Artificial intelligence-based jaw position guiding device and accessory planning method

Abstract

The invention discloses a jaw position guiding device and an accessory planning method based on artificial intelligence, and the device comprises a data preprocessing module which is used for preprocessing obtained three-dimensional oral data; the intelligent processing module is used for segmenting the three-dimensional oral cavity data obtained by preprocessing from the data preprocessing module to obtain segmentation data of each tooth; performing feature recognition based on the segmentation data of each tooth; and the attachment position planning module is used for determining the optimal position of the attachment based on the features obtained from the intelligent processing module, and when the optimal position of the attachment is determined, the optimal position is determined based on a dual-network structure constructed by a strategy network and a reward network. According to the invention, intelligent planning of the positions of the accessories is realized, the planning efficiency of the accessories is improved, and individual requirements of different users on the accessories are met.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence applications, and particularly to a jaw position guiding device and an attachment planning method based on artificial intelligence. Background Art

[0002] In the design process of traditional jaw position guiding devices, the following technical steps are mainly adopted: First, a three-dimensional oral data of a patient is obtained using an oral scanner, including the morphological information and spatial position relationship of the upper and lower jaw teeth. Then, the doctor needs to manually mark the key feature points of each tooth in professional software, including anatomical landmarks such as the highest point of the tooth crown and the cervical margin line. After the marking is completed, the doctor plans the placement positions of the attachments one by one on the digital model according to the treatment target, and at the same time, the mutual relationship between the attachments and the interference with the opposing teeth need to be considered.

[0003] For the specific design of the attachments, the prior art adopts a preset standardized attachment library. The doctor selects a suitable basic shape from it and then makes simple size adjustments through software tools. After determining the approximate shape and position of the attachment, the system performs basic surface fitting calculations to try to make the bottom of the attachment match the tooth surface morphology. Finally, the doctor needs to manually check and fine-tune the position and morphology of each attachment to ensure that it meets the clinical use requirements.

[0004] The method in the related art has the defect of poor design efficiency, because the doctor needs to manually complete multiple links such as feature point marking and attachment position planning, resulting in the design time for a single case often exceeding 1 - 2 hours. This low efficiency directly affects the treatment efficiency of the clinic and the patient experience. There is also the defect of poor personalization, because the preset standardized attachment library is difficult to meet the personalized needs of complex cases. Especially when dealing with atypical tooth morphologies, the existing attachment morphologies often cannot achieve the best results. Summary of the Invention

[0005] The main object of the present invention is to provide a jaw position guiding device and an attachment planning method based on artificial intelligence to solve the deficiencies in the related art.

[0006] To achieve the above object, according to the first aspect of the present invention, a jaw position guiding device based on artificial intelligence is provided, including: a data preprocessing module for preprocessing the obtained three-dimensional oral data; an intelligent processing module for segmenting the three-dimensional oral data preprocessed and obtained from the data preprocessing module to obtain the segmentation data of each tooth; performing feature recognition based on the segmentation data of each tooth; an attachment position planning module for determining the optimal position of the attachment based on the features obtained from the intelligent processing module, wherein when determining the optimal position of the attachment, the optimal position is determined based on a dual network structure constructed by a policy network and a reward network.

[0007] Optionally, the device further includes:

[0008] An attachment form optimization module, configured to optimize the form of the attachment at the determined optimal position to determine the optimal attachment form. When performing form optimization, it is optimized through multi-level Boolean operations, including: when performing form optimization, cropping at the first level to determine the main shape of the tooth surface contour; at the second level, performing local feature optimization on the basis of the main shape to adjust the matching degree between the attachment and the tooth surface contour; at the third level, smoothing the edges of the attachment; wherein, during the optimization process, manufacturing process parameters are used as constraint conditions for optimization.

[0009] Optionally, the device further includes an evaluation module, configured to score at least one dimension information of the planned attachment, where the at least one dimension information includes the optimal position of the attachment, the mechanical property information obtained when determining the optimal position, and / or the matching accuracy between the attachment and the tooth surface obtained through the optimization.

[0010] Optionally, segmenting the preprocessed three-dimensional oral data obtained from the data preprocessing module includes: segmenting each tooth based on a three-dimensional point cloud segmentation network to obtain segmentation data of each tooth; using a feature recognition model to identify the specified features of each tooth's segmentation data, where the feature recognition model adopts a three-level cascaded feature extraction architecture based on the PointNet++ architecture; the three-level combined feature extraction architecture includes a local feature extraction layer, a regional feature fusion layer, and a global feature aggregation layer.

[0011] Optionally, a density perception module is introduced in the feature extraction layer, where the density perception module fuses local point cloud density features; an attention mechanism is integrated in the regional feature fusion layer to learn feature importance weights; when performing feature aggregation output in the global feature aggregation layer, it is output through a dual-head structure, where the dual-head structure includes a feature point coordinate prediction structure and a confidence evaluation structure.

[0012] Optionally, determining the optimal position of the attachment based on the features obtained from the intelligent processing module includes: determining the optimal position of the attachment based on a preset mechanical optimization model, where the initial position of the attachment is calculated based on the extracted specified features; based on the specified state information of the attachment at the initial position, iterative optimization is performed through the dual-network structure to determine the attachment at the final position.

[0013] Optionally, in each iteration, the policy network generates a position adjustment policy based on the specified state information of the attachment at the current position, and generates an adjustment action according to the position adjustment policy; execute the position adjustment action, and calculate the mechanical reward value based on the reward function, where the reward function includes evaluating the mechanical balance of the attachment, evaluating the matching degree between the predetermined movement direction and the actual movement direction of the attachment, and evaluating the resistance between the attachment and the tooth surface.

[0014] Optionally, in the iterative optimization process, optimize based on the set constraints until the information of the attachment at the optimal position is output when the convergence condition is met, where the set constraints are used to constrain the minimum distance between attachments in real time.

[0015] Optionally, after each position adjustment action is executed, transfer the specified state information of the attachment before and after adjustment, and the calculated mechanical reward value to a preset experience replay pool, so as to iteratively update the parameters of the policy network based on the information corresponding to each adjustment in the experience replay pool.

[0016] According to the second aspect of the present invention, an attachment planning method is provided, including: preprocessing the obtained three-dimensional oral data; segmenting the preprocessed three-dimensional oral data to obtain the segmentation data of each tooth; performing feature recognition based on the segmentation data of each tooth; determining the optimal position of the attachment based on the recognized features, where when determining the optimal position of the attachment, the optimal position is determined based on a dual-network structure constructed by a policy network and a reward network.

[0017] The jaw position guiding device and attachment planning method based on artificial intelligence in this embodiment, where the device includes a data preprocessing module for preprocessing the obtained three-dimensional oral data; an intelligent processing module for segmenting the preprocessed three-dimensional oral data obtained from the data preprocessing module to obtain the segmentation data of each tooth; performing feature recognition based on the segmentation data of each tooth; an attachment position planning module for determining the optimal position of the attachment based on the features obtained from the intelligent processing module, where when determining the optimal position of the attachment, the optimal position is determined based on a dual-network structure constructed by a policy network and a reward network. It realizes the intelligent planning of the attachment position, improves the planning efficiency of the attachment, and meets the personalized needs of different users for the attachment. Description of the Drawings

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 is a schematic structural diagram of a jaw position guiding device based on artificial intelligence according to an embodiment of the present invention;

[0020] Figure 2 is an application schematic diagram of a jaw position guiding device based on artificial intelligence according to an embodiment of the present invention;

[0021] Figure 3 is a flowchart of an accessory planning method according to an embodiment of the present invention. Detailed implementation manners

[0022] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so as to implement the embodiments of the present invention described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0025] According to an embodiment of the present invention, a jaw position guiding device based on artificial intelligence is provided, as Figure 1 shown, including a data preprocessing module for preprocessing the acquired three-dimensional oral data; an intelligent processing module for segmenting the three-dimensional oral data preprocessed and obtained from the data preprocessing module to obtain the segmentation data of each tooth; performing feature recognition based on the segmentation data of each tooth; an accessory position planning module for determining the optimal position of the accessory based on the features obtained from the intelligent processing module, wherein when determining the optimal position of the accessory, the optimal position is determined based on a dual network structure constructed by a policy network and a reward network.

[0026] In this embodiment, with reference to Figure 1 , when preprocessing is performed by the data preprocessing module, three-dimensional oral data of the patient is obtained from a high-precision oral scanner, where the scanning resolution requirement is not less than 0.05 mm, and a standard STL format file is output. The obtained original data is optimized by the preprocessing module, including three steps: mesh repair, noise removal, and data alignment. Among them, the mesh filling algorithm is used for mesh repair to automatically repair the breakage and holes in the scanned data; the bilateral filtering algorithm is used for noise removal to effectively remove noise while retaining the edge features; and data alignment is based on the feature point registration technology to achieve the precise spatial alignment of the upper and lower jaw data.

[0027] In the intelligent processing module, an improved three-dimensional point cloud segmentation network structure can be used to achieve the precise segmentation of each tooth through the encoder-decoder architecture. After segmentation, an improved PointNet++ network can be used to identify key feature points, including anatomical landmarks such as the highest point of the tooth crown, the gingival margin line, and the adjacent contact points, and the accuracy of feature point identification is controlled within 0.1 mm.

[0028] In the attachment position planning module, the initial position of the attachment can be calculated based on the identified features such as the highest point of the tooth crown and the gingival margin line feature, and the initial height can be set at 4 mm from the gingival margin line. On this basis, a dual-network structure is constructed by using a policy network and a reward network, and the final position is determined through iterative optimization.

[0029] In this embodiment, with a modular design architecture, data interaction between modules is carried out through standardized interfaces. The system shows significant performance advantages in practical applications: the design time is shortened from the traditional 1-2 hours to within 10 minutes, the automation level reaches more than 90%, the attachment matching accuracy is improved to 0.05 mm, and the position optimization accuracy rate exceeds 95%. The actual clinical application data shows that using this system can increase the design efficiency by 300%, reduce the attachment shedding rate by 50%, and shorten the treatment cycle by 20%, fully demonstrating the technological advancement and practical value of the present invention.

[0030] Furthermore, in the dual-network structure, the policy network is responsible for generating position adjustment strategies to guide the movement of the attachment. The role of the reward network is to evaluate the value of the action in the current state, that is, by calculating the optimization degree of the current attachment position, it provides feedback to the policy network.

[0031] As an alternative implementation of this embodiment, segmenting the preprocessed three-dimensional oral data obtained from the data preprocessing module includes: segmenting each tooth based on a three-dimensional point cloud segmentation network to obtain segmentation data for each tooth; using a feature recognition model to recognize the segmentation data of each tooth to recognize the specified features of each tooth, where the feature recognition model adopts a three-level cascaded feature extraction architecture based on the PointNet++ architecture; the three-level combined feature extraction architecture includes a local feature extraction layer, a regional feature fusion layer, and a global feature aggregation layer.

[0032] As an alternative implementation of this embodiment, a density perception module is introduced in the feature extraction layer, where the density perception module fuses local point cloud density features; an attention mechanism is integrated in the regional feature fusion layer to learn feature importance weights; when performing feature aggregation output in the global feature aggregation layer, it is output through a dual-head structure, where the dual-head structure includes a feature point coordinate prediction structure and a confidence evaluation structure.

[0033] In the above alternative implementation, in this alternative implementation, the network adopts a three-level cascaded feature extraction architecture based on the original PointNet++ architecture, including a local feature extraction layer (sampling radius 0.2), a regional feature fusion layer (sampling radius 0.4), and a global feature aggregation layer. In terms of technological innovation, first, a density perception module is introduced in the feature extraction layer, enhancing the network's adaptability to point cloud density changes by fusing local point cloud density features; second, an attention mechanism is integrated in the regional feature fusion layer, improving the recognition accuracy of key anatomical landmarks by learning feature importance weights; finally, a dual-head structure is designed at the output end for feature point coordinate prediction and confidence evaluation respectively, enhancing the reliability of feature point positioning. Through these improvements, the network can achieve a feature point recognition accuracy controlled within 0.1 mm. Especially when dealing with continuous features such as the cervical margin line, the combination of multi-scale feature fusion and the attention mechanism can effectively capture local-global relationships, ensuring the spatial continuity and morphological consistency of the recognition results.

[0034] In this alternative implementation, an improved three-dimensional U-Net network structure is used for feature point recognition and segmentation. Through data augmentation and model training methods, the accuracy of recognition and segmentation is improved, and at the same time, an optimization strategy is applied to further improve the quality of the segmentation results.

[0035] As an alternative implementation of this embodiment, the device further includes: an attachment form optimization module for optimizing the form of the attachment at the determined optimal position. When performing form optimization, it is optimized through multi-level Boolean operations, including: when performing form optimization, cropping is performed at the first level to determine the main shape of the attachment; local feature optimization is performed at the second level to adjust the matching degree between the attachment and the tooth surface; edge smoothing processing is performed on the attachment at the third level; and during the optimization process, manufacturing process parameters are used as constraint conditions for optimization.

[0036] In this alternative implementation, in the form optimization design stage, a parametric design method is adopted, and the attachment form is described based on B-spline surface technology. First, a basic form is generated according to the optimized position information, and then refined processing is performed through multi-level Boolean operations. The Boolean operation process is divided into three levels: the first level performs large-volume cropping to determine the main shape of the attachment; the second level performs local feature optimization to adjust the matching degree with the tooth surface; the third level performs edge smoothing processing to improve the wearing comfort. Further, the first level performs large-volume cropping based on the tooth surface contour, the second level performs fine matching based on local features, and the third level performs edge smoothing processing and optimizes the transition area. Optimization parameters and threshold adjustment methods between each level are preset. Exemplarily, the optimization parameters between each level may include cropping tolerance, matching degree threshold, and fillet radius.

[0037] Further, throughout the process, the system always considers the constraint conditions of the manufacturing process to ensure the realizability of the design result. Exemplarily, constraint conditions can be constructed based on the parameters of the manufacturing process, and the manufacturing process parameters include but are not limited to chamfer curvature, etc.

[0038] In this alternative implementation, a multi-level Boolean operation method is used for form optimization, which improves the matching accuracy between the attachment and the tooth surface to 0.02 mm, a 60% increase compared to the prior art, and significantly reduces the detachment rate of the attachment. It overcomes the defect in the related art that using a simple surface fitting algorithm cannot fully consider the local geometric features of the tooth surface, resulting in an unsatisfactory fitting accuracy between the attachment base and the tooth surface and possibly increasing the attachment detachment rate.

[0039] Reference Figure 2 Schematically shows a schematic diagram for determining the position and form of the attachment, including automatic segmentation and feature recognition of intraoral scan data, then determining the attachment position through a reinforcement learning method, and finally obtaining the final attachment through multi-level Boolean operations and a smoothing algorithm.

[0040] As an alternative implementation of this embodiment, the device further includes an evaluation module for scoring at least one dimension information of the planned attachment, where the at least one dimension information includes the optimal position of the attachment, the mechanical property information obtained when determining the optimal position, and / or the matching accuracy between the attachment and the tooth surface obtained through the optimization.

[0041] In this alternative implementation, during quality evaluation, a comprehensive evaluation of the design result is performed. The evaluation indicators include multiple dimensions such as the attachment matching accuracy (required ≤ 0.05 mm), the position rationality score (0 - 100 points), and the mechanical property evaluation. If the evaluation result does not reach the preset threshold, the system will automatically generate optimization suggestions and return them to the corresponding module for adjustment. The entire design process forms a closed-loop feedback mechanism, continuously optimizing until the quality requirements are met.

[0042] As an alternative implementation of this embodiment, determining the optimal position of the attachment based on the features obtained from the intelligent processing module includes: determining the optimal position of the attachment based on a preset mechanical optimization model, where the initial position of the attachment is calculated based on the extracted specified features; and based on the specified state information of the attachment at the initial position, iterative optimization is performed through the dual network structure to determine the attachment at the final position.

[0043] In this alternative implementation, the mechanical analysis model based on deep reinforcement learning realizes the global optimization of the attachment position, improving the uniformity of the force distribution by 50% and effectively shortening the treatment cycle by 20%.

[0044] As an alternative implementation of this embodiment, in each iteration process, the policy network generates a position adjustment policy based on the specified state information of the attachment at the current position, and generates an adjustment action according to the position adjustment policy; the position adjustment action is executed, and a mechanical reward value is calculated based on the reward function, where the reward function includes evaluating the mechanical balance of the attachment, evaluating the matching degree between the predetermined movement direction and the actual movement direction of the attachment, and evaluating the resistance magnitude between the attachment and the tooth surface, where the reward network evaluates the mechanical properties of the attachment at the position where it is located and uses the evaluation information as the basis for calculating the mechanical reward value.

[0045] In this alternative implementation, the specified state information of the attachment at the current position may include 3D position, direction such as angle, and mechanical features such as mechanical specified parameters, and the policy network generates a position adjustment action based on the specified state information. The adjustment is executed and the mechanical reward value is calculated, where the reward function comprehensively considers the mechanical balance (weight 2.0), the movement direction matching degree (weight 1.5), and the resistance magnitude (weight 1.0).

[0046] The degree of motion direction matching refers to the degree of coincidence between the predetermined motion direction of the attachment and the actual motion direction. The reward network is mainly used to evaluate and feedback the mechanical performance of the attachment at the current position, including the stability of the attachment, mechanical balance, etc., and provide a basis for calculating rewards for the policy network.

[0047] As an alternative implementation of this embodiment, during the iterative optimization process, optimization is performed based on set constraints until the information of the attachment at the optimal position is output when the convergence condition is met, where the set constraints are used to constrain the minimum distance between attachments in real time.

[0048] In this alternative implementation, it overcomes the problem in the related art of lacking the overall optimization ability for the mutual relationship between multiple attachments, which is prone to problems such as uneven force and unreasonable moment distribution, affecting the treatment effect.

[0049] As an alternative implementation of this embodiment, after each position adjustment action is executed, the transfer of the specified state information before and after the attachment adjustment and the calculated mechanical reward value are stored in a preset experience replay pool, so as to iteratively update the parameters of the policy network based on the information corresponding to each adjustment in the experience replay pool.

[0050] In the above alternative implementation, in reinforcement learning, after the state transition, that is, after the transition from one state to another state, the effect of the action can be evaluated, and this effect can be evaluated based on the specified state information, and the policy is continuously optimized using the evaluated effect. Exemplarily, the update of the policy network parameters can be determined based on the received reward value and the information obtained from the state transition.

[0051] Exemplarily, after each adjustment action is executed, the state changes of the attachment before and after the adjustment, such as position, angle, and mechanical characteristics, etc., and their corresponding reward information are stored in the experience replay pool together. These recorded information can be sampled and learned, and by comparing the effects brought by the state transition, the parameters of the policy network are continuously iteratively updated, so as to more accurately select the optimal adjustment action in subsequent decisions and improve the overall optimization effect, and update the policy network parameters.

[0052] This embodiment is based on the deep reinforcement learning method. First, the system calculates the initial position of the attachment according to the characteristics of the highest point of the crown and the gingival margin line, and the initial height is set at 4 mm from the gingival margin line. On this basis, a dual-network structure is constructed by using a policy network and a reward network, and the final position is determined through iterative optimization. Each iteration includes four key steps: action selection, state transition, reward calculation, and policy update. During the whole process, the system checks for interference between attachments in real time by setting a minimum spacing constraint (2 mm), and outputs the optimal position plan when the convergence condition is met. In this way, it overcomes the defect that the preset standardized attachment library is difficult to meet the personalized needs of complex cases, especially when dealing with atypical tooth shapes, and the existing attachment shapes often cannot achieve the best treatment effect.

[0053] According to an embodiment of the present invention, there is also provided an attachment planning method, including preprocessing the acquired three-dimensional oral data; segmenting the preprocessed three-dimensional oral data to obtain the segmentation data of each tooth; performing feature recognition based on the segmentation data of each tooth; and determining the optimal position of the attachment based on the recognized features. When determining the optimal position of the attachment, the optimal position is determined based on a dual-network structure constructed by a policy network and a reward network.

[0054] As an optional implementation manner of this embodiment, the method further includes optimizing the shape of the attachment whose optimal position is determined to determine the optimal attachment shape. When performing shape optimization, it is optimized through multi-level Boolean operations, including: when performing shape optimization, cropping is performed at the first level to determine the main shape of the tooth surface contour; local feature optimization is performed on the basis of the main shape at the second level to adjust the matching degree between the attachment and the tooth surface contour; and edge smoothing processing is performed on the attachment at the third level. During the optimization process, the manufacturing process parameters are used as constraint conditions for optimization.

[0055] As an optional implementation manner of this embodiment, at least one dimension information of the planned attachment is scored, where the at least one dimension information includes the optimal position of the attachment, the mechanical property information obtained when determining the optimal position, and / or the matching accuracy between the attachment and the tooth surface obtained through the optimization.

[0056] As an optional implementation manner of this embodiment, segmenting the preprocessed three-dimensional oral data obtained from the data preprocessing module includes: segmenting each tooth based on a three-dimensional point cloud segmentation network to obtain the segmentation data of each tooth; and using a feature recognition model to identify the segmentation data of each tooth to identify the specified features of each tooth, where the feature recognition model adopts a three-level cascaded feature extraction architecture based on the PointNet++ architecture; the three-level combined feature extraction architecture includes a local feature extraction layer, a regional feature fusion layer, and a global feature aggregation layer.

[0057] As an optional implementation manner of this embodiment, a density perception module is introduced into the feature extraction layer, wherein the density perception module fuses local point cloud density features; an attention mechanism is integrated in the region feature fusion layer to learn feature importance weights; when feature aggregation output is performed in the global feature aggregation layer, it is output through a dual-head structure, wherein the dual-head structure includes a feature point coordinate prediction structure and a confidence evaluation structure.

[0058] As an optional implementation manner of this embodiment, determining the optimal position of the attachment based on the features obtained from the intelligent processing module includes: determining the optimal position of the attachment based on a preset mechanical optimization model, wherein the initial position of the attachment is calculated based on the extracted specified features; based on the specified state information of the attachment at the initial position, iterative optimization is performed through the dual-network structure to determine the attachment at the final position.

[0059] As an optional implementation manner of this embodiment, in each iteration process, the policy network generates a position adjustment policy based on the specified state information of the attachment at the current position, so as to generate an adjustment action according to the position adjustment policy; execute the position adjustment action, and calculate a mechanical reward value based on the reward function, wherein the reward function includes evaluating the mechanical balance of the attachment, evaluating the matching degree between the predetermined movement direction and the actual movement direction of the attachment, and evaluating the resistance magnitude between the attachment and the tooth surface, wherein the reward network evaluates the mechanical performance of the attachment at the position where it is located and uses the evaluation information as the basis for calculating the mechanical reward value.

[0060] As an optional implementation manner of this embodiment, in the iterative optimization process, optimization is performed based on the set constraints until the information of the attachment at the optimal position is output when the convergence condition is satisfied, wherein the set constraints are used to continuously constrain the minimum distance between attachments.

[0061] As an optional implementation manner of this embodiment, after each position adjustment action is executed, the transfer of the specified state information of the attachment before and after adjustment and the calculated mechanical reward value are stored in a preset experience replay pool, so as to iteratively update the parameters of the policy network based on the information corresponding to each adjustment in the experience replay pool.

[0062] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A jaw position guidance device based on artificial intelligence, characterized in that: include: A data preprocessing module, used for preprocessing the acquired three-dimensional oral data; An intelligent processing module, used for segmenting the three-dimensional oral data obtained through preprocessing by the data preprocessing module to obtain segmentation data of each tooth; Perform feature recognition based on the segmentation data of each tooth; The attachment position planning module is used to determine the optimal position of the attachment based on the features obtained from the intelligent processing module, wherein when determining the optimal position of the attachment, the optimal position is determined based on a dual network structure constructed by a policy network and a reward network.

2. The artificial intelligence-based jaw position guidance device according to claim 1, characterized in that: The device also includes: The attachment morphology optimization module is used to optimize the morphology of the attachments with the optimal position determined to determine the optimal attachment morphology. When performing morphology optimization, optimization is performed through multi-level Boolean operations, including: when performing morphology optimization, cutting is performed at the first level to determine the main shape of the tooth surface contour; at the second level, local feature optimization is performed on the basis of the main shape to adjust the matching degree between the attachment and the tooth surface contour; at the third level, the edges of the attachment are rounded; and during the optimization process, manufacturing process parameters are optimized as constraints.

3. The artificial intelligence-based jaw guidance device according to claim 2, characterized in that: The device also includes an evaluation module for scoring at least one dimensional information of the planned attachment, wherein the at least one dimensional information includes the optimal position of the attachment, the mechanical property information obtained when determining the optimal position, and / or the matching accuracy of the attachment and the tooth surface obtained by the optimization.

4. The artificial intelligence-based jaw position guidance device according to claim 1, characterized in that: Segmenting the three-dimensional oral cavity data obtained by preprocessing from the data preprocessing module includes: Segment each tooth based on the 3D point cloud segmentation network to obtain segmentation data of each tooth; The segmentation data of each tooth is identified by a feature recognition model to identify the specified features of each tooth, wherein the feature recognition model adopts a three-layer cascaded feature extraction architecture based on the PointNet++ architecture; the three-layer joint feature extraction architecture includes a local feature extraction layer, a regional feature fusion layer and a global feature aggregation layer.

5. The artificial intelligence-based jaw position guidance device according to claim 4, characterized in that: Introducing a density perception module into the feature extraction layer, wherein the density perception module fuses local point cloud density features; Integrating an attention mechanism in the regional feature fusion layer to learn feature importance weights; When the global feature aggregation layer performs feature aggregation output, the output is performed through a dual-head structure, wherein the dual-head structure includes a feature point coordinate prediction structure and a confidence assessment structure.

6. The artificial intelligence-based jaw guidance device according to claim 1, characterized in that: Determining an optimal position of an attachment based on the features obtained from the intelligent processing module includes: Determining the optimal position of the attachment based on a preset mechanical optimization model, wherein the initial position of the attachment is calculated based on the extracted specified features; Based on the specified state information of the attachment at the initial position, iterative optimization is performed through the dual network structure to determine the attachment at the final position.

7. The artificial intelligence-based jaw position guidance device according to claim 6, characterized in that: During each iteration, the policy network generates a position adjustment policy based on the specified state information of the attachment at the current position, so as to generate an adjustment action according to the position adjustment policy; Execute the position adjustment action and calculate the mechanical reward value based on the reward function, wherein the reward function includes evaluating the mechanical balance of the attachment, evaluating the matching degree between the predetermined movement direction and the actual movement direction of the attachment, and evaluating the resistance between the attachment and the tooth surface, wherein the reward network evaluates the mechanical properties of the attachment at the position, and uses the evaluation information as the basis for calculating the mechanical reward value.

8. The artificial intelligence-based jaw position guiding device according to claim 7, characterized in that: In the iterative optimization process, optimization is performed based on set constraints until the information of the attachment at the optimal position is output when the convergence condition is met, wherein the set constraints are used to constrain the minimum spacing between attachments in real time.

9. The artificial intelligence-based jaw position guiding device according to claim 8, characterized in that: After performing each position adjustment action, the change in the specified state information before and after the attachment adjustment and the calculated mechanical reward value are stored in a preset experience replay pool to iteratively update the parameters of the strategy network based on the information corresponding to each adjustment in the experience replay pool.

10. An attachment planning method, characterized in that: include: Preprocessing the acquired three-dimensional oral data; Segment the three-dimensional oral data obtained by preprocessing to obtain segmentation data of each tooth; Perform feature recognition based on the segmentation data of each tooth; The optimal position of the attachment is determined based on the identified features, wherein when determining the optimal position of the attachment, the optimal position is determined based on a dual network structure constructed by a policy network and a reward network.

Images