Self-propelled delivery method and device for vascular intervention guide wire

Abstract

The application provides a vascular interventional operation guide wire autonomous conveying method and device, wherein the method comprises the following steps: acquiring a real-time image of a guide wire in an interventional operation, and performing segmentation processing on the image through a guide wire segmentation algorithm based on deep learning to obtain a mask of the guide wire; performing shape analysis on the mask through a guide wire shape analysis algorithm based on a B-spline curve to obtain state quantities of the guide wire; based on the state quantities of the guide wire and a navigation target position, performing decision-making on guide wire conveying control through a decision-making algorithm based on reinforcement learning to obtain control quantities of guide wire conveying; and controlling the motion state of the guide wire according to the control quantities, conveying the guide wire to the target position, and realizing autonomous conveying of the guide wire.

Description

Technical Field

[0001] This application relates to the fields of devices for vascular interventional surgery and artificial intelligence technology, and in particular to a method and device for autonomous delivery of guidewires in vascular interventional surgery. Background Technology

[0002] Interventional vascular surgery, as a minimally invasive treatment technique, is widely used in the diagnosis and treatment of cardiovascular diseases. In this procedure, the guidewire is a crucial tool, primarily used for navigation within the blood vessel and guiding other instruments to the lesion site. However, current guidewire delivery methods mainly rely on manual operation by the surgeon, which demands a high level of experience and skill, making the procedure difficult and time-consuming. Especially when dealing with complex vascular structures, manual operation struggles to guarantee the accuracy and stability of guidewire delivery, and surgeons face a high risk of radiation exposure during the procedure. Therefore, there is an urgent need to design a method that can automatically deliver the guidewire to the target location to effectively solve this problem.

[0003] Regarding the acquisition of guidewire state variables, due to the guidewire's slender shape and sparse features, it is easily confused with the vessel edge. Existing methods mostly rely on guidewire tip position identification, neglecting the extraction of overall guidewire morphological features (such as curvature and orientation angle). However, these morphological features play a crucial role in estimating the guidewire's mechanical state and predicting its motion trend. In terms of guidewire autonomous navigation control, current methods largely depend on pre-programmed path planning or decision logic based on fixed rules, using classical control methods to control the guidewire. However, the interaction between the guidewire and the vessel has complex nonlinear dynamic characteristics, which classical control methods cannot accurately describe using linear models, resulting in insufficient control robustness. Summary of the Invention

[0004] This application provides a method and device for autonomous delivery of guidewires in vascular interventional surgery to solve one of the technical problems in related technologies.

[0005] The first aspect of this application provides a method for autonomous delivery of a guidewire in vascular interventional surgery, comprising:

[0006] Real-time images of the guidewire during interventional surgery are acquired, and the images are segmented using a deep learning-based guidewire segmentation algorithm to obtain a guidewire mask.

[0007] The mask is morphologically analyzed using a guidewire morphology analysis algorithm based on B-spline curves to obtain the state variables of the guidewire.

[0008] Based on the state variables of the guidewire and the position of the navigation target, a decision-making algorithm based on reinforcement learning is used to make decisions on the guidewire delivery control, thereby obtaining the control variables for guidewire delivery.

[0009] The movement state of the guidewire is controlled according to the control quantity, and the guidewire is delivered to the target position.

[0010] A second aspect of this application provides an autonomous guidewire delivery device for vascular interventional surgery, comprising:

[0011] The main control unit is used to acquire real-time images of the guidewire during interventional surgery and to segment the images using a deep learning-based guidewire segmentation algorithm to obtain a mask for the guidewire.

[0012] The main control unit is also used to perform morphological analysis on the mask using a guide wire morphology analysis algorithm based on B-spline curves to obtain the state variables of the guide wire;

[0013] The main control unit is also used to make decisions on the guide wire delivery control based on the state variables of the guide wire and the position of the navigation target, and obtain the control variables for guide wire delivery by using a decision algorithm based on reinforcement learning.

[0014] A guide wire delivery device is used to control the movement state of the guide wire according to the control quantity and deliver the guide wire to the target position.

[0015] The method and apparatus for autonomous guidewire delivery in vascular interventional surgery provided in this application extracts the guidewire mask from the interventional surgical image using a deep learning-based image segmentation algorithm, and extracts the guidewire state variables from the mask using a B-spline curve-based guidewire morphology analysis algorithm. A reinforcement learning-based decision-making algorithm determines the control variable for the next step of the guidewire based on the guidewire state variables. The guidewire delivery device responds to the control variable and delivers the guidewire to the target position. This scheme combines a deep learning-based image segmentation algorithm and a B-spline curve-based guidewire morphology analysis algorithm to obtain the guidewire state variables, considering the overall morphological characteristics of the guidewire, thus obtaining more accurate guidewire state variables. The reinforcement learning-based decision-making algorithm for determining the next step of the guidewire control variable considers the complex nonlinear dynamic characteristics of the interaction between the guidewire and the blood vessel, improving the robustness of the control variable. This scheme overcomes the limitations of traditional control methods in guidewire delivery tasks and provides an effective method for autonomous guidewire delivery in vascular interventional surgery.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0018] Figure 1A schematic flowchart illustrating a method for autonomous delivery of guidewires in vascular interventional surgery, provided in an embodiment of this application;

[0019] Figure 2 This is a model framework diagram of the guide wire image segmentation algorithm according to an embodiment of the present invention.

[0020] Figure 3 This is a flowchart of the filament morphology analysis algorithm in an embodiment of the present invention.

[0021] Figure 4 This is a framework diagram of the reinforcement learning-based decision-making algorithm in this embodiment of the invention.

[0022] Figure 5 This is an overall frame diagram of the autonomous guidewire delivery device for vascular interventional surgery, as exemplified by the present invention.

[0023] Figure 6 This is a flowchart illustrating the implementation steps of an example of the present invention.

[0024] Figure 7 This is a block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0026] The following description, with reference to the accompanying drawings, describes an embodiment of a method, apparatus, and device for autonomous delivery of guidewires in vascular interventional surgery.

[0027] Figure 1 This is a schematic flowchart illustrating a method for autonomous guidewire delivery in vascular interventional surgery, provided in an embodiment of this application. Figure 1 As shown, the autonomous guidewire delivery method for vascular interventional surgery includes the following steps:

[0028] Step S101: Obtain a real-time image of the guidewire during the interventional procedure, and segment the image using a deep learning-based guidewire segmentation algorithm to obtain a mask for the guidewire.

[0029] In this embodiment, a deep learning-based guidewire segmentation algorithm is obtained by training a deep learning network, such as... Figure 2As shown, this deep learning network adopts an encoder-decoder architecture. The encoder contains four downsampling modules, each consisting of two cascaded convolutional units. The number of channels between the four downsampling modules increases exponentially. The encoder's end, the fourth-level downsampling, performs downsampling through max pooling. The decoder contains four upsampling modules, each improving the feature map resolution through nearest-neighbor interpolation. The number of channels between the four upsampling modules decreases exponentially. A grouped dual-axis attention module is embedded in the skip connections between the four downsampling and upsampling modules to enhance the guidewire edge features.

[0030] In this embodiment, the grouped dual-axis attention module divides the input feature map into four channels, and performs adaptive average pooling and max pooling on the height and width axes for the features of each channel; using the Sigmoid function and Convolution generates dual-axis attention weights; the attention weights generated by each channel group are then concatenated and fused with the original features after element-wise weighting.

[0031] This step employs an encoder-decoder architecture for wire segmentation, which can accurately extract the mask information of the wire from the image.

[0032] As an example, the encoder contains four levels of downsampling, each consisting of two cascaded convolutional units. Each convolutional unit includes a convolutional layer (3×3 kernel, stride 1), a batch normalization layer, and a ReLU activation layer. The number of channels increases progressively between the four downsampling levels (64→78→256→57). The encoder ends with downsampling via max pooling (2×2 pooling kernel, stride 2).

[0033] The decoder contains four levels of upsampling. Each level of upsampling doubles the resolution of the feature map through nearest neighbor interpolation, and then connects to a convolutional layer (3×3 kernel, stride 1) for feature reconstruction. The number of channels decreases progressively between the four levels of upsampling (57→256→78→64).

[0034] A grouped dual-axis attention module is embedded in the skip connection between the fourth-level downsampling and the fourth-level upsampling. The feature map is divided into four groups according to the number of channels. Adaptive average pooling and max pooling are performed on each group of features along the height axis and the width axis, respectively, to generate dual-axis attention weights. The calculation formula is as follows:

[0035]

[0036] in, For input features, This represents the Sigmoid function. and for Convolutional layer.

[0037] The generated dual-axis attention weights are used to element-wise weight the original input feature vector, thereby enhancing the features of key regions. This operation can be represented as:

[0038]

[0039] in, This represents element-wise multiplication, and ultimately enhances the guidewire edge features through channel splicing and weighted fusion.

[0040] Step S102: The mask is morphologically analyzed using a guidewire morphology analysis algorithm based on B-spline curves to obtain the state variables of the guidewire.

[0041] As one implementation method, the method of extracting the state variables of the guidewire through guidewire morphology analysis algorithm, such as... Figure 3 As shown, it includes the following steps:

[0042] Step 201: Refine the binarized mask to obtain the center skeleton line point set with a single pixel width.

[0043] Step 202: Sort the central skeleton line point set according to the connected paths, and use the cumulative arc length parameterization method to parameterize the sorted central skeleton line point set to obtain the parameterized central skeleton line point set.

[0044] Step 203: By fitting the parameterized central skeleton line point set with a B-spline curve, the fitted B-spline curve formula is obtained.

[0045] Step 204: Perform mathematical processing on the fitted B-spline curve formula to obtain the state variables of the guidewire; the state variables of the guidewire include the position of the guidewire tip, the direction angle of the guidewire tip, the curvature distribution of the guidewire, and the length of the guidewire.

[0046] One implementation method involves fitting a parameterized set of central skeleton line points to a B-spline curve to obtain the fitted B-spline curve formula. This includes: constructing a third-order B-spline curve using the Cox-de-Boor recursive formula; constructing a least-squares problem to determine the optimal positions of the control points in the B-spline curve; and fitting the B-spline curve to the parameterized set of central skeleton line points to obtain the fitted B-spline curve formula.

[0047] As one implementation method, the guidewire tip position is determined by the value of the B-spline curve at the tip parameter, the guidewire tip direction angle is determined by the tangent vector at the tip of the B-spline curve, the guidewire curvature distribution is obtained by differentiating the B-spline curve, and the guidewire length is obtained by integrating the first derivative of the B-spline curve over the preset parameter range.

[0048] In this embodiment, the guidewire skeleton is extracted from the mask using a guidewire morphology analysis algorithm based on B-spline curves, and the morphological parameters of the guidewire, i.e., the state variables of the guidewire, are obtained by fitting a third-order B-spline curve.

[0049] As an example, the method for obtaining the state variables of the guidewire includes the following steps:

[0050] Step 1: Refine the binarized mask to obtain the center skeleton line point set with a single pixel width.

[0051] Step 2, set the central skeleton line points Sort by connected paths and assign parameter values ​​to each point using the cumulative arc length parameterization method:

[0052]

[0053] By using parameterization, we can ensure that the parameter difference between adjacent sampling points is consistent with their true spatial distance, thus avoiding fitting distortion caused by uniform parameterization in sharply curved regions.

[0054] Step 3: To construct the B-spline, select the spline order. and in the parameter range The node vector is generated using the average node method:

[0055]

[0056] After obtaining the node vectors, B-spline basis functions are constructed using the Cox-de-Boor recursive formula. The recursive definition of the basis functions is:

[0057]

[0058] Therefore, the B-spline curve can be represented as:

[0059]

[0060] in, For the first One control point, To determine the number of control points.

[0061] To determine the optimal control point locations, a least-squares problem is constructed to minimize the sum of the squared distances from all sampling points to the curve:

[0062]

[0063] Let matrix Each behavior in the parameters The values ​​of each basis function are given, and the skeleton pixel coordinates form a vector. Then a system of linear equations can be obtained. The least squares problem has the following normal equation:

[0064]

[0065] To each and Solving the above normal equation using the coordinate components yields a smooth and continuously differentiable B-spline curve.

[0066] Step 4: After skeletonization and B-spline curve fitting, the discrete pixel representation of the guidewire in the image is converted into a continuously differentiable function. .

[0067] The guidewire tip position can be determined by adjusting the tip parameters using a B-spline curve. The value at that location determines:

[0068]

[0069] coordinates of this location It directly reflects the specific position of the guidewire tip in the image, which is crucial for accurate positioning and path planning.

[0070] The direction of the guidewire tip can be determined by the tangent vector at the tip of the curve:

[0071]

[0072] Due to basis functions If the vector is differentiable, normalizing the tangent vector will yield the tip direction vector. Furthermore, the direction angle of the guidewire tip is obtained through the following formula, thereby achieving a quantitative description of the guidewire orientation:

[0073]

[0074] This is especially critical for intravascular navigation, as the orientation of the guidewire tip often determines the next insertion direction and potential branch selection.

[0075] The curvature characteristics of a guidewire reflect its degree of tortuosity within a vascular passage. According to the curvature formula for a plane curve, for a parametric curve... Its curvature can be written as:

[0076]

[0077] The higher-order continuity of the B-spline basis functions guarantees the stability of the calculation of the first and second derivatives, thus enabling the relatively accurate acquisition of the curvature distribution of the curve at various points. Analyzing this curvature distribution further yields the following indices:

[0078] Maximum curvature By searching across the entire parameter range The maximum value is used to determine the most curved area of ​​the guidewire.

[0079] Mean curvature near the tip Select the parameter range near the tip of the guidewire. Calculate the mean curvature:

[0080]

[0081] The mean curvature is used to assess whether there is a risk of excessive contact force at the guidewire tip due to excessive bending.

[0082] Guide wire length can be determined by adjusting the curve within the parameter range. Integrating the first derivative over the given information yields:

[0083]

[0084] The above formula can be approximated using numerical integration. Length information can be used to determine the depth of guidewire insertion or retraction, thereby enabling more accurate operational decisions.

[0085] In summary, the integrity parameters of the guidewire extracted from the mask are:

[0086]

[0087] Step S103: Based on the state variables of the guidewire and the position of the navigation target, a decision is made on the guidewire delivery control using a reinforcement learning-based decision algorithm to obtain the control variable for guidewire delivery.

[0088] In this embodiment, the reinforcement learning-based decision-making algorithm includes the definition of a state space, an action space, and a reward function. The state space is a six-dimensional state information vector, including the x and y coordinates of the guidewire tip position, the guidewire tip orientation angle, the maximum curvature of the guidewire, the average curvature near the guidewire tip, and the guidewire length. The action space consists of two degrees of freedom: axial motion and rotational motion during guidewire delivery. The reward function is a composite function, including a distance penalty term, a morphological penalty term, and a task settlement reward.

[0089] In this embodiment, the reinforcement learning-based decision-making algorithm is obtained by training a decision model including a policy network and a value network using the PPO algorithm. The framework of this decision model is as follows: Figure 3 As shown.

[0090] As an example, state space The six-dimensional state information vector is defined as the six-dimensional state variables extracted from the mask of the guidewire.

[0091]

[0092] The motion space is set to two degrees of freedom: axial motion and rotational motion of the guidewire.

[0093]

[0094] The reward function is set as a composite function, consisting of three parts: a distance penalty, a shape penalty, and a task settlement reward. The distance penalty... for:

[0095]

[0096] in, The Euclidean distance representing the guidewire tip and the target position. An exponential reward representing distance. This represents the weighting coefficient.

[0097] Morphological penalty item for:

[0098]

[0099] in, This represents the ideal length of the guidewire. This represents the maximum curvature of the guidewire. This represents the average curvature of the guidewire tip. This represents the weighting coefficient.

[0100] Task settlement rewards for:

[0101]

[0102] A positive reward of 50 is given upon successful completion of the task; this reward is 0 while the task is in progress; and a negative penalty of -50 is given upon task failure. The further compound reward function is set as follows:

[0103]

[0104] Based on the definitions of state space, action space and reward function above, the reinforcement learning PPO algorithm is used for training to obtain a decision algorithm model for autonomous guide wire delivery. It consists of a policy network (Actor) and a value network (Critc), and can output the control quantity of the guide wire according to the state quantity of the guide wire and the position of the navigation target.

[0105] The training process for the decision algorithm model is described below:

[0106] First, the algorithm interacts with the environment to collect a set of data. Interaction trajectory, in which It's a state. It is in state The following actions were taken. It is the reward received. The next state involves storing these trajectories in an experience pool and updating the network in small batches.

[0107] In each round of training, the algorithm selects actions based on the current policy network (Actor). That is, from a given state Sampling from the action distribution below is represented as:

[0108]

[0109] in, Representative strategy The network parameters are then used. After action sampling, the algorithm interacts with the environment and obtains rewards, which are then used to update the policy.

[0110] To improve training efficiency, the algorithm introduces an advantage function to estimate the advantage of a given action relative to the current policy. Advantage function The calculation relies on the Generalized Advantage Estimation (GAE) method, which estimates advantage based on experience over multiple time steps. The calculation formula is as follows:

[0111]

[0112] in, It is a time-difference, It is a discount factor. It is a hyperparameter of GAE, which controls the trade-off between bias and variance.

[0113] The objective function of the decision algorithm is shown below:

[0114]

[0115] in, Represents mathematical expectation, It is the ratio of the current strategy to the old strategy. Used to control the shearing range.

[0116] Value Network It will also be updated. The loss function of the value network is the mean squared error loss, and the calculation formula is:

[0117]

[0118] in, Represents mathematical expectation, Representing state The cumulative expected return is calculated using the time-difference method.

[0119] Step S104: Control the movement state of the guidewire according to the control quantity, and deliver the guidewire to the target position.

[0120] In this embodiment, the guide wire delivery device responds to the control quantity to control the guide wire advance and rotation, and delivers the guide wire to the target position; wherein, the guide wire delivery device obtains the control quantity through a Restful API interface built with the Fast API framework.

[0121] As one implementation method, the guide wire image segmentation algorithm, guide wire morphology analysis algorithm, and reinforcement learning-based decision-making algorithm are integrated with the guide wire delivery device using the Fast API framework to build a RESTful API interface, enabling communication between algorithms and between algorithms and devices, and responding in real time to the control quantities issued by the decision-making algorithm.

[0122] This application's embodiment of the autonomous guidewire delivery method for vascular interventional surgery extracts the guidewire mask from the interventional surgical image using a deep learning-based image segmentation algorithm, and extracts the guidewire state variables from the mask using a B-spline curve-based guidewire morphology analysis algorithm. A reinforcement learning-based decision-making algorithm determines the next control variable for the guidewire based on these state variables. The guidewire delivery device then responds to the control variable and delivers the guidewire to the target position. This scheme combines a deep learning-based image segmentation algorithm and a B-spline curve-based guidewire morphology analysis algorithm to obtain the guidewire state variables, considering the overall morphological characteristics of the guidewire and thus obtaining more accurate state variables. The reinforcement learning-based decision-making algorithm considers the complex nonlinear dynamics of the interaction between the guidewire and the blood vessel, improving the robustness of the control variable. This scheme overcomes the limitations of traditional control methods in guidewire delivery tasks, providing an effective method for autonomous guidewire delivery in vascular interventional surgery.

[0123] This application also provides an autonomous guidewire delivery device for vascular interventional surgery, comprising:

[0124] The main control unit is used to acquire real-time images of the guidewire during interventional surgery and to segment the images using a deep learning-based guidewire segmentation algorithm to obtain a mask for the guidewire.

[0125] The main control unit is also used to perform morphological analysis on the mask using a guide wire morphology analysis algorithm based on B-spline curves to obtain the state variables of the guide wire;

[0126] The main control unit is also used to make decisions on the guide wire delivery control based on the guide wire state variables and the navigation target position, and obtain the control variables for guide wire delivery through a reinforcement learning-based decision algorithm.

[0127] The guide wire feeding device is used to control the movement state of the guide wire according to the control quantity and to feed the guide wire to the target position.

[0128] It should be noted that the foregoing explanation of the autonomous guidewire delivery method for vascular interventional surgery also applies to the autonomous guidewire delivery device for vascular interventional surgery in this embodiment, and will not be repeated here.

[0129] To clearly illustrate the above embodiments, specific examples will now be used for explanation.

[0130] The specific object in this example is a human blood vessel simulation model. Figure 4 This diagram illustrates the overall framework of the autonomous guidewire delivery device for vascular interventional surgery in this example. An industrial camera connects to a GPU server via a GigE interface to capture real-time images of the guidewire within a human vascular simulation model. The GPU server acts as the main control center, running guidewire image segmentation algorithms, guidewire morphology analysis algorithms, and reinforcement learning-based decision-making algorithms. These algorithms communicate using a RESTful API interface. The guidewire delivery device responds to control requests through this RESTful API interface, receiving control signals from the decision-making algorithms to control the guidewire's movement within the human vascular simulation model.

[0131] The autonomous guidewire delivery method for vascular interventional surgery in this example includes the following specific operational steps:

[0132] Step 1: Train the guide wire image segmentation algorithm using a manually pre-annotated guide wire segmentation dataset.

[0133] The guidewire image segmentation algorithm in this example is written using the PyToch framework, and its model framework is as follows: Figure 2 As shown, the Dice loss function was used, the Adam optimizer was used, and the training was performed on an NVIDIA 4090 GPU with a learning rate of 0.001, a batch size of 16, and a training duration of 200 epochs.

[0134] Step 2: Use the trained image segmentation algorithm and guidewire morphology analysis algorithm to extract guidewire state variables, and train a reinforcement learning-based decision-making algorithm in the vascular interventional surgery simulation model.

[0135] The guidewire morphology analysis algorithm in this example is written in Python, and its algorithm flow is as follows: Figure 5 As shown in the figure. The B-spline curve fitting order is 3, and the smoothing coefficient is set to 10.

[0136] The reinforcement learning-based decision-making algorithm in this example is implemented with reference to the PPO algorithm, and its algorithm framework is as follows: Figure 3 As shown, in the actual training process, the state space, action space, and reward function are specifically defined as follows:

[0137] (1) State space

[0138] An industrial camera was used to capture an image of a human blood vessel simulation model from directly above, measuring [size missing]. The guide wire image is used to obtain the guide wire mask by calling the guide wire image segmentation algorithm deployed on the GPU server, and then the guide wire morphology analysis algorithm is used to obtain the state variables of the guide wire.

[0139] (2) Action space

[0140] The guide wire feeding device can achieve axial displacement control and rotation control of the guide wire. The range of axial displacement of the guide wire is... The range of rotational motion is .

[0141] (3) Reward function

[0142] The reward function is implemented according to the expression in the above embodiments of this application, wherein the weight coefficients are respectively set as follows: , , , .

[0143] Based on the detailed definition of the reinforcement learning-based decision-making algorithm above, the algorithm is written using the PyTorch framework, with the learning rate set to 0.001, the discount factor set to 0.97, the SGD optimizer used, 78 steps of data collected per batch, the policy updated 10 times, and the training performed for 20,000 time steps.

[0144] Step 3: Deploy the trained algorithm model on the GPU server to verify the autonomous guidewire delivery method.

[0145] This example deploys pre-trained guidewire image segmentation, guidewire morphology analysis, and reinforcement learning-based decision-making algorithm models on a GPU server to verify the autonomous guidewire delivery method. The verification is performed on the left subclavian artery branch of the LSA on a human vascular simulation model, with the guidewire starting area directly below the LSA branch.

[0146] The following is for reference. Figure 6 Explain the overall process of the autonomous guidewire delivery method:

[0147] (1) Initialize the wire guide and feeder device, and the industrial camera;

[0148] (2) Deploy the guide wire image segmentation algorithm, guide wire morphology analysis algorithm and reinforcement learning-based decision algorithm on the GPU server, and use the FastAPI framework to build a Restful API interface for the above three algorithms and the guide wire delivery device to realize mutual calls between the algorithms;

[0149] (3) Start the guidewire initialization program and push the guidewire to the starting position;

[0150] (4) Call the guidewire image segmentation algorithm and guidewire morphology analysis algorithm to obtain the state variables of the guidewire;

[0151] (5) The decision-making algorithm based on reinforcement learning outputs the guide wire control quantity according to the state quantity;

[0152] (6) The guide wire feeding device responds to the control quantity and feeds the guide wire;

[0153] (7) The reinforcement learning-based decision algorithm determines whether the guide wire tip has reached the target area. If not, return to step (4) to continue execution. If it has reached the target area, the program ends.

[0154] To implement the above embodiments, this application also proposes an electronic device. Please see [link to relevant documentation]. Figure 7 , Figure 7 This is a block diagram of the electronic device provided in an embodiment of this application. For example... Figure 7 As shown, the electronic device 700 includes: a processor 701 and a memory 702 communicatively connected to the processor 701; the memory 702 stores computer execution instructions; the processor 701 executes the computer execution instructions stored in the memory to implement the method provided in the foregoing embodiments.

[0155] To implement the above embodiments, this application also proposes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided in the foregoing embodiments.

[0156] To implement the above embodiments, this application also proposes a computer program product, including a computer program that, when executed by a processor, implements the methods provided in the foregoing embodiments.

[0157] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0158] It should be noted that personal information collected from users should be used for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. Furthermore, such collection / sharing should only be conducted after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization that includes authorization of relevant user information before the user uses the function. In addition, any necessary steps must be taken to protect and safeguard access to such personal information data and ensure that others with access to personal information data comply with their privacy policies and procedures.

[0159] This application is intended to provide an implementation scheme for users to selectively prevent the use or access to their personal information data. Specifically, this disclosure is intended to provide hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, risks can be minimized by restricting data collection and deleting data. Furthermore, where applicable, such personal information is de-identified to protect user privacy.

[0160] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0161] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0162] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A self-delivery device for guidewires in vascular interventional surgery, characterized in that, include: The main control unit is used to acquire real-time images of the guidewire during interventional surgery and to segment the images using a deep learning-based guidewire segmentation algorithm to obtain a mask for the guidewire. The main control unit is also used to perform morphological analysis on the mask using a guide wire morphology analysis algorithm based on B-spline curves to obtain the state variables of the guide wire; The main control unit is further configured to make decisions on the guidewire delivery control based on the state variables of the guidewire and the position of the navigation target, using a reinforcement learning-based decision algorithm, to obtain the control variables for guidewire delivery. The reinforcement learning-based decision algorithm includes the definitions of a state space, an action space, and a reward function. The state space is a six-dimensional state information vector, including the x and y coordinates of the guidewire tip position, the guidewire tip orientation angle, the maximum curvature of the guidewire, the average curvature near the guidewire tip, and the guidewire length. The action space consists of two degrees of freedom: axial motion and rotational motion during guidewire delivery. The reward function is a composite function, including a distance penalty term, a shape penalty term, and a task settlement reward. A guide wire delivery device is used to control the movement state of the guide wire according to the control quantity, and to deliver the guide wire to the target position; The main control unit is also used for: The binarized mask is thinned to obtain the center skeleton line point set with a single pixel width; The central skeleton line point set is sorted according to the connected paths, and the sorted central skeleton line point set is parameterized using the cumulative arc length parameterization method to obtain the parameterized central skeleton line point set. The formula for the fitted B-spline curve is obtained by fitting the parameterized central skeleton line point set with a B-spline curve. The fitted B-spline curve formula is mathematically processed to obtain the state variables of the guidewire; the state variables of the guidewire include the guidewire tip position, the guidewire tip direction angle, the guidewire curvature distribution, and the guidewire length.

2. The apparatus according to claim 1, characterized in that, The deep learning-based guide wire segmentation algorithm is obtained by training a deep learning network. The deep learning network adopts an encoder-decoder architecture. The encoder contains four downsampling modules, each consisting of two cascaded convolutional units. The number of channels between the four downsampling modules increases exponentially. The encoder end performs downsampling through max pooling. The decoder contains four upsampling modules, each improving the feature map resolution through nearest neighbor interpolation. The number of channels between the four upsampling modules decreases exponentially. A grouped dual-axis attention module is embedded in the skip connections between the four downsampling and upsampling modules.

3. The apparatus according to claim 2, characterized in that, The grouped dual-axis attention module is used to divide the input feature map into four channels, and performs adaptive average pooling and max pooling on the height and width axes for the features of each channel; through the Sigmoid function and Convolution generates dual-axis attention weights; the attention weights generated by each channel group are then concatenated and fused with the original features after element-wise weighting.

4. The apparatus according to claim 1, characterized in that, The main control unit is also used for: Constructing a third-order B-spline curve using the Cox-de-Boor recursive formula; Construct a least squares problem to determine the optimal location of control points in a B-spline curve; The B-spline curve is fitted to the parameterized central skeleton line point set to obtain the fitted B-spline curve formula.

5. The apparatus according to claim 1, characterized in that, The position of the guidewire tip is determined by the value of the B-spline curve at the tip parameter, the direction angle of the guidewire tip is determined by the tangent vector of the tip of the B-spline curve, the curvature distribution of the guidewire is obtained by differentiating the B-spline curve, and the length of the guidewire is obtained by integrating the first derivative of the B-spline curve over a preset parameter range.

6. The apparatus according to claim 4, characterized in that, The reinforcement learning-based decision-making algorithm is obtained by training a decision model including a policy network and a value network using the PPO algorithm.

7. The apparatus according to claim 1, characterized in that, The guide wire feeding device is also used for: The guide wire delivery device responds to the control quantity, controls the guide wire to advance and rotate, and delivers the guide wire to the target position; the guide wire delivery device obtains the control quantity through a Restful API interface built using the Fast API framework.

Images