dense blocks use β dense blocks, each of which consists of δ The convolutional layers are composed of, and the output of each convolutional layer is directly connected to all subsequent layers; in each dense block, the convolutional layer uses k 2× k The convolution kernel of 2 is padded with 1 to keep the feature map size unchanged; have to β Feature maps ; 3) Transition layer The transition layer is located between two dense blocks The transition layer consists of a 1×1 convolutional layer to reduce the number of channels, followed by a p 1× p 1 average pooling layer with a step size of s 2. Compress the feature map size to obtain the feature map F 3; 4) Fully connected layer and classifier Using a fully connected layer for chest CT images The segmentation tasks of tumors, blood vessels, trachea, and bones are performed using softmax The function calculates the category probability of each pixel: （1） in It's a pixel i The probability of being classified into the γth class, is the number of categories, The value is {0 , 1 , 2 , 3 , 4}, representing background, tumor, blood vessel, trachea and bone respectively, s γ is of category γ logit value, s i For category i of logit Value; select the category with the highest probability as the pixel i The final category : （2） in Represents pixel points i Belongs to background, tumors, blood vessels, trachea and bones; 5) Loss Function Due to the uneven distribution of background, tumor, blood vessel, trachea and bone categories, the module uses cross entropy loss during training, which is defined as follows: （3） in It's a pixel i The true label of the corresponding category γ, 1 means that the pixel belongs to this category, and 0 means that it does not belong to this category; N is the number of pixels.

5. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 1 is characterized in that: The path planning module performs the following steps: Input: Image segmentation result ;Process: Use dynamic adaptive ant colony optimization algorithm to select the best path or the approximate global best path; Output: optimal path or approximate global optimal path.

6. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 5 is characterized in that: The dynamic adaptive ant colony optimization algorithm dynamically adjusts the number of ants and the pheromone update strategy according to the path exploration progress. The dynamic adaptive ant colony optimization algorithm limits the path selection to a straight line path from the starting point to the end point. The equation of the straight line from the starting point to the end point is: For each ant, from the starting point to the end point P start ( x 0 ,y 0 ,z 0) To the end P end ( x n ,y n ,z n ) is expressed by the following formula: （4） The parameters t Indicates different positions on the straight line from the starting point to the end point; when t =0, ( x ( t ) ,y ( t ) ,z ( t ))= P start ( x 0 ,y 0 ,z 0); when t =1, ( x ( t ) ,y ( t ) ,z ( t ))= P end ( x n ,y n ,z n ).

7. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 6 is characterized in that: The steps of dynamic adaptive ant colony optimization algorithm are: 1) Initialization Set the start and end points: Set the start point P start ( x 0 ,y 0 ,z 0) and the end point P end ( x n ,y n ,z n ), and the end point P end ( x n ,y n , z n ) belongs to the tumor category only; Initialize the number of ant colonies to m ; Initialize pheromone: the pheromone of all paths is initialized to be equal; 2) Ant path selection strategy In each iteration, the ant selects a path based on the straight line equation between the current position and the target point, ensuring that the path is along a straight line in three-dimensional space; the path selection is based on the pheromone concentration. and heuristic information To decide, ants choose from nodes i To Node j The probability of the path is: （5） in It is a slave node i To Node j The pheromone intensity reflects the path ij the importance of Is the path ij The heuristic information of , the calculation process is shown in step 4), and when ants choose a path, they tend to choose a path with greater heuristic information; λ and μ are the weight coefficients of pheromone and heuristic information, respectively, which are used to balance the influence of pheromone on path selection and the role of heuristic information; k is the set of next-step nodes that the ant can choose; 3) Pheromone Update The rules for pheromone updates are as follows: （6） in ρ It is the pheromone evaporation factor, which is used to control the decay rate of pheromones; Represents the pheromone update amount, , d ij Is the path ij length; Q is a constant used to control the pheromone increment; 4) Calculation of heuristic information: （7） in d ij Representation node i To Node j The Euclidean distance of f ( ij ) is the obstacle function; （8） （9）； 5) Termination After multiple iterations or reaching the set maximum number of iterations, it will eventually converge to the global optimal or near-optimal path, that is, adopt P ij The largest path ij .

8. The robotic arm thoracentesis path planning system based on dynamic adaptive ant colony optimization algorithm according to claim 1 is characterized in that: The visualization module displays the searched optimal path or approximate global optimal path on the original chest CT image. I raw Mark on.

9. A method for planning a path for a thoracotomy with a robotic arm based on a dynamic adaptive ant colony optimization algorithm, which executes the steps in the module of the system for planning a path for a thoracotomy with a robotic arm based on a dynamic adaptive ant colony optimization algorithm according to any one of claims 1 to 8 based on chest CT images.

Images