A method, apparatus, device, medium, and product for implant member placement planning

Abstract

The present disclosure relates to a method, device, equipment, medium and product for planning the arrangement of an implant member, the method comprising: building a reinforcement learning agent for a fracture site, taking three-dimensional image data of the fracture site as input data, defining tool calling actions in the whole arrangement planning process from the image preprocessing link, the arrangement planning link to the verification feedback link as an action space, recording tool calling in the current decision round to update the state space, and through a decision model, according to the state space and the action space, sequentially executing each planning link to output verification feedback data of the current decision round; using a planning reward mechanism to optimize the parameters of the decision model for the next decision round until the verification feedback data passes the test, obtaining a target arrangement scheme for the fracture site, solving the problem that traditional arrangement planning relies on artificial experience, and realizing a repeatable, interpretable and verifiable arrangement planning system.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, medium and product for planning the placement of implanted components. Background Technology

[0002] The human skeletal system plays a crucial role in providing mechanical support and force transmission. When bones are subjected to high-energy external impacts, they are prone to fractures and injuries. Taking the pelvis as an example, as an important mechanical transmission structure connecting the trunk and lower limbs, it is composed of various bony tissues such as the ilium, ischium, pubis, sacrum, sacroiliac joint, and acetabulum. It is also surrounded by important anatomical structures such as the internal and external iliac vessels, sacral nerve plexus, obturator neurovascular bundle, and urogenital tissues. Pelvic fractures are often accompanied by bleeding, bodily stress response, multiple injuries, and significant displacement of bone structures.

[0003] Internal fixation implants are the main means of achieving fracture reduction and reconstructing skeletal mechanical support. However, complex bone structures such as the pelvis, which have concealed operating paths, limited visibility, small working areas, and low tolerance for deviations in the placement path, pose greater challenges to the placement planning of implants.

[0004] Currently, the planning and deployment of implantable components relies heavily on personnel experience, which is not only highly subjective and lacks repeatability, but also time-consuming due to the need for repeated comparisons of image data and past case studies, leading to analytical fatigue. More importantly, given the complex and varied individual anatomical differences, it is difficult to establish a standardized, verifiable, and interpretable deployment planning system. Summary of the Invention

[0005] This disclosure provides a method, apparatus, equipment, medium, and product for planning the placement of implantable components, thereby solving the problem that traditional placement planning relies on human experience and realizing a placement planning system that is repeatable, interpretable, and verifiable.

[0006] One aspect of this disclosure provides a method for planning the placement of implantable components, including: The three-dimensional image data of the fracture site is input into the reinforcement learning agent; wherein, the reinforcement learning agent is built based on a decision model, a tool library and a deployment knowledge base, the tool library includes a set of tools corresponding one-to-one with the planning steps in the deployment planning process, and the deployment planning process includes an image preprocessing step, a deployment planning step for implanted components and a verification feedback step. The reinforcement learning agent updates its state space based on the deployed knowledge base, the 3D image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and uses the tool call actions corresponding to the tool base as the action space; wherein, the tool call records include tool call actions and the tool output data corresponding to the tool call actions; Based on the decision model, the planning steps in the deployment planning process are executed sequentially according to the state space and the action space, and the verification feedback data of the current decision round is output. If the verification feedback data shows that the decision fails, the planning reward mechanism corresponding to the reinforcement learning agent is adopted to determine the planning reward value of the current decision round. The decision model is then optimized based on the planning reward value, and the optimized decision model is used as the decision model for the next decision round. Until the verification feedback data shows a pass, the placement scheme of the implanted component output in the placement planning stage is taken as the target placement scheme for the fracture site.

[0007] Another aspect of this disclosure provides a placement planning device for implantable components, comprising: A three-dimensional image data input module is used to input three-dimensional image data of the fracture site into a reinforcement learning agent; wherein, the reinforcement learning agent is built based on a decision model, a tool library and a deployment knowledge base, the tool library includes a set of tools corresponding one-to-one with the planning stages in the deployment planning process, and the deployment planning process includes an image preprocessing stage, a deployment planning stage for implanted components and a verification feedback stage. The state space update module is used by the reinforcement learning agent to update the state space based on the deployed knowledge base, the 3D image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and to use the tool call actions corresponding to the tool base as the action space; wherein, the tool call records include tool call actions and the tool output data corresponding to the tool call actions; The verification feedback data output module is used to execute the planning steps in the deployment planning process sequentially according to the state space and the action space through the decision model and output the verification feedback data of the current decision round. The decision model optimization module is used to determine the planning reward value of the current decision round by adopting the planning reward mechanism corresponding to the reinforcement learning agent when the verification feedback data shows that it fails, optimize the parameters of the decision model according to the planning reward value, and use the optimized decision model as the decision model for the next decision round. The target deployment scheme determination module is used to take the deployment scheme of the implanted component output by the deployment planning stage as the target deployment scheme for the fracture site until the verification feedback data shows that it has passed.

[0008] Another aspect of this disclosure provides an electronic device, comprising: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the implantation planning method for any embodiment of the present disclosure.

[0009] Another aspect of this disclosure provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the implantation planning method for any embodiment of this disclosure.

[0010] Another aspect of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the implantation planning method for any embodiment of this disclosure.

[0011] The technical solutions of the embodiments disclosed herein can achieve at least the following technical effects: 1. By building a reinforcement learning agent targeting the fracture site, and using the 3D image data of the fracture site as input data, a closed-loop optimization mechanism is provided for the entire deployment planning process, from image preprocessing, deployment planning to verification feedback. While ensuring the accuracy of deployment planning, the deployment planning time is significantly shortened, and a repeatable and verifiable deployment planning system is achieved.

[0012] 2. By decoupling the planning stage from the deployment planning process dimension, incorporating an interpretable toolchain into the reinforcement learning framework, defining the tool call actions of each planning stage as the action space, and using the tool call records of the current decision round to update the state space, the action strategies of each decision round are made traceable, auditable, and interventionable, and the deployment scheme is made verifiable, reproducible, and interpretable. This overcomes the trust crisis of traditional deployment planning and improves the stability and reliability of the deployment scheme.

[0013] 3. By incorporating the image preprocessing and verification feedback stages into the closed-loop optimization mechanism of the reinforcement learning agent, and using the original 3D image data as the initial input for policy iteration, adaptive iterative optimization of image processing parameters is achieved. This eliminates the negative impact of preprocessed image quality fluctuations on deployment planning results. Simultaneously, the verification feedback data is used to anchor the policy iteration convergence criterion, breaking the optimization mode where planning and verification are mutually isolated. This supports iterative updates of verification feedback standards, overcomes the inherent lag and one-sidedness of fixed verification mechanisms, and enables deployment schemes to more comprehensively adapt to the anatomical differences of different individuals, thereby improving the generalization ability and robustness of deployment planning.

[0014] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a method for planning the placement of an implantable component according to an embodiment of this disclosure; Figure 2 A flowchart illustrating another method for planning the placement of implantable components according to an embodiment of this disclosure; Figure 3 A flowchart illustrating another method for planning the placement of implantable components according to an embodiment of this disclosure; Figure 4 This is a schematic diagram of the structure of an implantable component placement planning device provided in one embodiment of the present disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present disclosure. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present disclosure, the technical solutions of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present disclosure, and not all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present disclosure.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] Figure 1 This is a flowchart illustrating a method for planning the placement of implantable components according to an embodiment of this disclosure. This embodiment is applicable to the planning of placement of implantable components applied to fracture structures. The method can be executed by a placement planning device for implantable components, which can be implemented in hardware and / or software and can be configured in a terminal device. Figure 1 As shown, the method includes: S110. Input the three-dimensional image data of the fracture site into the reinforcement learning agent.

[0020] Specifically, the fracture site refers to the anatomical structure where bone structure is broken or damaged. For example, the fracture site may be the femoral neck, proximal humerus, distal radius, tibial plateau, or pelvis, but is not limited to the examples given above.

[0021] The three-dimensional image data may come from medical imaging equipment such as CT (Computed Tomography), CBCT (Cone Beam Computed Tomography), MRI (Magnetic Resonance Imaging), DR (Digital Radiography), and DSA (Digital Subtraction Angiography), but is not limited to the examples given above.

[0022] Reinforcement learning refers to the process by which an agent, within the framework of a Markov decision process, continuously interacts with the environment, iteratively optimizes its strategy based on the state representation and quantified reward value returned by the environment, and learns autonomously and converges to the optimal strategy with the goal of maximizing long-term cumulative reward.

[0023] In this embodiment, the reinforcement learning agent is constructed based on a decision model, a tool library, and a deployment knowledge base. The decision model is an algorithmic model that completes the deployment planning process by calling planning tools from the tool library. For example, the tool calling action of the decision model can be implemented through a tool router. The tool library includes a set of tools corresponding one-to-one with the planning stages in the deployment planning process. Each set of tools contains one or more preset planning tools, which are encapsulated callable functions used to complete the functional tasks of the corresponding planning stage.

[0024] The deployment planning process includes an image preprocessing stage, an implant component deployment planning stage, and a verification and feedback stage. Specifically, the image preprocessing stage is used to perform preprocessing tasks on the 3D image data, providing high-quality, standardized anatomical modeling constraints for the deployment planning stage. Preprocessing tasks include, but are not limited to, image enhancement, 3D reconstruction, anatomical structure segmentation, and fracture structure recognition. The deployment planning stage is used to plan the deployment of the implant component in the 3D space of the fracture site, outputting a deployment scheme. This includes, but is not limited to, the selection, quantity, size parameters, spatial positioning, and implantation path of the implant component. The verification and feedback stage is used to perform multi-dimensional verification, evaluation, and feedback on the deployment scheme from the deployment planning stage, outputting verification feedback data, such as evaluation dimensions covering biomechanical adaptability, anatomical safety, operability, and long-term stability.

[0025] The toolsets for different planning stages are functionally decoupled, but are progressively coupled at the data flow processing logic level. Standardized interface protocols are used to ensure compatibility of cross-stage calls and smooth data flow.

[0026] Specifically, the deployment knowledge base represents a structured set of knowledge regarding deployment specifications and scenario constraints. This knowledge base supports the decision-making model in performing knowledge-driven parameter matching and quantitative screening during the deployment planning process, providing objective parameter constraints for the model's tool invocation actions. For example, the deployment knowledge base may include anatomical constraint specifications, fracture classification standards, implant component selection guidelines, a typical deployment case library, and a path matching sample set, but is not limited to the examples given above.

[0027] S120. The reinforcement learning agent updates the state space based on the deployed knowledge base, the three-dimensional image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and uses the tool call actions corresponding to the tool base as the action space.

[0028] In this embodiment, the tool invocation record includes the tool invocation action and the corresponding tool output data. The tool invocation action represents the action configuration used by the decision model to execute the tool invocation process, including the tool identifier and parameter configuration of the invoked target planning tool; the tool output data records the result data generated by the target planning tool.

[0029] For example, in the current decision round, the first Tool call action It can be represented as ,in, Indicates the first The tool identifier of the goal planning tool corresponding to each tool call action. , Indicates tool library, This indicates the parameter configuration of the target planning tool.

[0030] Tool Invocation Action Tool call logs after occurrence It can be represented as: in, Indicates the tool invocation action The tool outputs data after the event occurs.

[0031] For example, in the first In the first round of decision-making... Tool call action state space It can be represented as ,in, This represents 3D image data and the deployment of a knowledge base. Indicates the first The validation feedback data obtained from each round of decision-making.

[0032] S130. Using the decision model, based on the state space and the action space, the planning steps in the deployment planning process are executed sequentially to output the verification feedback data of the current decision round.

[0033] Specifically, through the aforementioned decision model, in the first... In each planning stage, the state space corresponding to the current planning stage is received. According to the state space From the toolkit corresponding to the current planning stage, determine the optimal tool invocation action. Actions invoked based on the aforementioned tools Initiate a tool invocation request to the tool router and receive the tool output data returned by the tool router. .

[0034] If the current planning stage is not the last planning stage, then the binary tool for this interaction will be used. Append to the tool call record to obtain the updated state space. Continue executing the next planning stage until the entire deployment planning process is completed, and output the verification feedback data for the current decision-making round. Through the above iterative cycle, the decision-making model can dynamically and interactively complete the entire deployment planning process.

[0035] S140. Determine whether the verification feedback data indicates failure. If yes, proceed to S150; otherwise, proceed to S160.

[0036] In an optional embodiment, determining whether the verification feedback data indicates failure includes: determining whether the verification feedback data indicates failure based on the null value state of the verification feedback data.

[0037] Specifically, if the verification feedback data is empty, it indicates that the verification feedback data has passed; if the verification feedback data is not empty, it indicates that the verification feedback data has failed. Here, the verification feedback data not being empty means that the deployment plan for the current decision-making round has an anomaly in at least one verification dimension, which is recorded in the verification feedback data, or that a correction suggestion for this anomaly is recorded in the verification feedback data.

[0038] In another optional embodiment, the verification feedback data includes feedback quantification features, which are used to quantify the compliance performance of the deployment scheme in the current decision-making round on at least one verification dimension. Specifically, the feedback quantification features include verification feedback labels and / or verification feedback scores. The verification feedback labels are used to indicate the compliance judgment results of the deployment scheme in the current decision-making round on each verification dimension, and the verification feedback scores are used to quantify the degree of compliance of the deployment scheme in the current decision-making round on each verification dimension.

[0039] In this embodiment, determining whether the verification feedback data is unsuccessful includes: determining whether the verification feedback data is unsuccessful based on the feedback quantification characteristics.

[0040] In a specific example, if the verification feedback label is unqualified or the verification feedback score is lower than a preset score threshold, it indicates that the verification feedback data is unqualified; conversely, if the verification feedback label is qualified and the verification feedback score is not lower than the preset score threshold, it indicates that the verification feedback data is qualified.

[0041] S150. Using the planning reward mechanism corresponding to the reinforcement learning agent, determine the planning reward value for the current decision round, optimize the parameters of the decision model based on the planning reward value, and use the optimized decision model as the decision model for the next decision round.

[0042] The planning reward value is used to drive the parameter updates and strategy iterations of the decision model, and the optimization objective of the decision model is to maximize the expected planning reward value. For example, the optimization objective of the decision model can be expressed as: in, This represents the learnable parameters within the decision-making model. Indicates the learnable parameters Defined decision model, Represents the expectation function, Indicates the first The planning reward value corresponding to each decision round.

[0043] In an optional embodiment, the verification feedback data includes feedback quantification features, and the step of using the planning reward mechanism corresponding to the reinforcement learning agent to determine the planning reward value of the current decision round includes: using the feedback quantification features as the planning reward value of the current decision round.

[0044] In another optional embodiment, the step of using the planning reward mechanism corresponding to the reinforcement learning agent to determine the planning reward value of the current decision round includes: for each tool call action in the tool call record, determining the process reward value of the tool call action based on the tool output data corresponding to the tool call action; summing the process reward values ​​according to the discount factor, and determining the planning reward value of the current decision round based on the summation result.

[0045] Wherein, the process reward value characterizes the immediate feedback efficiency of the tool invocation behavior on the deployment scheme optimization process, and the discount factor characterizes the time decay effect of the process reward value in the action sequence. For example, the first... The planning reward value corresponding to each decision round satisfies The following formula: in, Indicates the number of times the tool is called. Let be the discount factor, representing the th Discount weight for each tool call action Indicates the first The reward value for each tool's invocation action.

[0046] In an optional embodiment, determining the planning reward value for the current decision round based on the obtained summation result includes: using the summation result as the planning reward value for the current decision round. In another optional embodiment, determining the planning reward value of the current decision round based on the obtained summation result includes: performing a weighted operation on the summation result and the feedback quantification feature to obtain the planning reward value of the current decision round.

[0047] Specifically, based on the optimized decision model, proceed to the next decision round and repeat S120-S150.

[0048] S160. The placement scheme of the implanted component output by the placement planning stage shall be used as the target placement scheme for the fracture site.

[0049] Based on the above embodiments, the method may optionally further include: before entering the next decision round, determining whether the number of decision iterations corresponding to the current decision round has reached the upper limit of the number of iterations; if so, terminating the reinforcement learning process, and outputting abnormal diagnostic data according to the verification feedback data corresponding to each decision round.

[0050] Specifically, the abnormal diagnostic data characterizes the source information of the reasons why the reinforcement learning agent failed to train. For example, the abnormal diagnostic data includes, but is not limited to, the reasons for planning failure, the classification of failure types, bottleneck planning links, and suggestions for manual review.

[0051] The technical solution of this embodiment adopts a closed-loop optimization mechanism for the entire deployment planning process, from image preprocessing and deployment planning to verification feedback. This significantly shortens the deployment planning time while ensuring the accuracy of deployment planning, and realizes a repeatable, interpretable, and verifiable deployment planning system. By incorporating an interpretable toolchain into the reinforcement learning framework, it overcomes the trust crisis in traditional deployment planning and improves the stability and reliability of the deployment scheme. By incorporating the image preprocessing and verification feedback stages into the closed-loop optimization mechanism of the reinforcement learning agent, the deployment scheme can more comprehensively adapt to the anatomical differences of different individuals, improving the generalization ability and robustness of deployment planning.

[0052] Figure 2This is a flowchart illustrating another method for planning the deployment of an implanted component according to an embodiment of this disclosure. This embodiment refines the step of "using the planning reward mechanism corresponding to the reinforcement learning agent to determine the planning reward value of the current decision round" in the previous embodiment. The step of using the planning reward mechanism corresponding to the reinforcement learning agent to determine the planning reward value of the current decision round includes: obtaining feedback quantification features from the verification feedback data; determining the final reward value based on the feedback quantification features and operational statistical parameters; and determining the planning reward value of the current decision round based on the final reward value. Figure 2 As shown, the method includes: S210. Input the three-dimensional image data of the fracture site into the reinforcement learning agent.

[0053] S220. The reinforcement learning agent updates the state space based on the deployed knowledge base, the three-dimensional image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and uses the tool call actions corresponding to the tool base as the action space.

[0054] S230. Using the decision model, based on the state space and the action space, the planning steps in the deployment planning process are executed sequentially to output the verification feedback data of the current decision round.

[0055] S210-S230 in this embodiment are the same as those in the above embodiment. Figure 1 The steps S110-S130 shown are logically consistent and will not be described again in this embodiment.

[0056] S240. Determine whether the verification feedback data indicates failure. If yes, proceed to S250; otherwise, proceed to S280.

[0057] S250. Obtain the feedback quantification features in the verification feedback data, and determine the final reward value based on the feedback quantification features and the running statistical parameters.

[0058] In this embodiment, the feedback quantification features include verification feedback labels and / or verification feedback scores, and the operational statistics parameters include at least one of the following: the number of decision iterations corresponding to the current decision round, the number of invalid tool calls, and the cumulative tool call cost.

[0059] Specifically, the number of invalid tool calls refers to the number of times the target planning tool called in the current decision round fails to produce valid tool output data or causes the deployment planning process to stall. The cumulative tool call cost represents the sum of the call costs of all target planning tools in the current decision round. For example, the call cost includes, but is not limited to, the consumption of computing resources, data processing time, and memory resource occupation.

[0060] The feedback quantification feature refers to the reward item in the planning reward mechanism, and the operational statistical parameter refers to the penalty item in the planning reward mechanism. A larger reward value and a smaller penalty value indicate a better quality deployment plan in the current decision-making round; conversely, a smaller reward value and a larger penalty value indicate a worse quality deployment plan in the current decision-making round.

[0061] For example, the first The final reward value corresponding to each decision round Satisfy the following formula: in, Indicates the verification feedback score The corresponding reward value, Indicates verification feedback label The corresponding reward value, Indicates the number of decision iterations With the upper limit of the number of iterations Proportional information, Indicates the number of invalid tool calls. The corresponding penalty items, Indicates the cumulative cost of tool usage. The corresponding penalty items, This represents the weight coefficients corresponding to the five reward and punishment indicators mentioned above.

[0062] S260. Determine the planning reward value for the current decision round based on the final reward value.

[0063] In an optional embodiment, determining the planning reward value for the current decision round based on the final reward value includes: using the final reward value as the planning reward value for the current decision round.

[0064] In another optional embodiment, determining the planning reward value of the current decision round based on the final reward value includes: for each tool call action in the tool call record, determining the process reward value of the tool call action based on the tool output data corresponding to the tool call action; and determining the planning reward value of the current decision round based on each process reward value and the final reward value.

[0065] In this embodiment, the process reward value is obtained by weighting at least one of the output quality item, the call cost penalty item, the tool error penalty item, and the invalid call indication item.

[0066] Specifically, the output quality item describes the quality level of the tool's output data, the call cost penalty item is used to quantify the resource consumption caused by calling the planning tool, the tool error penalty item is used to count the number of error entries in the error set generated by the target planning tool corresponding to the tool call action, and the invalid call indication item is used to mark whether the planning tool was successfully called.

[0067] For example, the first The process reward value corresponding to each tool invocation action satisfies the following formula: in, This indicates the data quality of the tool's output data. Indicates the cost of calling the tool. This represents the set of errors generated by the invoked planning tool. Indicates an invalid call flag. This represents the weight coefficient corresponding to the reward and punishment items in each process.

[0068] For example, the first The planning reward value corresponding to each decision round satisfies The following formula: This embodiment adds reward and penalty constraints to each tool call action, which can provide real-time feedback on the quality of tool call actions in each round of decision iteration, alleviate the inherent lag of a single final reward, effectively accelerate the convergence speed of the model, reduce resource waste and frequency of invalid operations in the deployment planning process, and simultaneously optimize the generation efficiency and overall quality of deployment schemes.

[0069] S270. Optimize the parameters of the decision model based on the planned reward value, and use the optimized decision model as the decision model for the next decision round.

[0070] S280. The placement scheme of the implanted component output by the placement planning stage shall be used as the target placement scheme for the fracture site.

[0071] S270 in this embodiment is the same as that in the above embodiment. Figure 1 The logic of step S150 shown is the same as that in the above embodiment, and step S280 is the same as that in the above embodiment. Figure 1 The logic of step S160 shown is consistent, and will not be repeated in this embodiment.

[0072] Based on the above embodiments, optionally, the reinforcement learning agent further includes a scheme output model, which is used to implement at least one of the following: By calling the interface output tool corresponding to the preset application device, the target deployment scheme is converted into an interface data packet that can be recognized by the preset application device; By calling a visualization tool, the target layout scheme is rendered onto the real-time three-dimensional image corresponding to the fracture site to obtain a three-dimensional layout effect diagram; By invoking the report generation tool, a natural language report is generated based on the 3D deployment effect diagram, the target deployment scheme, the verification feedback data, and the interface data packet.

[0073] In an optional embodiment, the target deployment scheme includes at least one component deployment sequence corresponding to each implantable component. The component deployment sequence includes at least one deployment parameter among the following: component selection, size parameters, spatial pose, and implantation method. The spatial pose characterizes the implantation position and orientation of the implantable component in the coordinate system of the medical imaging device. The preset application device refers to a positioning device with orthopedic positioning capabilities. For example, the preset application device can be an orthopedic robot or an orthopedic navigation system, whose interface protocol is bidirectionally data-coupled with the reinforcement learning agent.

[0074] Specifically, through the interface output tool, the coordinate transformation matrix between the preset application device and the medical imaging device is obtained, and the spatial pose in the target deployment scheme is mapped to the local coordinate system of the preset application device, completing cross-device spatial consistency registration. Throughout the data processing, relying on classic geometric and mathematical algorithms such as rigid body registration, matrix transformation, unit conversion, and field integrity checks, precise conversion and compliance verification are completed sequentially for coordinate transformation, boundary verification, and format verification, ensuring the accuracy and effectiveness of data transmission and mapping.

[0075] For example, the first The in-point coordinates of the implanted component in the local coordinate system End point coordinates and direction vector The following mapping relationship is satisfied: in, This represents the coordinate transformation matrix corresponding to the CT device and the medical imaging device. Indicates the first The ingress coordinates of each implanted component in the CT coordinate system Indicates the first The endpoint coordinates of each implanted component in the CT coordinate system.

[0076] In one specific embodiment, the interface data includes, but is not limited to, planning number, component number, implantation method, selection, size parameters, spatial pose, local coordinate system parameters, coordinate transformation matrix, registration error, and feedback quantization features. The registration error is used to describe the degree of deviation in spatial mapping between the medical imaging device and the preset application device. For example, when the preset application device is an orthopedic robot, the registration error can be determined based on the acquisition spatial parameters of the medical imaging device and the body spatial parameters of the orthopedic robot. When the preset application device is an orthopedic navigation system, the registration error can be determined based on the spatial deviation between the three-dimensional image data and the real-time navigation data under the orthopedic navigation system.

[0077] For example, the interface data packet can be encapsulated in JSON (JavaScript Object Notation), XML (Extensible Markup Language), CSV (Comma-Separated Values), and DICOM (Digital Imaging and Communications in Medicine) formats to adapt to the standard data interfaces of various pre-defined application devices. After the interface data packet is imported into the pre-defined application device, the device parses the packet to obtain the implantation path data for each implantable component. This path information is used in component placement operations such as spatial navigation, robotic arm spatial positioning, and actuator-assisted positioning. The implantation path includes spatial pose, implantation method, and dimensional parameters.

[0078] For example, the 3D layout rendering includes, but is not limited to, visualization elements corresponding to objects such as fracture areas, implantation positions of implanted components, spatial poses, safe areas, restricted areas, and key parameters. The visualization elements include, but are not limited to, primitive types such as points, lines, surfaces, volumes, colors, transparency, textures, and dynamic annotations.

[0079] For example, the report generation tool may be a multimodal driving engine, a large language model, or a rule text generator, etc. The natural language report may be used to describe the annotation content corresponding to the 3D deployment effect diagram, the deployment logic corresponding to the target deployment scheme, the parameter basis, the structured content corresponding to the interface data packet, and the planning interpretation content corresponding to the verification feedback data, etc.

[0080] Based on the above embodiments, optionally, the scheme output model is also used to cross-validate the natural language report with the target deployment scheme and interface data packets by calling a consistency verification tool, so as to avoid semantic deviations or logical contradictions that may occur in the natural language report.

[0081] In traditional reinforcement learning paradigms, agent policy optimization relies on a final reward mechanism related to business objectives. However, such final rewards cannot constrain intermediate decision-making behavior during the agent's iteration process. The irrationality of intermediate decision-making behavior cannot be identified and corrected in a timely manner, which can easily lead to the continuous accumulation of wrong decisions and problems such as sparse training gradients and poor model stability.

[0082] The technical solution of this embodiment incorporates the feedback quantification features in the verification feedback data into the final reward mechanism, binding the deployment planning stage of the agent with the final reward signal. This enables refined evaluation and constraint of the deployment planning stage, effectively solving the problem of the lack of intermediate process supervision in traditional reward mechanisms. It can accurately identify and punish unreasonable deployment behaviors in the deployment planning stage, avoiding the accumulation of deviations, and provide complete, dense, and accurate gradient supervision signals for the agent's strategy iteration. This enhances the convergence speed and iterative stability of the learning agent, making the final deployment strategy take into account both process standardization and global optimality, and improving the deployment accuracy of the implanted components.

[0083] Figure 3 This is a flowchart of another implantable component deployment planning method provided in one embodiment of this disclosure. This embodiment further refines the "through the decision model, according to the state space and the action space, sequentially execute the planning stages in the deployment planning process to output the verification feedback data of the current decision round" in the above embodiment. The step of sequentially executing the planning stages in the deployment planning process according to the state space and the action space to output the verification feedback data of the current decision round through the decision model includes: sequentially executing the planning stages in the deployment planning process according to the state space and the action space through the decision model; when the planning stage is a verification feedback stage, verifying the implantable component deployment scheme output by the deployment planning stage by calling the target planning tool corresponding to the planning verification node to obtain scheme verification data; and parsing the scheme verification data by calling the target planning tool corresponding to the verification feedback node to obtain verification feedback data. Figure 3 As shown, the method includes: S310. Input the three-dimensional image data of the fracture site into the reinforcement learning agent.

[0084] S320. The reinforcement learning agent updates the state space based on the deployed knowledge base, the three-dimensional image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and uses the tool call actions corresponding to the tool base as the action space.

[0085] S310-S320 in this embodiment are the same as those in the above embodiments. Figure 1 The steps S110-S120 shown are logically consistent and will not be described again in this embodiment.

[0086] S330. Using the decision model, the planning steps in the deployment planning process are executed sequentially according to the state space and the action space.

[0087] In one optional embodiment, the preprocessing nodes in the image preprocessing stage include a 3D reconstruction node, a fracture identification node, a fracture classification node, and a region segmentation node. The tool set corresponding to the image preprocessing stage contains a node tool set that corresponds one-to-one with the preprocessing nodes. The tool output data corresponding to the image preprocessing stage includes a 3D geometric model, a set of fracture structures, a fracture classification, and a set of prohibited regions corresponding to the fracture site.

[0088] Specifically, the 3D reconstruction node is configured to receive 3D image data and perform 3D reconstruction processing to generate a 3D geometric model matching the fracture site. The fracture identification node is used to perform semantic segmentation and feature recognition on the 3D geometric model and output a set of fracture structures, which includes bony injury structures in the fracture site. For example, the bony injury structures include fracture lines and / or fracture fragments. The fracture lines are used to depict the trajectory of bone fracture at the fracture site, and the fracture fragments are used to depict independent bone fragments formed by bone fracture at the fracture site. The fracture classification node retrieves multi-dimensional feature parameters such as position parameters, size parameters, and spatial angle parameters of each fracture structure in the fracture structure set to complete automatic classification and output the fracture classification corresponding to the fracture site. For example, the classification system used for fracture classification includes, but is not limited to, AO classification, Letournel-Judet classification, etc. The region division node combines the 3D geometric model and the fracture structure set to perform spatial segmentation operations, dividing the 3D geometric model to generate a set of forbidden regions. The set of forbidden regions contains multiple forbidden regions corresponding to the forbidden region type, and the forbidden region type can be one or more.

[0089] For example, the segmentation toolset corresponding to the segmentation process may include residual connections, spatial attention mechanisms, and 3DU-net networks, but is not limited to the given example.

[0090] Based on the above embodiments, optionally, the image preprocessing stage further includes a quality verification node. The quality verification node is used to perform quality verification on the image processing results output by each preprocessing stage, and outputs the quality verification result and terminates the deployment planning process if the quality verification result fails.

[0091] For example, the quality verification includes motion artifacts, incomplete field of view, excessive layer thickness, metal artifacts, or insufficient resolution, but is not limited to the examples given above.

[0092] This embodiment refines the preprocessing nodes in the image preprocessing stage, configures dedicated node tools for each preprocessing node, and incorporates multiple preprocessing nodes into the iterative optimization process of the reinforcement learning agent. This enables independent parameter tuning, independent function control, and adaptive iterative optimization for each preprocessing node, breaking the limitations of traditional preprocessing schemes that are rigid overall and have uniform parameter optimization. It improves the accuracy, adaptability, and controllability of image preprocessing, provides reliable data support for subsequent planning stages, improves the stability, accuracy, and robustness of deployment planning, and effectively reduces planning deviations caused by image preprocessing defects.

[0093] S340. When the planning stage is a verification feedback stage, the deployment scheme of the implanted component output by the deployment planning stage is verified by calling the target planning tool corresponding to the planning verification node to obtain scheme verification data.

[0094] In this embodiment, the execution nodes in the verification feedback stage include planning verification nodes and verification feedback nodes, and the stage toolset corresponding to the verification feedback stage contains node toolsets that correspond one-to-one with the execution nodes.

[0095] The node toolset includes at least two planning tools. The two types of node toolsets are decoupled and independent at the business function level, but are progressively coupled at the data flow processing logic level.

[0096] In one optional embodiment, the scheme verification data includes at least one of anatomical safety, mechanical stability, scheme compliance, image quality consistency, and iteration efficiency.

[0097] The anatomical safety refers to the distance information between the implanted component and the prohibited area of ​​the fracture site. For example, the prohibited area includes, but is not limited to, neurovascular bundles, joint cavities, nerve distribution areas, soft tissue areas, and existing scar areas. If the implanted component is too close to the neurovascular bundles, it may cause nerve compression or vascular damage; if too close to the joint cavity, it may lead to limited joint movement or induce traumatic arthritis; if too close to the nerve distribution areas, it may induce limb numbness or movement disorders; and if too close to the soft tissue areas, it is prone to causing local inflammatory reactions or tissue necrosis.

[0098] The mechanical stability refers to the degree of matching between the placement parameters of the implanted component and the anatomical mechanics specifications in the placement knowledge base. Specifically, the anatomical mechanics specifications characterize the stability requirements of the fracture site under physiological loads. For example, the shear angle between the implanted component and the fracture line is ≥60° to ensure that the fracture ends do not shift under axial compression and torsional loads. The effective length of the implanted component covers at least 1.5 times the bone diameter on both sides of the fracture ends to distribute stress and prevent loosening of the internal fixation. The implantation method of the implanted component is included in the standardized operating procedure for the corresponding fracture classification.

[0099] The compliance of the proposed scheme characterizes the degree of matching between the placement scheme of the implanted components and the placement specifications in the placement knowledge base. Specifically, the placement specifications characterize the standardized operational requirements for fracture sites in clinical practice. For example, the placement specifications include, but are not limited to, the selection of implanted components conforming to the scope of the national medical device registration certificate, and the number of implanted components conforming to the specified number range for the corresponding fracture type in the "Guidelines for Clinical Application of Orthopedic Implant Devices".

[0100] The image quality consistency refers to the degree of consistency between the quality information of the three-dimensional image data and the image processing results. Specifically, the processing results of the three-dimensional image data refer to the image data obtained by performing image processing on the three-dimensional image data, and the processing results include, but are not limited to, three-dimensional geometric models, fracture enhancement images, fracture segmentation masks, etc.

[0101] The iterative efficiency represents the efficiency quantification information corresponding to at least one of the following: the number of decision iterations, the number of invalid tool calls, and the type of repeated errors for the current decision round.

[0102] S350. By calling the target planning tool corresponding to the verification feedback node, the verification data of the scheme is parsed to obtain the verification feedback data.

[0103] Specifically, the feedback quantification features in the verification feedback data are determined by weighted calculation of at least one scheme verification index in the scheme verification data.

[0104] For example, the first The validation feedback score corresponding to each decision round Satisfy the following formula: in, The parameter value represents the safety of the dissection. The parameter values ​​represent the mechanical stability. The parameter values ​​represent the compliance of the solution. The parameter values ​​represent the consistency of image quality. The parameter value represents the iteration efficiency. This represents the weight coefficients corresponding to the verification indicators of the above five schemes.

[0105] Specifically, when the verification feedback score is greater than a preset score threshold, the verification feedback label is set to 1; when the verification feedback score is less than or equal to the preset score threshold, the verification feedback label is set to 0.

[0106] In an optional embodiment, when the verification feedback data shows a failure, the verification feedback data includes the tool error set corresponding to the tool library and / or correction suggestions for the deployment scheme of the implanted component output in the deployment planning stage.

[0107] For example, the first The validation feedback data corresponding to each decision round constitutes a quadruple: in, Indicates whether the verification passed. Indicates the first The set of tool errors corresponding to all tools invoked in each decision round. This indicates a suggested revision.

[0108] The tool error set represents the invoked tool that caused the error and the error type information. For example, the correction suggestions include, but are not limited to, adjusting the implantation position, increasing the safety distance, changing the angle of the implantation component, increasing the length of the implantation component, changing the implantation method, recalculating the safety zone, and requiring manual review.

[0109] S360. Determine whether the verification feedback data indicates failure. If yes, proceed to S370; otherwise, proceed to S380.

[0110] S370. Using the planning reward mechanism corresponding to the reinforcement learning agent, determine the planning reward value for the current decision round, optimize the parameters of the decision model based on the planning reward value, and use the optimized decision model as the decision model for the next decision round.

[0111] S380. The placement scheme of the implanted component output by the placement planning stage shall be used as the target placement scheme for the fracture site.

[0112] S360-S380 in this embodiment are the same as those in the above embodiments. Figure 1 The logic of steps S140-S160 shown is consistent with, or is consistent with, those in the above embodiments. Figure 2 The steps S240-S280 shown are logically consistent and will not be described again in this embodiment.

[0113] The technical solution of this embodiment refines the execution nodes in the verification feedback process, configures dedicated node tools for each execution node, and incorporates multiple execution nodes into the iterative optimization process of the reinforcement learning agent. This enables independent parameter tuning, independent functional control, and adaptive iterative optimization for each execution node, breaking the limitations of traditional preprocessing schemes that are rigid as a whole and have uniform parameter optimization. It improves the accuracy, adaptability, and controllability of verification feedback data, provides reliable data support for the selection of deployment schemes, improves the stability, accuracy, and robustness of deployment planning, and effectively reduces planning deviations caused by the selection of verification indicators.

[0114] It is understood that the node division within the image preprocessing stage, deployment planning stage, and verification feedback stage in the embodiments of this disclosure is merely a preferred example and not the only implementation method. Without departing from the core innovative principles of this disclosure, other adaptive adjustments made to suit different application scenarios, different device models, and different data acquisition conditions, as long as the technical solutions can achieve the aforementioned technical effects, are all within the protection scope of this disclosure.

[0115] Taking a left sacroiliac joint separation or sacral fracture as an example, firstly, the acquired pelvic CT images are input into a reinforcement learning agent. Through the decision model within this agent, the pelvic bone structure and fracture line are identified in the pelvic CT images during image preprocessing, and the left sacroiliac joint separation area is delineated. In the placement planning stage, candidate paths for S1 and S2 screws are matched based on the output data from the image preprocessing stage and the placement knowledge base, and a placement plan is output. In the verification and correction stage, the placement plan is verified for anatomical safety, mechanical stability, compliance, image quality consistency, and iteration efficiency. If the distance between the S1 screw and the sacral foramen is less than the safe distance threshold, a correction suggestion is output: "Adjust the entry point 2mm backward and upward, increase the direction vector relative to the cross-section by 3°, and recalculate the implantation length." If the verification feedback data indicates failure, the model parameters and state space of the decision model are updated, and the next decision round continues.

[0116] If the verification feedback data shows that the plan passes, the output includes an interface data package that the orthopedic robot can recognize, a structured parameter table corresponding to the target deployment plan, a 3D deployment effect diagram, a natural language report, and a plan verification log for each decision round. The number of decision iterations corresponding to the current decision round is equal to the upper limit of the number of iterations. The output also includes anomaly handling suggestions, such as the reasons for planning failure, the risk locations of the plan, the deployment specifications that the plan does not meet, and suggestions for manual review.

[0117] It should be noted that the collection, use, storage, sharing and transfer of user personal information involved in the technical solution disclosed herein all comply with the provisions of relevant laws and regulations, and require notification to users and obtaining their consent or authorization. Where applicable, user personal information has been subjected to de-identification and / or anonymization and / or encryption technical processing.

[0118] The following are embodiments of the implantable component layout planning device provided in this disclosure. This device and the implantable component layout planning method of the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the implantable component layout planning device, please refer to the content of the implantable component layout planning method in the above embodiments.

[0119] Figure 4 This is a schematic diagram of a layout planning device for implantable components provided in one embodiment of the present disclosure. Figure 4 As shown, the device includes: a three-dimensional image data input module 410, a state space update module 420, a verification feedback data output module 430, a decision model optimization module 440, and a target deployment scheme determination module 450.

[0120] The three-dimensional image data input module 410 is used to input the three-dimensional image data of the fracture site into the reinforcement learning agent. The reinforcement learning agent is built based on a decision model, a tool library and a deployment knowledge base. The tool library includes a set of tools corresponding to the planning steps in the deployment planning process. The deployment planning process includes an image preprocessing step, a deployment planning step for implanted components and a verification feedback step. The state space update module 420 is used by the reinforcement learning agent to update the state space based on the deployed knowledge base, the three-dimensional image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and to use the tool call actions corresponding to the tool base as the action space; wherein, the tool call records include tool call actions and tool output data corresponding to the tool call actions; The verification feedback data output module 430 is used to execute the planning stage in the deployment planning process sequentially according to the state space and the action space through the decision model and output the verification feedback data of the current decision round. The decision model optimization module 440 is used to determine the planning reward value of the current decision round by adopting the planning reward mechanism corresponding to the reinforcement learning agent when the verification feedback data shows that it fails, optimize the parameters of the decision model according to the planning reward value, and use the optimized decision model as the decision model for the next decision round. The target layout scheme determination module 450 is used to take the layout scheme of the implanted component output by the layout planning stage as the target layout scheme of the fracture site until the verification feedback data shows that it has passed.

[0121] The technical solution of this embodiment adopts a closed-loop optimization mechanism for the entire deployment planning process, from image preprocessing and deployment planning to verification feedback. This significantly shortens the deployment planning time while ensuring the accuracy of deployment planning, and realizes a repeatable, interpretable, and verifiable deployment planning system. By incorporating an interpretable toolchain into the reinforcement learning framework, it overcomes the trust crisis in traditional deployment planning and improves the stability and reliability of the deployment scheme. By incorporating the image preprocessing and verification feedback stages into the closed-loop optimization mechanism of the reinforcement learning agent, the deployment scheme can more comprehensively adapt to the anatomical differences of different individuals, improving the generalization ability and robustness of deployment planning.

[0122] In an optional embodiment, the decision model optimization module 440 includes: A feedback quantification feature acquisition unit is used to acquire feedback quantification features in the verification feedback data, wherein the feedback quantification features include verification feedback labels and / or verification feedback scores. The final reward value determination unit is used to determine the final reward value based on the feedback quantification characteristics and operational statistical parameters; wherein, the operational statistical parameters include at least one of the following: the number of decision iterations corresponding to the current decision round, the number of invalid tool calls, and the cumulative tool call cost; The planning reward value determination unit is used to determine the planning reward value for the current decision round based on the final reward value.

[0123] In one optional embodiment, the planning reward value determination unit is specifically used for: For each tool call action in the tool call record, determine the process reward value of the tool call action based on the tool output data corresponding to the tool call action; Based on the process reward value and the final reward value, determine the planning reward value for the current decision round; The process reward value is obtained by weighting at least one of the following: output quality item, call cost penalty item, tool error penalty item, and invalid call indication item.

[0124] In an optional embodiment, the execution nodes in the verification feedback process include planning verification nodes and verification feedback nodes, and the process tool set corresponding to the verification feedback process includes a node tool set that corresponds one-to-one with the execution nodes. The verification feedback data output module 430 includes: A deployment planning process execution unit is used to execute the planning steps in the deployment planning process sequentially based on the state space and the action space, according to the decision model. The scheme verification data determination unit is used to verify the layout scheme of the implanted component output by the layout planning stage and obtain scheme verification data when the planning stage is a verification feedback stage. The verification feedback data determination unit is used to parse the scheme verification data to obtain verification feedback data by calling the target planning tool corresponding to the verification feedback node.

[0125] In an optional embodiment, the scheme verification data includes at least one of anatomical safety, mechanical stability, scheme compliance, image quality consistency, and iteration efficiency. The anatomical safety characterizes the distance information between the implanted component and the prohibited area of ​​the fracture site. The mechanical stability characterizes the degree of matching between the placement parameters of the implanted component and the anatomical mechanical specifications in the placement knowledge base. The scheme compliance characterizes the degree of matching between the placement scheme of the implanted component and the placement specifications in the placement knowledge base. The image quality consistency characterizes the degree of consistency between the quality information of the three-dimensional image data and the image processing results. The iteration efficiency characterizes the efficiency quantification information corresponding to at least one of the following: the number of decision iterations, the number of invalid tool calls, and the repeated error type corresponding to the current decision round.

[0126] In an optional embodiment, when the verification feedback data shows a failure, the verification feedback data includes the tool error set corresponding to the tool library and / or correction suggestions for the deployment scheme of the implanted component output in the deployment planning stage.

[0127] In one optional embodiment, the preprocessing nodes in the image preprocessing stage include a 3D reconstruction node, a fracture identification node, a fracture classification node, and a region segmentation node. The tool set corresponding to the image preprocessing stage contains a node tool set that corresponds one-to-one with the preprocessing nodes. The output data of the image preprocessing stage includes a 3D geometric model, a set of fracture structures, a fracture classification, and a set of prohibited regions corresponding to the fracture site.

[0128] In an optional embodiment, the reinforcement learning agent further includes a scheme output model, which is used to implement at least one of the following: By calling the interface output tool corresponding to the preset application device, the target deployment scheme is converted into an interface data packet that can be recognized by the preset application device; By calling a visualization tool, the target layout scheme is rendered onto the real-time three-dimensional image corresponding to the fracture site to obtain a three-dimensional layout effect diagram; By invoking the report generation tool, a natural language report is generated based on the 3D deployment effect diagram, the target deployment scheme, the verification feedback data, and the interface data packet.

[0129] The implantable component layout planning device provided in this disclosure can execute the implantable component layout planning method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the method execution.

[0130] Figure 5 This is a schematic diagram of an electronic device provided according to one embodiment of the present disclosure. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0131] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor 11. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0132] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information or data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0133] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the implantation planning method provided in the above embodiments.

[0134] In some embodiments, the implantation planning method provided in the above embodiments can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the implantation planning method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the implantation planning method by any other suitable means (e.g., by means of firmware).

[0135] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of embodiments of this disclosure.

[0136] Various embodiments of the systems and techniques described above can be implemented in the following systems or combinations thereof: digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard parts (ASSPs), system-on-chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0137] Computer programs used to implement the placement planning method for implantable components of this disclosure can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable storage medium. Examples of machine-readable storage media include, based on an electrical connection of at least one wire, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0139] To provide interaction with a user, the systems and techniques described herein can be implemented on a terminal device having: a display device for displaying information to the user (e.g., a cathode-ray tube (CRT) or liquid crystal display (LCD) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the terminal device. Other types of devices can also provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0140] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0141] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0142] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this disclosure can be achieved, and this is not limited herein.

[0143] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for planning the layout of implantable components, characterized in that, include: The three-dimensional image data of the fracture site is input into the reinforcement learning agent; wherein, the reinforcement learning agent is built based on a decision model, a tool library and a deployment knowledge base, the tool library includes a set of tools corresponding one-to-one with the planning steps in the deployment planning process, and the deployment planning process includes an image preprocessing step, a deployment planning step for implanted components and a verification feedback step. The reinforcement learning agent updates its state space based on the deployed knowledge base, the 3D image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and uses the tool call actions corresponding to the tool base as the action space; wherein, the tool call records include tool call actions and the tool output data corresponding to the tool call actions; Based on the decision model, the planning steps in the deployment planning process are executed sequentially according to the state space and the action space, and the verification feedback data of the current decision round is output. If the verification feedback data shows that the decision fails, the planning reward mechanism corresponding to the reinforcement learning agent is adopted to determine the planning reward value of the current decision round. The decision model is then optimized based on the planning reward value, and the optimized decision model is used as the decision model for the next decision round. Until the verification feedback data shows a pass, the placement scheme of the implanted component output in the placement planning stage is taken as the target placement scheme for the fracture site.

2. The method according to claim 1, characterized in that, The step of using the planning reward mechanism corresponding to the reinforcement learning agent to determine the planning reward value for the current decision round includes: Obtain the feedback quantification features from the verification feedback data, wherein the feedback quantification features include verification feedback labels and / or verification feedback scores; The final reward value is determined based on the feedback quantification characteristics and operational statistical parameters; wherein, the operational statistical parameters include at least one of the following: the number of decision iterations corresponding to the current decision round, the number of invalid tool calls, and the cumulative tool call cost; Based on the final reward value, determine the planning reward value for the current decision round.

3. The method according to claim 2, characterized in that, The step of determining the planning reward value for the current decision round based on the final reward value includes: For each tool call action in the tool call record, determine the process reward value of the tool call action based on the tool output data corresponding to the tool call action; Based on the process reward value and the final reward value, determine the planning reward value for the current decision round; The process reward value is obtained by weighting at least one of the following: output quality item, call cost penalty item, tool error penalty item, and invalid call indication item.

4. The method according to claim 1, characterized in that, The execution nodes in the verification feedback process include planning verification nodes and verification feedback nodes, and the process tool set corresponding to the verification feedback process contains a node tool set that corresponds one-to-one with the execution nodes. The step of using the decision model, based on the state space and the action space, to sequentially execute the planning stages of the deployment planning process and output verification feedback data for the current decision round includes: Based on the decision model, the planning steps in the deployment planning process are executed sequentially according to the state space and the action space. When the planning stage is a verification and feedback stage, the deployment scheme of the implanted component output by the deployment planning stage is verified by calling the target planning tool corresponding to the planning verification node to obtain scheme verification data. Verification feedback data is obtained by parsing the scheme verification data by calling the target planning tool corresponding to the verification feedback node.

5. The method according to claim 4, characterized in that, The scheme verification data includes at least one of anatomical safety, mechanical stability, scheme compliance, image quality consistency, and iteration efficiency. Anatomical safety characterizes the distance information between the implanted component and the prohibited area of ​​the fracture site. Mechanical stability characterizes the degree of matching between the placement parameters of the implanted component and the anatomical mechanical specifications in the placement knowledge base. Scheme compliance characterizes the degree of matching between the placement scheme of the implanted component and the placement specifications in the placement knowledge base. Image quality consistency characterizes the degree of consistency between the quality information of the three-dimensional image data and the image processing results. Iteration efficiency characterizes the efficiency quantification information corresponding to at least one of the following: the number of decision iterations, the number of invalid tool calls, and the type of repeated errors corresponding to the current decision round.

6. The method according to claim 4, characterized in that, When the verification feedback data indicates a failure, the verification feedback data includes the tool error set corresponding to the tool library and / or correction suggestions for the deployment scheme of the implanted component output in the deployment planning stage.

7. The method according to claim 1, characterized in that, The preprocessing nodes in the image preprocessing stage include 3D reconstruction nodes, fracture recognition nodes, fracture classification nodes, and region segmentation nodes. The tool set corresponding to the image preprocessing stage contains a node tool set that corresponds one-to-one with the preprocessing nodes. The output data of the image preprocessing stage includes the 3D geometric model, fracture structure set, fracture classification, and forbidden region set corresponding to the fracture site.

8. The method according to any one of claims 1-7, characterized in that, The reinforcement learning agent further includes a scheme output model, which is used to achieve at least one of the following: By calling the interface output tool corresponding to the preset application device, the target deployment scheme is converted into an interface data packet that can be recognized by the preset application device; By calling a visualization tool, the target layout scheme is rendered onto the real-time three-dimensional image corresponding to the fracture site to obtain a three-dimensional layout effect diagram; By invoking the report generation tool, a natural language report is generated based on the 3D deployment effect diagram, the target deployment scheme, the verification feedback data, and the interface data packet.

9. A device for planning the placement of implanted components, characterized in that, include: A three-dimensional image data input module is used to input three-dimensional image data of the fracture site into a reinforcement learning agent; wherein, the reinforcement learning agent is built based on a decision model, a tool library and a deployment knowledge base, the tool library includes a set of tools corresponding one-to-one with the planning stages in the deployment planning process, and the deployment planning process includes an image preprocessing stage, a deployment planning stage for implanted components and a verification feedback stage. The state space update module is used by the reinforcement learning agent to update the state space based on the deployed knowledge base, the 3D image data, the tool call records of the current decision round, and the verification feedback data obtained in the previous decision round, and to use the tool call actions corresponding to the tool base as the action space; wherein, the tool call records include tool call actions and the tool output data corresponding to the tool call actions; The verification feedback data output module is used to execute the planning steps in the deployment planning process sequentially according to the state space and the action space through the decision model and output the verification feedback data of the current decision round. The decision model optimization module is used to determine the planning reward value of the current decision round by adopting the planning reward mechanism corresponding to the reinforcement learning agent when the verification feedback data shows that it fails, optimize the parameters of the decision model according to the planning reward value, and use the optimized decision model as the decision model for the next decision round. The target deployment scheme determination module is used to take the deployment scheme of the implanted component output by the deployment planning stage as the target deployment scheme for the fracture site until the verification feedback data shows that it has passed.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the placement planning method for the implantable component according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the placement planning method for the implantable component as described in any one of claims 1-8.

12. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the placement planning method for the implantable component according to any one of claims 1-8.

Images