Orthodontic treatment scheme intelligent design and prediction system driven by artificial intelligence

Abstract

The invention relates to the field of medical information technology and artificial intelligence, and provides an artificial intelligence-driven orthodontic treatment scheme intelligent design and prediction system, which comprises an oral data acquisition module, an intelligent planning module, a biomechanical simulation module, a scheme optimization module and a prediction generation module. The system generates a candidate movement scheme set through a deep reinforcement learning algorithm, verifies the scheme safety through biomechanical simulation, selects an optimal scheme through multi-target collaborative optimization, and generates a predicted three-dimensional oral cavity model. Multi-objective optimization of the occlusion relation, the aesthetic effect and the treatment cycle is achieved, and an intelligent solution is provided for complex malocclusion.

Description

Technical Field

[0001] This invention relates to the fields of medical information technology and artificial intelligence, and in particular to an artificial intelligence-driven intelligent design and prediction system for orthodontic treatment plans. Background Technology

[0002] Currently, digital orthodontic technology is widely used in clinical practice. For example, Chinese patent application CN112690913A discloses a method and system for generating orthodontic plans. This technical solution acquires a user's scalp image and a 3D oral model, obtains the user's first characteristic parameters based on the scalp image, determines whether orthodontic treatment is needed, then generates an orthodontic plan based on the orthodontic type, and generates a predictive 3D oral model. This technical solution overcomes the shortcomings of using only two-dimensional parameter analysis and plays a role in predicting orthodontic effects.

[0003] However, the aforementioned technical solutions have the following shortcomings in practical applications: They primarily rely on pre-defined orthodontic types and fixed planning processes, lacking the ability to intelligently optimize tooth movement paths. When dealing with complex malocclusions, due to the involvement of multiple teeth in coordinated movement, dynamic adjustment of occlusal relationships, comprehensive consideration of aesthetic effects, and reasonable planning of treatment cycles, traditional planning methods struggle to find the optimal balance among multiple treatment goals. Especially in determining the specific tooth movement paths and parameters, doctors often need to make repeated adjustments based on experience, resulting in low efficiency and outcomes dependent on the doctor's individual experience level.

[0004] Furthermore, existing orthodontic solutions typically do not adequately consider the biomechanical response of periodontal tissues during tooth movement when generating orthodontic plans. Tooth movement is achieved by applying orthodontic forces to the teeth, causing tissue remodeling of the periodontal ligament and alveolar bone. If the applied orthodontic force is too large, it may lead to excessive stress on the periodontal tissues, causing complications such as root resorption and periodontal inflammation; if the orthodontic force is too small, it may lead to prolonged treatment time and affect treatment efficiency. Therefore, there is an urgent need for an orthodontic treatment plan design system that can intelligently optimize the tooth movement path while taking into account biomechanical safety and multi-objective collaborative optimization. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide an AI-driven intelligent design and prediction system for orthodontic treatment plans. By introducing deep reinforcement learning algorithms, the system achieves intelligent optimization of tooth movement paths and combines a biomechanical simulation module to verify and adjust the movement plan for safety. This allows the system to find the optimal balance between occlusal relationship, aesthetic effect, and treatment cycle, significantly improving the personalization level and clinical safety of orthodontic treatment plans.

[0006] To achieve the above objectives, this invention provides an artificial intelligence-driven intelligent design and prediction system for orthodontic treatment plans, comprising:

[0007] The oral data acquisition module is used to receive the patient's oral scan data, perform three-dimensional reconstruction processing on the oral scan data to obtain an initial three-dimensional oral model, the initial three-dimensional oral model including a tooth model, a gingival model and a periodontal tissue model; and acquire the tooth feature information corresponding to the initial three-dimensional oral model, the tooth feature information including tooth position information, occlusal relationship information and facial contour information.

[0008] The intelligent planning module, connected to the oral data acquisition module, is used to determine the target tooth and treatment target if the tooth feature information meets the abnormal feature range. The treatment target includes the target occlusal relationship, the target aesthetic effect, and the target treatment cycle. Based on a deep reinforcement learning algorithm, a set of candidate movement schemes is generated. Each candidate movement scheme in the set of candidate movement schemes includes the movement path and movement parameters of the target tooth in multiple treatment stages.

[0009] A biomechanical simulation module, connected to the intelligent planning module, is used to simulate the stress distribution of periodontal tissues during the movement of the target tooth for each candidate movement scheme in the candidate movement scheme set, and obtain the maximum stress value corresponding to the stress distribution; if the maximum stress value does not meet the stress safety range, the candidate movement scheme is adjusted.

[0010] The optimal solution module, connected to the intelligent planning module and the biomechanical simulation module, is used to calculate the comprehensive reward value corresponding to each candidate movement solution in the adjusted candidate movement solution set. The comprehensive reward value is calculated based on the occlusal relationship score, aesthetic effect score, and treatment cycle score. If the comprehensive reward value meets the reward threshold range, the candidate movement solution with the highest comprehensive reward value is selected as the optimal movement solution.

[0011] The prediction generation module, connected to the scheme selection module, is used to simulate the movement of the initial three-dimensional oral model according to the optimal movement scheme to generate a predicted three-dimensional oral model, which displays the oral state after treatment.

[0012] In one possible implementation, the oral data acquisition module is specifically used to preprocess the oral scan data to remove noise points and redundant data; perform three-dimensional reconstruction on the processed oral scan data based on a point cloud registration algorithm to generate the initial three-dimensional oral model; perform semantic segmentation on the initial three-dimensional oral model to identify the tooth region, gingival region and periodontal tissue region; and extract the tooth feature information from the initial three-dimensional oral model.

[0013] In one possible implementation, the intelligent planning module includes an anomaly identification submodule, a target setting submodule, and a reinforcement learning submodule. The reinforcement learning submodule establishes a reinforcement learning environment and uses the deep reinforcement learning algorithm to explore and optimize within the reinforcement learning environment to generate the candidate movement scheme set.

[0014] In one possible implementation, the biomechanical simulation module establishes a finite element model of teeth and periodontal tissues, simulates the stress and displacement distribution of periodontal tissues under orthodontic force based on the finite element model, calculates the maximum stress value, and adjusts the candidate movement scheme according to the relationship between the maximum stress value and the stress safety range.

[0015] In one possible implementation, the scheme optimization module calculates the occlusal relationship score, the aesthetic effect score, and the treatment cycle score respectively using an occlusal relationship evaluation algorithm, an aesthetic effect evaluation algorithm, and a treatment cycle evaluation algorithm, and then weights and sums the three according to a preset weight coefficient to obtain the comprehensive reward value.

[0016] The beneficial effects of this invention include:

[0017] First, this invention achieves intelligent optimization of tooth movement paths by introducing deep reinforcement learning algorithms. Compared with traditional methods that rely on preset rules and doctor experience, deep reinforcement learning algorithms can explore extensively in the state space, automatically discovering the optimal balance between occlusal relationship, aesthetic effect, and treatment cycle, significantly improving the personalization level of orthodontic plans and treatment outcomes.

[0018] Secondly, this invention innovatively introduces a biomechanical simulation module to verify the stress distribution of periodontal tissues in real time during the treatment plan generation process, ensuring biomechanical safety during tooth movement. By setting a safe stress range and automatically adjusting the plan if it exceeds the range, complications such as root resorption and periodontal inflammation caused by excessive orthodontic force are effectively avoided, thus improving the clinical safety of treatment.

[0019] Third, this invention achieves multi-objective collaborative optimization through a scheme optimization module. The comprehensive reward value comprehensively considers three key dimensions: occlusal relationship, aesthetic effect, and treatment cycle, and supports dynamic adjustment of weight coefficients based on patient preferences. It can flexibly balance different treatment goals, providing patients with highly personalized treatment plans and improving patient satisfaction and treatment compliance.

[0020] Fourth, this invention achieves real-time updates of the treatment plan during the treatment process through a dynamic adjustment module. During treatment, the patient's actual oral data is periodically collected and compared with the expected state. If the deviation is significant, the treatment plan is automatically regenerated, ensuring that the treatment plan always matches the patient's actual condition, further improving the accuracy and success rate of treatment. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall architecture of the AI-driven intelligent design and prediction system for orthodontic treatment plans provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the structure of the intelligent planning module provided in an embodiment of the present invention.

[0023] Figure 3 The flowchart illustrates the state-action-reward interaction of the reinforcement learning environment provided in this embodiment of the invention.

[0024] Figure 4 A flowchart illustrating the workflow of the biomechanical simulation module provided in this embodiment of the invention.

[0025] Figure 5 A flowchart illustrating the scoring calculation process of the preferred solution module provided in this embodiment of the invention.

[0026] Figure 6 A system workflow diagram provided for embodiments of the present invention. Detailed Implementation

[0027] Please refer to the attached document. Figure 1-6 To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] It should be noted that the terms "first," "second," etc., used in the specification and claims of this invention 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 where appropriate so that the embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein.

[0029] like Figure 1As shown, this embodiment of the invention provides an artificial intelligence-driven intelligent design and prediction system for orthodontic treatment plans. The system includes an oral data acquisition module 1, an intelligent planning module 2, a biomechanical simulation module 3, a plan optimization module 4, and a prediction generation module 5. These modules are connected via data interfaces to form a complete orthodontic treatment plan design and prediction process.

[0030] The oral cavity data acquisition module 1 receives the patient's oral cavity scan data, which can be acquired through an intraoral scanner, cone-beam computed tomography (CBCT), or other three-dimensional scanning equipment. In one possible implementation, the oral cavity scan data is in point cloud format, containing the three-dimensional spatial coordinate information of the patient's oral cavity. This module first preprocesses the oral cavity scan data, including removing noise points generated during the scanning process, eliminating data redundancy, and filling in scanning blind areas, to improve the accuracy of subsequent processing.

[0031] Then, the module performs 3D reconstruction of the processed oral cavity scan data based on a point cloud registration algorithm. Specifically, the module uses the Iterative Closest Point (ICP) algorithm or its variant to register and fuse point cloud data acquired from multiple angles, generating a complete initial 3D oral cavity model. The initial 3D oral cavity model includes a tooth model, a gingival model, and a periodontal tissue model, which can realistically reflect the 3D geometry of the patient's oral cavity.

[0032] Furthermore, this module performs semantic segmentation on the initial 3D oral cavity model to identify tooth regions, gingival regions, and periodontal tissue regions. In one possible implementation, this module employs a deep learning-based semantic segmentation network, such as U-Net or PointNet++, to classify the point cloud data point by point, labeling each point as belonging to a tooth, gingiva, or periodontal tissue. Through semantic segmentation, this module can accurately locate the spatial position of each tooth, laying the foundation for subsequent feature extraction.

[0033] Finally, the module extracts tooth feature information from the initial 3D oral model. This tooth feature information includes tooth position information, occlusal relationship information, and facial contour information. For example, the tooth position information includes geometric parameters such as the 3D coordinates of each tooth, the direction of the tooth's long axis, and the crown inclination angle; the occlusal relationship information includes the position of the occlusal contact points of the upper and lower teeth, the number of occlusal contact points, and the distribution of occlusal forces; and the facial contour information includes aesthetically relevant indicators such as the facial midline position, the shape of the dental arch curve, and smile curve parameters.

[0034] The intelligent planning module 2 is connected to the oral data acquisition module 1, receives the tooth feature information, and generates a set of candidate repositioning schemes based on a deep reinforcement learning algorithm. For example... Figure 2As shown, the intelligent planning module 2 includes an anomaly identification submodule 21, a target setting submodule 22, and a reinforcement learning submodule 23.

[0035] The anomaly identification submodule 21 compares the tooth feature information with the normal tooth feature range to determine whether the tooth feature information conforms to the abnormal feature range. In one possible implementation, the abnormal feature range includes the range of tooth misalignment deviation, the range of abnormal occlusion, and the range of facial symmetry deviation. The range of tooth misalignment deviation is a single tooth positional deviation greater than 2 mm or an angular deviation greater than 10 degrees; the range of abnormal occlusion is the number of occlusal contact points of the upper and lower teeth less than a preset contact point threshold (e.g., 12 contact points), or the unevenness of occlusal force distribution greater than a preset unevenness threshold (e.g., coefficient of variation greater than 0.3); the range of facial symmetry deviation is the degree of deviation of the facial midline greater than 3 mm. If the tooth feature information conforms to the abnormal feature range, the submodule determines that the patient needs orthodontic treatment and transmits the relevant information to the target setting submodule 22.

[0036] The target setting submodule 22 determines the target tooth and treatment target based on the degree of deviation between the tooth feature information and the normal tooth feature range. The target tooth is the tooth that needs to be repositioned, and may include one or more teeth. The treatment target includes the target occlusal relationship, the target aesthetic effect, and the target treatment period. In one possible implementation, the target occlusal relationship is defined as having more than 18 occlusal contact points between the upper and lower teeth, and a occlusal force distribution unevenness of less than 0.2; the target aesthetic effect is defined as a facial midline deviation of less than 1 mm and a tooth alignment score greater than 90 points (out of 100); the target treatment period is estimated based on the tooth movement distance and the patient's age, preferably controlled between 12 and 24 months.

[0037] The reinforcement learning submodule 23 establishes a reinforcement learning environment and explores and optimizes within this environment using deep reinforcement learning algorithms to generate a set of candidate movement schemes. For example... Figure 3 As shown, the reinforcement learning environment includes three core elements: state space, action space, and reward function.

[0038] The state space describes the current spatial position and occlusal relationship of the teeth. In one possible implementation, the state space includes the three-dimensional coordinate vectors of all target teeth, the major axis direction vector of the teeth, the distance vector between adjacent teeth, and the occlusal contact point vectors of the upper and lower teeth, forming a high-dimensional state vector.

[0039] ,

[0040] in For the target number of teeth, This represents the number of adjacent spacings. This represents the number of contact points.

[0041] The motion space defines the movement operations that can be performed on the teeth. In one possible implementation, the motion space includes six basic movement types: tilting movement, overall movement, torque movement, rotational movement, depressing movement, and elongating movement. Each movement type corresponds to a different torque and force application method. For each target tooth, the submodule can select a combination of one or more movement types and specify parameters such as movement distance, movement angle, and movement direction. For example, overall movement corresponds to applying a horizontal force to translate the tooth in the horizontal direction, with a movement distance ranging from 0.1 to 3 mm; torque movement corresponds to applying a torque to rotate the tooth root, with a movement angle ranging from 1 to 30 degrees.

[0042] The reward function evaluates the degree of improvement in the state after performing an action. In one possible implementation, the reward function... It is obtained by weighted summation of the three sub-rewards:

[0043] ,

[0044] in, , , These are the weighting coefficients. Assessment status The degree of improvement in occlusal relationship, Assess the degree of improvement in aesthetic effect. According to the action The corresponding time cost is given a negative reward (the longer the time, the greater the negative reward).

[0045] This submodule employs deep reinforcement learning algorithms, such as Deep Q-Network (DQN), Proximal Policy Optimization (PPO), or the Actor-Critic algorithm, and is trained in the reinforcement learning environment. Specifically, this submodule initializes a neural network policy that takes the current state as input and outputs the probability distribution or Q-value of each selectable action. Then, the submodule selects an action according to the policy, observes the new state and the reward obtained after executing the action, and updates the neural network parameters based on the reward feedback, optimizing the policy towards obtaining higher cumulative rewards.

[0046] After extensive iterative training, this submodule acquires an optimized strategy that, given an initial state, progressively selects action sequences to move the tooth from an abnormal state to a state close to the target state. It should be noted that, due to the exploratory nature of reinforcement learning algorithms, this submodule generates multiple different candidate movement schemes. These schemes may have different emphases on occlusal relationships, aesthetic effects, and treatment cycles, forming a set of candidate movement schemes.

[0047] Each candidate repositioning scheme includes the repositioning path and parameters of the target tooth across multiple treatment phases. For example, a candidate repositioning scheme might include: a first phase of overall repositioning of the maxillary first premolar, moving it posteriorly by 1.5 mm; a second phase of torque repositioning of the maxillary central incisor, adjusting the root angle by 8 degrees; a third phase of rotational repositioning of the mandibular lateral incisor, rotating it by 15 degrees; and so on, until the treatment goal is achieved. The number of treatment phases depends on the complexity of the tooth repositioning, typically ranging from 10 to 40 phases.

[0048] The biomechanical simulation module 3 is connected to the intelligent planning module 2, receives a set of candidate movement schemes, and performs biomechanical simulations on each candidate movement scheme to verify its safety. For example... Figure 4 As shown, the workflow of this module includes four steps: establishing a finite element model, applying orthogonal force, calculating stress distribution, and adjusting the scheme.

[0049] First, this module establishes a finite element model of the teeth and periodontal tissues. The finite element model includes tooth elements, periodontal ligament elements, and alveolar bone elements. In one possible implementation, based on the initial three-dimensional oral cavity model, the module meshes the teeth, periodontal ligament, and alveolar bone, generating a finite element mesh composed of a large number of tetrahedral or hexahedral elements. The tooth elements adopt an isotropic elastic material model with an elastic modulus set to 20,000 MPa and a Poisson's ratio set to 0.3; the periodontal ligament elements adopt a hyperelastic material model to simulate their nonlinear mechanical properties, with an elastic modulus set to 0.68 MPa; the alveolar bone elements adopt an anisotropic elastic material model, with elastic moduli varying in different directions, ranging from 1,000 to 14,000 MPa.

[0050] Then, the module determines the magnitude and direction of the orthodontic force applied to the target tooth based on the candidate movement schemes. Different movement types correspond to different force application methods. For example, overall movement requires applying a horizontal force on the crown, the magnitude of which is determined based on the movement distance and tooth mass, preferably ranging from 0.5 to 2 Newtons; torque movement requires applying opposite forces to the crown and root to form a torque, with the torque ranging from 5 to 20 Newton-millimeters. The module applies these orthodontic forces as boundary conditions to the corresponding nodes of the finite element model.

[0051] Next, based on the finite element analysis method, this module solves for the stress and displacement distribution of the periodontal tissues under the orthodontic force. Specifically, this module solves the finite element equilibrium equations. ,in For the overall stiffness matrix, Let be the nodal displacement vector. This represents the nodal force vector. After obtaining the displacement distribution, this module further calculates the stress distribution of each element, including principal stresses, shear stresses, and equivalent stresses. This module pays particular attention to the stress distribution of the periodontal ligament, because the stress level of the periodontal ligament directly affects the bone remodeling process and treatment safety.

[0052] This module calculates the maximum stress value corresponding to the stress distribution, where the maximum stress value is the maximum equivalent stress of all units in the periodontal ligament. It is understood that if the maximum stress value is too high, it may lead to adverse reactions such as periodontal ligament pressure necrosis and excessive alveolar bone resorption; if the maximum stress value is too low, tooth movement efficiency will be poor, and the treatment period will be prolonged.

[0053] Therefore, this module sets a stress safety range to determine the maximum stress value. In one possible implementation, the stress safety range is the maximum stress value of the periodontal ligament between 0.02 MPa and 0.26 MPa. This range is determined based on biomechanical research and clinical experience, ensuring effective tooth movement while avoiding tissue damage caused by excessive stress. It should be noted that the stress safety range can be dynamically adjusted according to the patient's age, periodontal health status, and bone density. For example, for young patients, due to active bone metabolism, the upper limit of stress can be appropriately increased to 0.30 MPa; for middle-aged and elderly patients or patients with periodontitis, the upper limit of stress should be decreased to 0.20 MPa to ensure treatment safety.

[0054] If the maximum stress value does not meet the stress safety range, the module adjusts the candidate movement scheme. Specifically, if the maximum stress value exceeds the upper limit of the stress safety range, the module reduces the movement distance or adjusts the movement direction in the candidate movement scheme to reduce the applied orthodontic force and the generated stress; if the maximum stress value is below the lower limit of the stress safety range, the module increases the movement distance in the candidate movement scheme to improve treatment efficiency while ensuring that the stress remains within the safety range. After adjustment, the module performs a new finite element simulation to verify whether the adjusted scheme meets the stress safety requirements. This process may require multiple iterations until movement parameters that meet the stress safety range are found.

[0055] The optimal solution module 4 is connected to the intelligent planning module 2 and the biomechanical simulation module 3. It receives a set of candidate movement solutions that have been biomechanically verified and adjusted, and comprehensively evaluates each candidate solution to select the optimal one. For example... Figure 5 As shown, the scoring calculation process of this module includes four steps: occlusal relationship scoring, aesthetic effect scoring, treatment cycle scoring, and comprehensive reward value calculation.

[0056] For candidate repositioning schemes, this module first calculates an occlusal relationship score using an occlusal relationship evaluation algorithm. The occlusal relationship score is calculated based on the number of occlusal contact points after treatment, the uniformity of occlusal force distribution, and occlusal stability. In one possible implementation, the module applies the candidate repositioning scheme to an initial three-dimensional oral model to simulate the tooth position after treatment, and then analyzes the occlusal contact of the upper and lower teeth. The module counts the number of occlusal contact points; the more contact points, the more stable the occlusal relationship, and the higher the score. The module also calculates the uniformity of occlusal force distribution by calculating the coefficient of variation of the force at each occlusal contact point; the smaller the coefficient of variation, the more uniform the distribution, and the higher the score. Furthermore, the module evaluates occlusal stability, including indicators such as the overbite / overjet relationship of the anterior teeth and the cusp-fossa relationship of the posterior teeth. Combining these indicators, the module calculates an occlusal relationship score, ranging from 0 to 1, with 1 indicating the optimal occlusal relationship.

[0057] For candidate tooth repositioning schemes, this module calculates an aesthetic effect score using an aesthetic effect evaluation algorithm. The aesthetic effect score is calculated based on post-treatment tooth alignment, facial symmetry, and smile curve coordination. In one possible implementation, the module evaluates tooth alignment by measuring parameters such as the gap between adjacent teeth and the tilt angle of the teeth to calculate an alignment index. The module evaluates facial symmetry by calculating the deviation of the facial midline from the dental arch midline and the symmetry of the positions of the teeth on both sides. The module also evaluates smile curve coordination, where the smile curve is the curve connecting the incisal edges of the maxillary anterior teeth; an ideal smile curve should be parallel to the lower lip curve. Combining these aesthetic indicators, the module calculates an aesthetic effect score, ranging from 0 to 1, with 1 indicating the optimal aesthetic effect.

[0058] For candidate mobility plans, this module calculates a treatment cycle score based on the total duration of multiple treatment phases. In one possible implementation, the module counts the number of treatment phases included in the candidate mobility plan and calculates the total treatment cycle based on the time required for each phase (typically 2 to 4 weeks). The shorter the treatment cycle, the higher the score. For example, a treatment cycle of 12 months results in a score of 1; a treatment cycle of 24 months results in a score of 0.7; and a treatment cycle of 36 months results in a score of 0.4. This module uses a function mapping to convert the treatment cycle into a score value of 0 to 1.

[0059] Finally, the module performs a weighted summation of the occlusal relationship score, aesthetic effect score, and treatment cycle score based on preset weighting coefficients to obtain a comprehensive reward value. In one possible implementation, the formula for calculating the comprehensive reward value is:

[0060] ,

[0061] in, For the overall reward value, To score the occlusal relationship, Rate the aesthetic effect. Scoring for the treatment cycle, , , These are the weighting coefficients, and For example, when patients are more concerned about treatment outcomes, settings can be configured... , , When patients are more concerned about the treatment cycle, settings can be made... , , By adjusting the weighting coefficients, this module can generate the optimal treatment plan based on the patient's individual preferences.

[0062] This module determines whether the overall reward value falls within a reward threshold range. In one possible implementation, the reward threshold range is an overall reward value greater than 0.85. Only candidate movement schemes with overall reward values ​​exceeding this threshold are included in the final preferred range. If the overall reward values ​​of all candidate movement schemes do not fall within the reward threshold range, this module sends feedback to the intelligent planning module 2, requesting the generation of more candidate movement schemes, or adjusting the reinforcement learning parameters to explore a better solution space.

[0063] For candidate movement schemes that fall within the reward threshold range, this module selects the candidate movement scheme with the highest overall reward value as the optimal movement scheme. The optimal movement scheme represents the best balance between occlusal relationship, aesthetic effect, and treatment cycle under the current constraints, and can provide patients with personalized, high-quality orthodontic treatment plans.

[0064] The prediction generation module 5 is connected to the scheme optimization module 4, receives the optimal movement scheme, and performs simulated movement processing on the initial three-dimensional oral model according to the optimal movement scheme to generate a predicted three-dimensional oral model. The predicted three-dimensional oral model shows the oral state after treatment, including the final position of the teeth, occlusal relationship, and facial contour.

[0065] In one possible implementation, the module progressively adjusts the spatial position of the target tooth in the initial 3D oral model according to the movement parameters of each treatment stage in the optimal movement scheme. For each treatment stage, the module calculates the position transformation matrix of the target tooth based on the movement type, movement distance, movement angle, and movement direction, and applies this transformation matrix to update the 3D coordinates of the tooth model. The module also simulates the adaptive remodeling of periodontal tissues, including the stretching of the periodontal ligament, alveolar bone resorption and regeneration, making the predicted 3D oral model more realistic.

[0066] After generating the predictive 3D oral model, this module can be displayed to doctors and patients through a visual interface. Doctors can observe the predictive 3D oral model from multiple angles to assess whether the treatment effect meets expectations. Patients can intuitively see the before-and-after comparison, understand the post-treatment tooth alignment, occlusion, and facial aesthetics, which helps improve patients' understanding and acceptance of the treatment plan and enhances treatment compliance.

[0067] Furthermore, this module can generate design data for personalized orthodontic appliances based on a predicted 3D oral model. For example, for invisible orthodontics, the module generates a corresponding clear aligner model for each treatment stage, with the inner surface shape of the aligner matching the expected tooth position for that stage, allowing for gradual tooth movement through progressive aligner replacement. For fixed orthodontics, the module generates bracket bonding positions and archwire bending parameters to guide the dentist in precise bracket bonding and archwire adjustments.

[0068] In a preferred embodiment, the system of the present invention further includes a dynamic adjustment module 6, such as... Figure 1 As shown in the dashed box. The dynamic adjustment module 6 is connected to the prediction generation module 5 and is used to dynamically update the treatment plan during the treatment process, further improving the accuracy and success rate of the treatment.

[0069] Orthodontic treatment is a process that can last for months or even years, during which the patient's oral condition will constantly change. Due to individual differences, the actual speed and path of tooth movement may deviate from the prediction. Furthermore, patient compliance (such as the duration of appliance wear) and oral hygiene can also affect treatment progress. Therefore, regular monitoring of treatment progress and adjustments to the treatment plan based on the actual situation are crucial to ensuring treatment effectiveness.

[0070] The dynamic adjustment module 6 periodically receives actual oral scan data from the patient during treatment. In one possible implementation, this module receives oral scan data every 4 to 8 weeks, and the scanning frequency can be adjusted according to treatment progress and tooth movement speed. The actual oral scan data is acquired using the same equipment and methods as the initial scan and is preprocessed to form an actual three-dimensional oral model.

[0071] This module compares the actual oral scan data with the expected oral state at the corresponding treatment stage in the optimal movement plan. Specifically, this module registers and compares the actual 3D oral model with the predicted 3D oral model, calculating indicators such as positional deviation, angular deviation, and occlusal relationship deviation of the corresponding teeth. For example, this module calculates the Euclidean distance between the actual and expected positions of each target tooth, as well as the angular difference between the actual and expected angles.

[0072] This module determines whether the deviation exceeds a preset deviation threshold. In one possible implementation, the preset deviation threshold is set to a positional deviation greater than 0.5 mm or an angular deviation greater than 5 degrees. If the deviation is within the threshold range, it indicates that the treatment progress is as expected, and the module does not adjust the plan, continuing to execute the original optimal movement plan.

[0073] If the deviation exceeds the preset deviation threshold, it indicates a significant difference between the actual treatment progress and the prediction, requiring an update to the treatment plan. Understandably, such deviations can be caused by various factors, such as insufficient time spent wearing the orthodontic appliance leading to slower-than-expected tooth movement, greater-than-expected resistance to tooth movement, or unforeseen bone remodeling abnormalities.

[0074] At this point, the module re-triggers the intelligent planning module 2, biomechanical simulation module 3, and solution selection module 4 to generate updated optimal relocation plans. Specifically, this module uses the actual 3D oral cavity model as the new initial state, re-extracts tooth feature information, and, based on the current state and remaining treatment goals, calls the reinforcement learning algorithm of the intelligent planning module 2 to generate a new set of candidate relocation plans. Then, the biomechanical simulation module 3 verifies the safety of the new candidate relocation plans, and the solution selection module 4 performs a comprehensive evaluation and selects the new optimal relocation plan.

[0075] After generating the updated optimal movement plan, the prediction generation module 5 generates a new predictive 3D oral model based on the updated plan and notifies the doctor and patient that the plan has been updated. The doctor can view the updated treatment plan, assess the arrangement of the remaining treatment stages, and adjust the orthodontic appliances as needed. The patient can also understand the reasons for the plan adjustment and the new treatment expectations, maintaining confidence in the treatment.

[0076] Understandably, through the dynamic adjustment module 6, the system of this invention can achieve closed-loop management of planning-execution-monitoring-adjustment, ensuring that the treatment plan always matches the patient's actual condition. This dynamic adjustment mechanism significantly improves the accuracy and success rate of treatment, especially in the face of complex cases or fluctuations in patient compliance, enabling timely correction and avoiding poor treatment results or prolonged treatment cycles.

[0077] like Figure 6 As shown, the complete workflow of the AI-driven intelligent design and prediction system for orthodontic treatment plans of the present invention is as follows:

[0078] Step S1: The oral data acquisition module 1 receives the patient's oral scan data, performs three-dimensional reconstruction and feature extraction, and obtains an initial three-dimensional oral model and tooth feature information.

[0079] Step S2: The intelligent planning module 2 determines whether the tooth feature information falls within the abnormal feature range. If it does not, the patient does not require orthodontic treatment, and the process ends; if it does, proceed to step S3.

[0080] Step S3: The intelligent planning module 2 determines the target tooth and treatment goal, and generates a set of candidate relocation schemes based on a deep reinforcement learning algorithm.

[0081] Step S4: Biomechanical simulation module 3 performs biomechanical simulation for each scheme in the candidate movement scheme set, calculates the maximum stress value, and adjusts schemes that do not meet the stress safety range.

[0082] Step S5: The scheme optimization module 4 comprehensively evaluates the adjusted candidate movement schemes, calculates the comprehensive reward value, and selects the scheme with the highest comprehensive reward value as the optimal movement scheme.

[0083] Step S6: The prediction generation module 5 generates a predictive 3D oral model based on the optimal movement scheme to show the oral state after treatment.

[0084] Step S7: The doctor reviews the predicted 3D oral model, communicates with the patient to confirm the treatment plan, and begins orthodontic treatment.

[0085] Step S8: During the treatment process, the dynamic adjustment module 6 periodically receives actual oral scan data and compares the actual state with the expected state.

[0086] Step S9: Determine whether the deviation exceeds the preset deviation threshold. If it does not exceed the threshold, continue with the original treatment plan and return to step S8; if it exceeds the threshold, return to step S3 to regenerate an updated treatment plan.

[0087] Step S10: Treatment completed, final oral condition achieved, treatment goal achieved.

[0088] Through the above workflow, the system of the present invention can provide each patient with a personalized, safe, and multi-objective optimized orthodontic treatment plan, and dynamically adjust it during the treatment process to ensure that the treatment effect reaches the expected level.

[0089] To more clearly illustrate the practical application effects of the present invention, a specific application example is given below.

[0090] A 28-year-old female patient presented with malocclusion and protrusion of the maxillary anterior teeth. The dentist used an intraoral scanner to obtain oral scan data from the patient and input the data into the system of this invention.

[0091] The oral data acquisition module 1 processes the scanned data, generates an initial three-dimensional oral model, and extracts tooth feature information. The extraction results show that the maxillary central incisor protrudes by 3.5 mm, the left lateral incisor is rotated by 18 degrees, and the right canine is labially misaligned by 2.8 mm; the number of occlusal contact points is 10, which is less than the normal range; and the facial midline is deviated to the right by 2.5 mm.

[0092] The anomaly identification submodule of the intelligent planning module 2 determines that these features meet the abnormal feature range and confirms that the patient needs orthodontic treatment. The goal setting submodule sets the treatment goals as follows: retract the maxillary anterior teeth to their normal positions, rotate and reposition the lateral incisors, and correct labial misalignment of the canines; increase the number of occlusal contact points to more than 18; reduce facial midline deviation to within 1 mm; and control the treatment cycle to within 18 months.

[0093] The reinforcement learning submodule establishes a reinforcement learning environment and trains the system using deep reinforcement learning algorithms to generate 20 candidate movement schemes. These schemes differ in movement paths, number of treatment phases, and focus.

[0094] The biomechanical simulation module 3 performed finite element simulations on each of the 20 candidate movement schemes. The simulation results showed that the maximum stress value of 5 of the schemes exceeded 0.26 MPa, which may lead to periodontal tissue damage. Therefore, the module adjusted these 5 schemes by reducing the movement distance or optimizing the movement direction to reduce the maximum stress value to a safe range.

[0095] The optimal solution module 4 comprehensively evaluates the 20 adjusted candidate movement schemes. The calculation results show that the comprehensive reward value for scheme A is 0.92, for scheme B it is 0.89, for scheme C it is 0.87, and the comprehensive reward values ​​for the remaining schemes are all below 0.85. This module selects scheme A as the optimal movement scheme.

[0096] Plan A comprises 32 treatment phases, with an estimated treatment period of 16 months. The main movement steps are as follows: In the first 8 phases, the right canine is moved as a whole, posteriorly by 2.8 mm and lingually by 1.5 mm; in phases 9 to 16, the left lateral incisor is rotated, gradually returning to its normal angle; in phases 17 to 28, the maxillary central incisor is moved as a whole, posteriorly by 3.5 mm, to adjust protrusion; in phases 29 to 32, all maxillary anterior teeth are fine-tuned to optimize occlusion and aesthetics.

[0097] The prediction generation module 5 generates a predictive 3D oral model based on Plan A. The doctor and patient view the before-and-after comparison on a visual interface. The predicted 3D oral model shows that after treatment, the teeth are aligned neatly, protrusion is completely corrected, facial midline deviation is reduced to 0.8 mm, the number of occlusal contact points increases to 20, and the occlusal relationship is stable. The patient is very satisfied with the prediction results and agrees to use Plan A for treatment.

[0098] After treatment began, the patient returned for follow-up appointments every 8 weeks, during which the dentist used an intraoral scanner to collect actual oral data. At the 16-week follow-up appointment, the dynamic adjustment module 6 compared the actual data with the expected data and found that the actual movement distance of the right canine was 2.2 mm, less than the expected 2.8 mm, with a positional deviation of 0.6 mm, exceeding the preset deviation threshold of 0.5 mm. Analysis revealed that the deviation was caused by insufficient appliance wearing time at certain periods, resulting in a slower-than-expected movement speed.

[0099] The dynamic adjustment module 6 retried the system, regenerating the treatment plan based on actual oral data. The updated plan adjusted the remaining treatment period from 12 months to 14 months, appropriately extending the movement phase of the right canine and optimizing the parameters of subsequent movement steps. The doctor explained the reasons for the plan adjustment to the patient, emphasizing the importance of appliance adherence, and the patient stated that they would strictly follow the wearing requirements.

[0100] During the subsequent treatment, patient compliance improved significantly, and the treatment progressed smoothly. After 18 months of treatment, the patient's teeth alignment achieved the expected results, protrusion was corrected, occlusion was good, and facial aesthetics were significantly improved. The final treatment outcome was highly consistent with the predicted 3D oral model, and both the patient and doctor expressed satisfaction with the treatment results.

[0101] This specific application case demonstrates that the system of the present invention can generate highly personalized treatment plans based on the individual patient's condition, ensure the safety of the plans through biomechanical simulation, and respond to changes in the treatment process through a dynamic adjustment mechanism, ultimately achieving high-quality treatment results.

[0102] In summary, the AI-driven intelligent design and prediction system for orthodontic treatment plans of the present invention has the following significant technical advantages:

[0103] First, this invention achieves intelligent optimization of tooth movement paths by introducing a deep reinforcement learning algorithm. Deep reinforcement learning algorithms can explore on a large scale in high-dimensional state and action spaces, automatically discovering the complex trade-offs between occlusal relationships, aesthetic effects, and treatment cycles, and generating diverse candidate movement schemes. Compared to traditional methods that rely on preset rules and physician experience, the system of this invention can explore a broader solution space and find better treatment plans.

[0104] Secondly, this invention innovatively introduces a biomechanical simulation module to accurately simulate the periodontal tissue stress distribution for each candidate repositioning scheme. By establishing a finite element model including the tooth, periodontal ligament, and alveolar bone, the system can predict the stress distribution under orthodontic force and ensure that all schemes remain within the biomechanical safety range through stress safety range judgment and automatic adjustment mechanisms. This forward-looking safety verification effectively avoids serious complications such as root resorption and periodontal ligament necrosis caused by excessive orthodontic force, significantly improving the clinical safety of treatment.

[0105] Third, this invention achieves multi-objective collaborative optimization through a scheme optimization module. The comprehensive reward value comprehensively considers three key dimensions: occlusal relationship, aesthetic effect, and treatment cycle, and supports dynamic adjustment of weight coefficients based on patient preferences. This multi-objective optimization strategy can flexibly balance different treatment goals, providing personalized optimal solutions for patients with different needs. For example, for young patients who value aesthetic effects, the weight of the aesthetic score can be increased; for busy patients who wish to shorten the treatment cycle, the weight of the treatment cycle score can be increased. This flexibility greatly improves patient satisfaction and treatment adherence.

[0106] Fourth, this invention achieves real-time updates to the treatment plan during the treatment process through a dynamic adjustment module. Orthodontic treatment is a long-term process, and the patient's actual oral condition may deviate from expectations due to individual differences, fluctuations in compliance, and other factors. The dynamic adjustment module periodically monitors the actual progress, and when a deviation exceeds a threshold, it automatically triggers a regeneration of the treatment plan, ensuring that the treatment always proceeds along the correct path. This closed-loop management mechanism significantly improves the accuracy and success rate of treatment, especially in the face of complex cases, enabling timely correction and avoiding treatment failure or poor results.

[0107] Fifth, the system of this invention has excellent scalability and generalization ability. The deep reinforcement learning algorithm can be trained on a large amount of case data to continuously improve the quality of the strategy and adapt to more types of malocclusion. The biomechanical simulation module can dynamically adjust material parameters and safety ranges according to individual characteristics such as age and periodontal health status of different patients, achieving truly personalized simulation. The weight coefficients of the scheme optimization module can be configured according to the preferences of different regions and doctors to meet diverse clinical needs.

[0108] Finally, the system of this invention provides patients with intuitive visual predictions. The predicted 3D oral model clearly shows the oral state after treatment, including tooth alignment, occlusion, and facial contours. This visual prediction not only helps doctors better assess the feasibility of the treatment plan but also helps patients intuitively understand the treatment effect, enhancing their confidence and compliance. Clinical practice shows that patients who can see the expected results have significantly higher treatment compliance than those who rely solely on verbal descriptions.

[0109] The technical solution of this invention is not only applicable to traditional fixed and invisible orthodontic treatments, but can also be extended to various orthodontic techniques such as lingual orthodontics and functional orthodontics. By adjusting the motion space and reward function, the system can adapt to the characteristics and constraints of different orthodontic techniques. Furthermore, the technical solution of this invention can be integrated with other digital dental technologies, such as combining with 3D printing technology to create personalized orthodontic devices, combining with virtual reality technology to provide immersive treatment plan demonstrations, and combining with telemedicine platforms to achieve online consultation and remote monitoring, forming a complete digital orthodontic ecosystem.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the above exemplary embodiments, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the invention is defined by the appended claims rather than the foregoing description, and all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An AI-driven intelligent design and prediction system for orthodontic treatment plans, characterized in that, include: The oral data acquisition module is used to receive oral scan data from patients, perform three-dimensional reconstruction processing on the oral scan data to obtain an initial three-dimensional oral model, the initial three-dimensional oral model including a tooth model, a gingival model and a periodontal tissue model; and acquire tooth feature information corresponding to the initial three-dimensional oral model, the tooth feature information including tooth position information, occlusal relationship information and facial contour information. The intelligent planning module is connected to the oral data acquisition module and is used to determine the target tooth and treatment target if the tooth feature information meets the abnormal feature range. The treatment target includes the target occlusal relationship, the target aesthetic effect and the target treatment cycle. A set of candidate movement schemes is generated based on a deep reinforcement learning algorithm. Each candidate movement scheme in the set includes the movement path and movement parameters of the target tooth in multiple treatment stages. A biomechanical simulation module, connected to the intelligent planning module, is used to simulate the stress distribution of periodontal tissues during the movement of the target tooth for each candidate movement scheme in the candidate movement scheme set, and to obtain the maximum stress value corresponding to the stress distribution. If the maximum stress value does not meet the stress safety range, the candidate movement scheme shall be adjusted. The optimal solution module, connected to the intelligent planning module and the biomechanical simulation module, is used to calculate the comprehensive reward value for each candidate movement solution in the adjusted candidate movement solution set. The comprehensive reward value is calculated based on the occlusal relationship score, aesthetic effect score, and treatment cycle score. If the comprehensive reward value meets the reward threshold range, the candidate movement solution with the highest comprehensive reward value is selected as the optimal movement solution. The prediction generation module, connected to the scheme selection module, is used to simulate the movement of the initial three-dimensional oral model according to the optimal movement scheme to generate a predicted three-dimensional oral model, which displays the oral state after treatment.

2. The system according to claim 1, characterized in that, The oral data acquisition module is specifically used for: The oral cavity scan data is preprocessed to remove noise points and redundant data; The processed oral cavity scan data is reconstructed into three dimensions based on the point cloud registration algorithm to generate the initial three-dimensional oral cavity model. The initial three-dimensional oral cavity model is semantically segmented to identify the tooth region, gingival region, and periodontal tissue region; The tooth feature information is extracted from the initial three-dimensional oral model. The tooth feature information includes at least the three-dimensional coordinates of each tooth, the distance between adjacent teeth, the occlusal contact points of the upper and lower teeth, and the position of the facial midline.

3. The system according to claim 1, characterized in that, The intelligent planning module includes: An anomaly identification submodule is used to compare the tooth feature information with the range of normal tooth features to determine whether the tooth feature information conforms to the range of abnormal features. The target setting submodule, connected to the anomaly recognition submodule, is used to determine the treatment target based on the degree of deviation between the tooth feature information and the normal tooth feature range. The reinforcement learning submodule, connected to the target setting submodule, is used to establish a reinforcement learning environment. The reinforcement learning environment uses the current tooth state as the state space, the tooth movement operation as the action space, and the degree of proximity to the treatment target as the reward function. The deep reinforcement learning algorithm is used to explore and optimize in the reinforcement learning environment to generate the set of candidate movement schemes.

4. The system according to claim 1 or 2, characterized in that, The range of abnormal features includes the range of tooth misalignment deviation, the range of abnormal occlusion, and the range of facial symmetry deviation; The range of tooth misalignment deviation is defined as a positional deviation of a single tooth greater than 2 mm or an angular deviation greater than 10 degrees. The abnormal range of occlusion is defined as the number of occlusal contact points of the upper and lower teeth being less than a preset contact point threshold, or the unevenness of the occlusal force distribution being greater than a preset unevenness threshold. The range of facial symmetry deviation is defined as a deviation of more than 3 millimeters from the facial midline.

5. The system according to claim 1 or 2, characterized in that, The movement parameters in the candidate movement scheme include movement type, movement distance, movement angle, and movement direction; The movement type includes at least one of tilting movement, overall movement, torque movement, rotational movement, depressurization movement, and elongation movement; The deep reinforcement learning algorithm explores diverse movement paths by setting action combinations of different movement types, and selects the best-performing movement paths through a reward feedback mechanism.

6. The system according to claim 1, characterized in that, The biomechanical simulation module is specifically used for: A finite element model of teeth and periodontal tissues is established, the finite element model including tooth elements, periodontal ligament elements and alveolar bone elements; The magnitude and direction of the orthodontic force applied to the target tooth are determined based on the candidate movement scheme; Based on the finite element model, the stress distribution and displacement distribution of periodontal tissues under orthodontic force are simulated, and the maximum stress value is calculated. If the maximum stress value exceeds the upper limit of the stress safety range, the movement distance in the candidate movement scheme is reduced or the movement direction is adjusted; if the maximum stress value is lower than the lower limit of the stress safety range, the movement distance in the candidate movement scheme is increased to improve treatment efficiency.

7. The system according to claim 1, characterized in that, The stress safety range is defined as the maximum stress value of the periodontal ligament between 0.02 MPa and 0.26 MPa. This stress safety range is dynamically adjusted according to the patient's age, periodontal health status, and bone density.

8. The system according to claim 1, characterized in that, The scheme optimization module is also used for: For the candidate relocation scheme, the occlusal relationship score is calculated using an occlusal relationship evaluation algorithm. The occlusal relationship score is calculated based on the number of occlusal contact points, the uniformity of occlusal force distribution, and occlusal stability after treatment. For the candidate repositioning scheme, an aesthetic effect score is calculated using an aesthetic effect evaluation algorithm. The aesthetic effect score is calculated based on the post-treatment tooth alignment, facial symmetry, and smile curve coordination. For the candidate movement scheme, the treatment cycle score is calculated based on the total duration of the multiple treatment stages; The comprehensive reward value is obtained by weighting and summing the occlusal relationship score, the aesthetic effect score, and the treatment cycle score according to preset weighting coefficients.

9. The system according to claim 1, characterized in that, The reward threshold range is when the comprehensive reward value is greater than 0.85, wherein the comprehensive reward value ranges from 0 to 1; The preset weighting coefficients are adjusted according to the patient's treatment preferences. If the patient is more concerned about the treatment effect, the weights of the occlusal relationship score and the aesthetic effect score are increased; if the patient is more concerned about the treatment cycle, the weight of the treatment cycle score is increased.

10. The system according to claim 1, characterized in that, The system also includes: The dynamic adjustment module, connected to the prediction generation module, is used to periodically receive the patient's actual oral scan data during treatment and compare the actual oral scan data with the expected oral state at the corresponding treatment stage in the optimal movement plan. If the deviation exceeds a preset deviation threshold, the intelligent planning module, the biomechanical simulation module, and the plan optimization module are re-triggered to generate an updated optimal movement plan.

Images