Mandibular third molar extraction risk assessment method, system, device and medium

Abstract

This application provides a method, system, device, and medium for assessing the extraction risk of the mandibular third molar. The method includes inputting target risk assessment information and target medical record information into a target extraction risk assessment model. This allows the model to comprehensively quantify the proximity of the target patient's mandibular third molar to the mandibular second molar and the inferior alveolar nerve canal based on the target risk assessment information. The model also cross-validates the target patient's mandibular second molar lesion status and extraction risk-related medical history based on the target risk assessment information and target medical record information. Furthermore, the target extraction risk assessment model determines the target extraction risk level based on a fusion of medical rules and AI scoring algorithms and outputs target extraction risk warning information. This solution integrates CBCT image analysis, intraoral scan information processing, and structured medical record information to automatically identify the spatial relationship and pathological condition of the mandibular third molar and output extraction risk warning information, providing relevant surgical references.

Description

Technical Field

[0001] This application relates to the field of oral medicine technology, and in particular to a method, system, device and medium for risk assessment of mandibular third molar extraction. Background Technology

[0002] The mandibular third molar is often impacted due to insufficient space in the jawbone, which can easily lead to pericoronitis, damage to adjacent teeth, and other conditions requiring surgical extraction. However, this surgery carries a high risk. The mandibular anatomy is complex, and the mandibular third molar is adjacent to important structures such as the mandibular canal. Complications such as numbness of the lower lip and damage to adjacent teeth are common during the procedure. Traditional methods have a high complication rate. Therefore, accurate preoperative risk assessment is crucial to reducing surgical risks and improving the success rate.

[0003] Accurate preoperative assessment not only helps doctors anticipate surgical challenges and optimize surgical plans, but also effectively reduces surgical trauma and complications, providing a core guarantee for surgical safety. Imaging examinations are a core basis for preoperative risk assessment, but traditional two-dimensional imaging examinations have significant limitations.

[0004] Traditional two-dimensional imaging examinations cannot accurately present the three-dimensional relationship between teeth and surrounding structures, which can easily lead to misjudgment. In contrast, cone-beam computed tomography (CBCT) images, with their high spatial resolution, can clearly show the morphology of the mandibular third molar and its positional relationship with surrounding structures, significantly improving the accuracy of assessment.

[0005] Although technologies such as CBCT provide a more accurate imaging basis for risk assessment, current technologies still have many shortcomings: First, they do not integrate multi-source information and the assessment dimensions are not comprehensive; second, they focus on single detection tasks and lack comprehensive identification capabilities and a complete risk assessment system; third, the interpretation of cone-beam CT relies on the doctor's experience, the risk assessment lacks a unified quantitative standard, and the surgical parameters are not clearly standardized, which can easily increase risks and trial-and-error costs.

[0006] Given the shortcomings of existing technologies and the rapid development of digital healthcare, traditional experience-based diagnosis and treatment models can no longer meet the clinical needs for surgical safety and precision. Therefore, there is an urgent need for a technical solution that integrates multi-source data, achieves automated and accurate risk assessment, and provides surgical references, thereby improving the level of surgical intelligence and reducing surgical risks. Summary of the Invention

[0007] To overcome the aforementioned technical deficiencies, this application provides a method, system, device, and medium for assessing the risk of mandibular third molar extraction. It integrates CBCT image analysis, intraoral scan information processing, and structured medical record information to automatically identify the spatial relationship and pathological condition of the mandibular third molar, and output extraction risk warning information, providing relevant surgical references. This can effectively improve the scientificity and accuracy of surgical planning, reduce the incidence of surgical complications, ensure patient safety, and improve clinical treatment efficiency.

[0008] To solve the above-mentioned technical problems, this application provides the following technical solution: According to a first aspect of the embodiments of this application, a method for assessing the risk of extraction of the mandibular third molar is provided, the method comprising: Acquire target acquisition information corresponding to the target object; the target acquisition information includes target intraoral scan information, target CBCT image information, and target medical record information; The target intraoral scan information and the target CBCT image information are input into the target information processing model so that the target information processing model can process the target intraoral scan information and the target CBCT image information to obtain target risk assessment information; The target risk assessment information and the target medical record information are input into the target extraction risk assessment model, so that the target extraction risk assessment model comprehensively quantifies the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target risk assessment information, and cross-validates the target object's mandibular second molar lesion status and extraction risk-related past history based on the target risk assessment information and the target medical record information; and so that the target extraction risk assessment model determines the target extraction risk level based on the fusion of medical rules and AI scoring algorithms and outputs target extraction risk warning information; the target extraction risk warning information includes the target extraction risk level corresponding to the target object and extraction-related information corresponding to the target object that meets preset rules.

[0009] In one exemplary embodiment, the step of inputting the target intraoral scan information and the target CBCT image information into a target information processing model, so that the target information processing model processes the target intraoral scan information and the target CBCT image information to obtain target risk assessment information, specifically includes: The target intraoral scan information and the target CBCT image information are input into the target information processing model, so that the target information processing model performs rigid registration on the target intraoral scan information and the target CBCT image information to establish a unified spatial coordinate system; and so that the target information processing model performs voxel-by-voxel automatic segmentation on the target CBCT image information based on a pre-trained three-dimensional deep convolutional neural network to extract the target structure corresponding to the target object, and performs tooth and soft tissue segmentation on the target intraoral scan information based on a pre-trained deep hierarchical neural network to extract the soft tissue morphology of the free gingiva, attached gingiva and oral mucosa corresponding to the target object, thereby obtaining the target risk assessment information.

[0010] In one exemplary implementation, the three-dimensional deep convolutional neural network employs an improved 3DnnUNet, which is based on the native nnUNet framework, embeds an attention mechanism, and adds loss functions at different levels of the decoder; the deep hierarchical neural network employs PointNet++.

[0011] In an exemplary embodiment, the target risk assessment information includes target acquisition information and a target visualization 3D model; the target visualization 3D model carries target 3D axial classification information, target horizontal displacement grading information, and target vertical position grading information; the target 3D axial classification information indicates the impaction direction type of the target object's mandibular third molar, obtained by combining the angle between the long axis of the target object's mandibular third molar and the normal vector of the standard occlusal plane with the tooth projection direction; the target horizontal displacement grading information indicates the horizontal displacement level of the target object's mandibular third molar, obtained by combining the horizontal overlap between the crown of the target object's mandibular third molar and the anterior edge of the mandibular ramus with the Pell & Gregory classification standard; the target vertical position grading indicates the depth level of the target object's mandibular third molar, obtained by combining the vertical distance from the apex of the crown of the target object's mandibular third molar to the occlusal surface of the adjacent tooth with the Pell & Gregory classification standard.

[0012] In an exemplary implementation, the step of inputting the target risk assessment information and the target medical record information into the target extraction risk assessment model allows the target extraction risk assessment model to comprehensively quantify the proximity of the target subject's mandibular third molar to the mandibular second molar and the inferior alveolar nerve canal based on the target risk assessment information, and to cross-validate the target subject's mandibular second molar lesion status and extraction risk-related past history based on the target risk assessment information and the target medical record information. Specifically, this includes: The target risk assessment information and the target medical record information are input into the target extraction risk assessment model so that the target extraction risk assessment model obtains the target parameters of the target object based on the target risk assessment information. The target extraction risk assessment model comprehensively quantifies the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target parameters. The target extraction risk assessment model also cross-validates the lesion status of the target object's mandibular second molar based on the target CBCT image information and the target intraoral scan information, and determines the target object's past history related to extraction risk based on the target medical record information. The target parameters include the shortest distance between the crown of the mandibular third molar and the distal surface of the mandibular second molar, the estimated contact area between the crown of the mandibular third molar and the mandibular second molar, the axial angle of the mandibular third molar, the surface contact volume between the proximal root of the mandibular third molar and the root surface of the mandibular second molar, the three-dimensional shortest path distance between the root apex of the mandibular third molar and the inferior alveolar nerve canal, the overlapping projection area of ​​the shadow of the mandibular third molar and the shadow of the inferior alveolar nerve canal, the angle difference between the long axis of the mandibular third molar and the direction of the inferior alveolar nerve canal, and the volume ratio of the overlapping area in three-dimensional space. The pathological state of the mandibular second molar of the target object is determined based on the proximal caries, residual tooth structure, external root resorption, restoration status, and the health status of the mandibular soft tissue of the target object. The proximal caries and residual tooth structure of the mandibular second molar of the target object are determined based on the target parameters. The low-density radiolucent area of ​​the mandibular second molar obtained from CBCT imaging information is used to determine the external resorption of the mandibular second molar root, which is determined by the root contour defect of the mandibular second molar obtained from the target CBCT imaging information. The restoration status of the target object is determined by whether the target object has a restoration or filling and the restoration method, which is determined by the target CBCT and intraoral scan information. The mandibular soft tissue health status of the target object is assessed by the gingival margin, free gingiva, attached gingiva, and oral mucosa of the mandibular third and second molars, which are determined by the target intraoral scan information. The target object's past history of tooth extraction risk is determined based on the target medical record information. The target medical record information includes the target object's oral medical history and non-oral medical history that meets the oral medical relevance criteria.

[0013] In one exemplary embodiment, the step of causing the target tooth extraction risk assessment model to determine the target tooth extraction risk level and output target tooth extraction risk warning information based on the fusion of medical rules and AI scoring algorithms specifically includes: The target tooth extraction risk assessment model extracts the rule judgment parameters of the target object, and independently matches each parameter in the rule judgment parameters of the target object in parallel based on the medical rules. After traversing all the medical rules, the target rule matching result is obtained. At the same time, the target tooth extraction risk assessment model extracts the AI ​​judgment parameters of the target object, inputs the AI ​​judgment parameters of the target object into the AI ​​scoring algorithm, outputs the target tooth extraction risk score, and obtains the target AI score result based on the target tooth extraction risk score and the AI ​​scoring classification rules. The target tooth extraction risk assessment model integrates the target rule matching result and the target AI scoring result to determine the target tooth extraction risk level and outputs the target tooth extraction risk warning information; The rule-based judgment parameters include the impaction angle of the mandibular third molar, the horizontal position of the mandibular third molar, the vertical position of the mandibular third molar, the contact state between the mandibular third molar and adjacent teeth, the distance between the mandibular third molar and the inferior alveolar nerve canal, the lesion state of the mandibular second molar, and the history related to extraction risk. The AI ​​judgment parameters include the impaction angle of the mandibular third molar, the vertical depth between the crown and the ulnar surface of the mandibular third molar, the shortest distance from the root of the mandibular third molar to the inferior alveolar nerve, the contact area between the mandibular third molar and adjacent teeth, the CT value change of the distal proximal surface of the mandibular third molar, the restoration type label, and the history related to extraction risk.

[0014] In one exemplary implementation, the AI ​​scoring algorithm employs a gradient boosting tree algorithm, with its training objective function being binary cross-entropy, and its optimization metrics being AUC and / or recall. An early stopping strategy is used to prevent overfitting.

[0015] In one exemplary embodiment, the method further includes: Based on the target tooth extraction risk warning information and the basic database, clinical demonstration information and surgical procedure recommendation information are generated. The basic database includes a literature database, a multimedia database, a technique knowledge base, and local technique data.

[0016] According to a second aspect of the embodiments of this application, a mandibular third molar extraction risk assessment system is provided, implemented using any of the mandibular third molar extraction risk assessment methods described above, the system comprising: The information acquisition module is used to acquire target acquisition information corresponding to the target object; the target acquisition information includes target intraoral scan information, target CBCT image information, and target medical record information; The information processing module is used to input the target intraoral scanning information and the target CBCT image information into the target information processing model, so that the target information processing model can process the target intraoral scanning information and the target CBCT image information to obtain target risk assessment information; The risk assessment module is used to input the target risk assessment information and the target medical record information into the target tooth extraction risk assessment model, so that the target tooth extraction risk assessment model comprehensively quantifies the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target risk assessment information, and cross-validates the target object's mandibular second molar lesion status and extraction risk-related past history based on the target risk assessment information and the target medical record information; and so that the target tooth extraction risk assessment model determines the target tooth extraction risk level based on the fusion of medical rules and AI scoring algorithms and outputs target tooth extraction risk warning information; the target tooth extraction risk warning information includes the target tooth extraction risk level corresponding to the target object and extraction-related information corresponding to the target object that meets preset rules.

[0017] According to a third aspect of the embodiments of this application, an electronic device is provided, including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement any of the above-described methods for assessing the risk of extraction of a mandibular third molar.

[0018] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided, the storage medium storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by a processor to implement the mandibular third molar extraction risk assessment method described above.

[0019] The above-mentioned technical solution of this application has the following beneficial effects: (1) This application innovatively integrates CBCT image information, intraoral scan information and structured medical record information to realize the automatic identification of the spatial relationship and pathological condition of the mandibular third molar, and outputs tooth extraction risk warning information to provide relevant surgical references. This can effectively improve the scientificity and accuracy of surgical planning, reduce the incidence of surgical complications, ensure patient safety and improve clinical treatment efficiency.

[0020] (2) This application relies on the multimodal information processing results that integrate CBCT image information, intraoral scan information, and structured medical record information. It uses medical rules and AI scoring algorithms to determine and output standardized tooth extraction risk warning information, which can replace the traditional manual assessment mode that relies on doctors' experience. It not only significantly shortens the assessment time and avoids assessment bias caused by individual experience differences, but also achieves accurate risk classification based on individualized anatomical structure and pathological data. The automatically output tooth extraction risk warning information can be combined with the basic database to generate clinical demonstration information and surgical procedure recommendation information, directly guiding surgical planning and effectively reducing the incidence of serious surgical complications such as nerve damage and adjacent tooth damage. It not only effectively protects the safety of patients' diagnosis and treatment, but also significantly improves the standardization and efficiency of clinical diagnosis and treatment processes, and is suitable for rapid and accurate assessment of clinical medical records.

[0021] (3) This application adapts a specific and accurate preprocessing scheme for different information features. For CBCT image information, an improved 3DnnUNet with an embedded attention mechanism and loss function added in multiple levels of the decoder is used to achieve accurate segmentation and differentiation of the mandibular third molar and its surrounding structures; for the pain point of sparse and irregular intraoral scan information, the characteristics of PointNet++ with strong edge preservation and high computational efficiency are adapted to complete the segmentation of teeth and soft tissues.

[0022] (4) This application cross-validates the lesion status of the second molar through multi-source data. Compared with the traditional single image processing or manual experience assessment method, it solves the problems of insufficient data dimensions, low segmentation accuracy, large irrelevant interference, and obvious subjective error. It greatly improves the accuracy of identifying the spatial relationship and pathological condition of the mandibular third molar, and lays a solid and comprehensive data foundation for risk level determination.

[0023] (5) This application can form a full-chain support of "textual explanation - three-dimensional visualization - evidence-based support - surgical procedure recommendation". Through standardized textual descriptions of tooth extraction risk warnings, it clearly explains the impaction type, contact pattern, lesion condition and risk source, etc.; at the same time, the textual description information is displayed on the constructed visualization three-dimensional model, realizing the observation of key anatomical sites such as tilt direction, adjacent tooth contact surface, and nerve course proximity area. Moreover, the visualization three-dimensional model can be dynamically rotated and dissected for observation, intuitively presenting complex anatomical relationships; through the API interface, it links the cloud literature database, multimedia database and surgical procedure knowledge base as well as local surgical procedure data, automatically generating clinical demonstration information and surgical procedure recommendation information for doctors' reference. Compared with the traditional technology that only outputs risk results, it greatly reduces the cost for doctors to obtain evidence-based evidence and surgical procedure references, and improves the rationality and operability of surgical planning; especially for high-difficulty cases (such as horizontal impaction combined with nerve apposition), it can accurately recommend appropriate surgical procedures and push key terms and related videos, diagrams and other reference materials, which has important guiding value for improving the clinical ability of junior doctors.

[0024] (6) The system of this application supports doctors to customize surgical procedure labels, upload local surgical procedure data, and establish a search index. The system administrator interface allows for flexible expansion of the rule base with new conditions or fine-tuning of risk assessment weights, ensuring that the technology can adapt to the diagnosis and treatment processes of different departments and levels of medical institutions, as well as doctors' personalized clinical habits. Simultaneously, the system can seamlessly connect with external CDSS, PACS, or intelligent imaging workstations to achieve interconnection and interoperability of diagnostic and treatment data. The open keyword-based surgical procedure search entry can link with external video platforms or medical literature engines, embedding search results. Compared to traditional assessment systems with fixed functions and poor scalability, this solution effectively extends the technology lifecycle, reduces the system replacement and integration costs for clinical institutions, and facilitates the large-scale promotion and application of the technology.

[0025] (7) This application can promote the transformation of mandibular third molar extraction treatment from "experience-dependent" to "precision evidence-based", and help optimize and upgrade the clinical treatment model of oral surgery and advance the industry's technology. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a method for assessing the risk of mandibular third molar extraction, provided in an embodiment of this application; Figure 2 A visualized three-dimensional model diagram provided for an embodiment of this application; Figure 3 A structural block diagram of a mandibular third molar extraction risk assessment system provided in this application embodiment; Figure 4 A hardware block diagram of an electronic device for performing a method for assessing the risk of mandibular third molar extraction, provided in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0029] The term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. In the description of the embodiments of this application, it should be understood that the terms "upper," "lower," "top," "bottom," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first" and "second" may explicitly or implicitly include one or more of that feature. Moreover, the terms "first," "second," etc., 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 this application described herein can be implemented in orders other than those illustrated or described herein.

[0030] Please see Figure 1 The diagram shown is a flowchart illustrating a method for assessing the risk of mandibular third molar extraction according to an embodiment of this application. This method includes: Step S1: Obtain the target acquisition information corresponding to the target object; the target acquisition information includes the target intraoral scan information, the target CBCT image information, and the target medical record information; Step S2: Input the target intraoral scan information and target CBCT image information into the target information processing model so that the target information processing model can process the target intraoral scan information and target CBCT image information to obtain target risk assessment information; Step S3: Input the target risk assessment information and target medical record information into the target extraction risk assessment model so that the target extraction risk assessment model can comprehensively quantify the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target risk assessment information, and cross-validate the target object's mandibular second molar lesion status and extraction risk-related past history based on the target risk assessment information and target medical record information; and so that the target extraction risk assessment model can determine the target extraction risk level based on the fusion of medical rules and AI scoring algorithms and output the target extraction risk warning information; the target extraction risk warning information includes the target extraction risk level corresponding to the target object and the extraction-related information corresponding to the target object that meets the preset rules.

[0031] Among them, the target information processing model is obtained by training, validating, and testing the sample collection information and sample information processing model corresponding to the sample object; the target tooth extraction risk assessment model is obtained by training, validating, and testing the sample risk assessment information, sample medical record information, and sample tooth extraction risk assessment model corresponding to the sample object.

[0032] In an optional embodiment, the target intraoral scan information in step S1 above includes the intraoral tooth surface morphology, restoration condition, and soft tissue condition of the target object; the target CBCT image information includes three-dimensional images of the target object's teeth and surrounding anatomical structures; and the target medical record information includes the target object's structured electronic medical record information, including basic information, oral disease history, and non-oral disease history that meets the oral disease association conditions. The oral disease history includes previous dental treatments, root canal treatments, periodontal treatments, restorative treatments, surgical treatment records, and postoperative complication records.

[0033] In an optional embodiment, step S2 above, which involves inputting the target intraoral scan information and the target CBCT image information into the target information processing model, allows the model to process the target intraoral scan information and the target CBCT image information to obtain target risk assessment information. Specifically, this includes: The target intraoral scan information and target CBCT image information are input into the target information processing model so that the target information processing model can rigidly register the target intraoral scan information and target CBCT image information to establish a unified spatial coordinate system; and the target information processing model can automatically segment the target CBCT image information on a voxel-by-voxel basis based on a pre-trained three-dimensional deep convolutional neural network to extract the target structure corresponding to the target object, and segment the teeth and soft tissues of the target intraoral scan information based on a pre-trained deep hierarchical neural network to extract the soft tissue morphology of the free gingiva, attached gingiva and oral mucosa corresponding to the target object, so as to obtain the target risk assessment information.

[0034] Preferably, the 3D deep convolutional neural network adopts an improved 3DnnUNet. The improved 3DnnUNet is based on the native nnUNet framework and embeds an attention mechanism, which enables the model to focus on small but critical areas such as tooth roots and nerve canals, suppress interference from irrelevant background information, and add loss functions at different levels of the decoder to guide shallow networks to learn effective features, accelerate model convergence and improve gradient backpropagation efficiency. The deep hierarchical neural network adopts PointNet++. Taking advantage of the strong edge preservation and high computational efficiency of PointNet++, it is used to segment the teeth and soft tissue of the target object and extract the soft tissue morphology of the free gingiva, attached gingiva and oral mucosa of the target object.

[0035] The attention mechanism uses CBAM, and the loss function uses a composite loss function combining DiceLoss and Cross-EntropyLoss. Under preset conditions, FocalLoss is introduced into the composite loss function or a preset category is given a higher weight than other categories. The preset conditions are that there is a class imbalance problem, such as the neural tube volume is much smaller than the bone (much smaller is defined as the neural tube volume and bone volume reaching a preset ratio). The preset category is the inferior alveolar nerve tube.

[0036] Specifically, the formula for calculating DiceLoss is as follows: ; in For the number of categories, The total number of voxels. It is the probability that voxel i belongs to class c. It is the true label value (0 or 1) of voxel i belonging to category c. This is a smoothing term.

[0037] Specifically, the formula for calculating Cross-EntropyLoss is as follows: ; The formula for calculating FocalLoss is: ; The formula for assigning a higher weight to a predefined category than to other categories is as follows: ; in Let be the weighting coefficient, satisfying .

[0038] Specifically, leveraging PointNet++'s strong edge preservation and high computational efficiency, it is used to segment the teeth and soft tissue of the target object, extracting the soft tissue morphology of the free gingiva, attached gingiva, and oral mucosa of the target object, including: Key points covering the coverage edge are quickly selected from the point cloud (teeth + soft tissue) of the target intraoral scan information using farthest point sampling (FPS). The calculation formula is as follows: Let the original point cloud set be P = {p1, p2, ..., p...} N (N is the total number of oral cavity point clouds, including teeth, free gingiva, attached gingiva, and mucosa), M core points need to be sampled (M≪N): Randomly select the first core point c1∈P; Select the set of already selected core points {c1,...,c k-1 The farthest point is taken as c. k ,Right now: ; in Euclidean distance is used to adapt spatial distance calculations for 3D oral cavity point clouds; BallQuery is used for local neighborhood grouping to construct a local neighborhood for each core point, thereby capturing the local morphology of soft tissue. The calculation formula is as follows: For each core point c k Select all neighboring points within radius r to form a local neighborhood N. k : ; Where r is the neighborhood radius; Using PointNet layers for each local neighborhood N k Feature encoding is performed to extract local features, and the calculation formula is as follows: Let each point p∈N k The original feature is x p The core point is c k The coordinates are c k : Local coordinate normalization: ; Spatial Transformation Network (T-Net) Alignment: Learning transformation matrix T∈R 3×3 Transform the normalized coordinates: ; T-Net's loss function: ; Where I is the identity matrix. It is the Frobenius norm; Multilayer perceptron (MLP) extracts local features + max pooling aggregation: splicing features for each point Perform MLP mapping, and then aggregate neighborhood features using max pooling: ; Feature propagation restores the segmentation features of the original point cloud, and its calculation formula is as follows: Let the core point features of the l-th layer be... The point cloud of layer l-1 is P l-1 Features are propagated through KNN interpolation: for P l-1 For each point p in F, find its position in F. l The k nearest core points {c1,...,c k The feature interpolation value is: ; Where d(p,ci) represents the distance between point p and core point c. iThe Euclidean distance; Cross-entropy loss is used to classify multi-class soft tissue, and its calculation formula is as follows: Suppose the original point cloud has N points, and the prediction probability distribution of the i-th point is: (corresponding to teeth, free gingiva, attached gingiva, and mucosa respectively), the true label is yi (one-hot encoded), and the loss function is: ; Where λ is the regularization coefficient; if class imbalance exists (e.g., few free gingival points), weighted cross-entropy can be introduced: ; w c For category weights.

[0039] In an optional embodiment, the target intraoral scan information and target CBCT image information in step S2 above can be further manually verified by an experienced dentist after information processing to ensure accuracy. The verification includes the integrity of the boundaries of each structure, volume connectivity, the accuracy of the segmentation of the nerve canal near the root apex, the segmentation of the root bifurcation area, and the consistency between the images.

[0040] In an optional embodiment, step S2 may further include: The target information processing model automatically distinguishes between manually verified intraoral scan information and processed CBCT image information of the target object to obtain the mandibular third molar category information. The classification results are output as structured labels, and a visualized 3D model (such as...) is generated. Figure 2 (As shown), for doctors to verify visually.

[0041] Among them, the mandibular third molar category information of the target object automatically includes three-dimensional axial classification (impact direction), horizontal displacement classification (relationship with mandibular ascending ramus), and vertical position classification (relationship with occlusal plane). The three-dimensional axial classification includes seven categories: vertical, mesial, distal, horizontal, inverted, lingual, and buccal. The classification result is obtained by automatically calculating the angle between the long axis of the mandibular third molar and the normal vector of the standard occlusal plane, combined with the tooth projection direction, to determine the impaction direction type. The horizontal displacement classification includes three levels: Class I (complete anterior), Class II (partial overlay), and Class III (complete impaction). The classification results are obtained by automatically calculating the horizontal overlap between the crown of the mandibular third molar and the anterior margin of the mandibular ramus, combined with the Pell & Gregory classification criteria. The vertical position classification includes three depth levels: high (A), middle (B), or low (C). The classification result is determined by automatically measuring the vertical distance from the crown apex of the mandibular third molar to the occlusal surface of the adjacent tooth, combined with the Pell & Gregory criteria.

[0042] In an optional embodiment, the target risk assessment information in step S2 above includes target acquisition information and a target visualization 3D model; the target visualization 3D model carries target 3D axial classification information, target horizontal displacement grading information, and target vertical position grading information; the target 3D axial classification information indicates the impaction direction type of the target object's mandibular third molar obtained by combining the angle between the long axis of the target object's mandibular third molar and the normal vector of the standard occlusal plane with the tooth projection direction; the target horizontal displacement grading information indicates the horizontal displacement level of the target object's mandibular third molar obtained by combining the horizontal overlap between the crown of the target object's mandibular third molar and the anterior edge of the mandibular ramus with the Pell & Gregory classification standard; the target vertical position grading indicates the depth level of the target object's mandibular third molar obtained by combining the vertical distance from the crown apex of the target object's mandibular third molar to the occlusal surface of the adjacent tooth with the Pell & Gregory classification standard.

[0043] In an optional embodiment, step S3 above, which involves inputting the target risk assessment information and target medical record information into the target extraction risk assessment model, allows the model to comprehensively quantify the proximity of the target subject's mandibular third molar to the mandibular second molar and the inferior alveolar nerve canal based on the target risk assessment information. Furthermore, the model cross-validates the target subject's mandibular second molar lesion status and extraction risk-related medical history based on the target risk assessment information and target medical record information. Specifically, this includes: The target risk assessment information and target medical record information are input into the target extraction risk assessment model so that the target extraction risk assessment model can obtain the target parameters of the target object based on the target risk assessment information. The target extraction risk assessment model can comprehensively quantify the proximity of the target object's mandibular third molar, mandibular second molar and inferior alveolar nerve canal based on the target parameters. The target extraction risk assessment model can also cross-validate the lesion status of the target object's mandibular second molar based on the target CBCT image information and the target intraoral scan information, and determine the target object's past history related to extraction risk based on the target medical record information. The target parameters include the shortest distance between the crown of the mandibular third molar and the distal surface of the mandibular second molar, the estimated contact area between the crown of the mandibular third molar and the mandibular second molar, the axial angle of the mandibular third molar, the surface contact volume between the proximal root of the mandibular third molar and the root surface of the mandibular second molar, the three-dimensional shortest path distance between the root apex of the mandibular third molar and the inferior alveolar nerve canal, the overlapping projection area of ​​the shadow of the mandibular third molar and the shadow of the inferior alveolar nerve canal, the angle difference between the long axis of the mandibular third molar and the direction of the inferior alveolar nerve canal, and the volume ratio of the overlapping area in three-dimensional space. The pathological status of the mandibular second molar of the target subject is determined based on the proximal caries, residual tooth structure, external resorption of the root, restoration status, and the health status of the mandibular soft tissue of the target subject. The physical structure of the target subject was determined by the low-density radiolucent area of ​​the mandibular second molar obtained from the target CBCT image information. The external resorption of the mandibular second molar root was determined by the root contour defect of the mandibular second molar obtained from the target CBCT image information. The restoration status of the target subject was determined by whether the target subject had restorations or fillings and the restoration method, obtained from the target CBCT and intraoral scan information. The soft tissue health status of the target subject's mandible was assessed by the gingival margin, free gingiva, attached gingiva, and oral mucosa of the mandibular third and second molars obtained from the target intraoral scan information. The target subject's past history of tooth extraction risk was determined based on the target medical record information. The target medical record information included the target subject's oral medical history and non-oral medical history that met the oral relevance criteria.

[0044] In an optional embodiment, step S3 above, specifically including enabling the target tooth extraction risk assessment model to determine the target tooth extraction risk level and output target tooth extraction risk warning information based on the fusion of medical rules and AI scoring algorithms, includes: The target tooth extraction risk assessment model extracts the rule judgment parameters of the target object and independently matches each parameter in the rule judgment parameters of the target object in parallel based on medical rules. After traversing all medical rules, the target rule matching result is obtained. At the same time, the target tooth extraction risk assessment model extracts the AI ​​judgment parameters of the target object, inputs the AI ​​judgment parameters of the target object into the AI ​​scoring algorithm, outputs the target tooth extraction risk score, and obtains the target AI score result based on the target tooth extraction risk score and the AI ​​scoring classification rules. The target tooth extraction risk assessment model integrates the target rule matching results and the target AI scoring results to determine the target tooth extraction risk level and outputs target tooth extraction risk warning information; The rule-based judgment parameters include the impaction angle of the mandibular third molar, the horizontal position of the mandibular third molar, the vertical position of the mandibular third molar, the contact state between the mandibular third molar and adjacent teeth, the distance between the mandibular third molar and the inferior alveolar nerve canal, the lesion state of the mandibular second molar, and the history of extraction risk. The AI ​​judgment parameters include the impaction angle of the mandibular third molar, the vertical depth between the crown and the ulnar surface of the mandibular third molar, the shortest distance from the root of the mandibular third molar to the inferior alveolar nerve, the contact area between the mandibular third molar and adjacent teeth, the CT value change of the distal proximal surface of the mandibular third molar, the restoration type label, and the history of extraction risk.

[0045] Specifically, the medical rules are constructed based on literature evidence and actual surgical experience, covering structural combination logic highly correlated with tooth extraction complications. They are modeled using an IF-THEN rule structure, with each rule triggered by a set of structural variable conditions, and then determined to a specific risk level. The rule set operates in parallel, with each rule matching independently and without priority over others. All rules are traversed in each case, and the results of all successfully matched rules are statistically analyzed. If multiple rules are activated, the "highest risk level" is the default output of the rule module; if no rule is matched, the rule module does not participate in the current round of fusion decision-making. After each rule judgment, the following will be output: the number and logical path of the successfully matched rule; and the activated key variables (including their specific values).

[0046] The risk level definition can be modified; only one method of setting the rules for a single risk level is shown here: High risk (meeting any one of the following): Impacted tooth with mesial tilt >40° and distal contact area with second molar >4.5mm², and adjacent tooth CBCT image shows distal caries or the presence of restoration; Impacted tooth root apex to the nearest distance from the inferior alveolar nerve canal <1.5mm, and there is an overlap area of ​​nerve canal-root surface >2mm³; Low vertical depth + Class III + Impacted tooth root apex to the nearest distance from the inferior alveolar nerve canal <1.5mm, and adjacent tooth has a restoration.

[0047] Medium risk (meeting any one of the following): The mandibular third molar is horizontally impacted at a moderate depth. The shortest distance from the root apex of the impacted tooth to the inferior alveolar nerve canal is <1.5mm but there is no overlap. The adjacent teeth are not decayed but have fillings. Part of the crown of the mandibular third molar is covered by the mandibular ramus, and part of the crown enters the mandibular ramus bone with less than half the space. The axial angle is 25°-40°. There is no damage to the adjacent teeth, but the shortest distance from the root apex of the impacted tooth to the inferior alveolar nerve canal is 1.5-2.5mm. There is no contact with adjacent teeth. Most or all of the crown of the third molar is located in the mandibular ramus bone, but the root apex is more than 3mm away from the nerve canal. The root morphology is abnormal.

[0048] Low risk (meeting all items): The third molar has a crown completely in front of the ascending ramus of the mandible. The mesiodistal diameter of the crown is entirely before the distal midline of the second molar, with sufficient space for eruption + high position + vertical impaction. The distance between the crown and adjacent teeth > 2 mm, not in contact; The closest distance from the apex of the impacted tooth root to the inferior alveolar nerve canal > 3 mm, without abutting / overlapping; There are no treatment records in the adjacent tooth images and medical records.

[0049] In an optional embodiment, the AI scoring algorithm in the above step S3 adopts the gradient boosting tree algorithm, whose training objective function is binary cross-entropy, and the optimization metrics are AUC and / or recall rate. An early stopping strategy is used to prevent overfitting.

[0050] Preferably, the AI scoring algorithm adopts the XGBoost algorithm, and its calculation formula is: First, obtain the logit through the ensemble tree: ; where, γ0: global bias term; K: the total number of decision trees in XGBoost; f k (x): the predicted value of the k-th decision tree for the input feature x; x: input feature vector; Then, convert it to a probability score P between 0 and 1 through the sigmoid function: .

[0051] Set the risk level classification rules according to the P value: P > 0.85 is determined as high risk, 0.65 < P ≤ 0.85 is medium risk, and P ≤ 0.65 is low risk. The model is supervised-trained on historical case data with clear outcome labels (such as surgical difficulty, postoperative nerve injury, adjacent tooth loosening, postoperative infection) after surgery. The training objective function is binary cross-entropy, and the optimization metrics are AUC and / or recall rate. The training set and the validation set are divided in an 8:2 ratio, and an early stopping strategy is used to prevent overfitting.

[0052] In an optional embodiment, to improve the model interpretability, the method may further include: Perform variable influence decomposition on the output of the AI scoring algorithm.

[0053] Specifically, each scoring output is accompanied by a dominant feature ranking, listing the top n variables that contribute the most to the risk value composition and calculating their relative weights. For example, for a certain case with an output of P = 0.91, the system will display "Risk dominant factors: relationship between the cusp of the third molar and the nerve canal (27%), adjacent tooth contact area (22%), impaction angle (18%)". The interpretation result can be presented through a graphical interface to assist the doctor in understanding the scoring basis.

[0054] In an optional embodiment, the target tooth extraction risk assessment model in step S3 above determines the target tooth extraction risk level based on a fusion of medical rules and AI scoring algorithms, and outputs target tooth extraction risk warning information. The specific fusion determination rules are as follows: If both AI and medical rules determine the same risk level (e.g., both "high risk"), that level will be used as the final output. If the results differ, a pre-defined fusion strategy will be applied: if the AI ​​score P exceeds a set confidence threshold (e.g., P>0.85), the risk level can be upgraded even if the medical rule only outputs "medium risk". If the AI ​​score is between [0.65, 0.85], it will check for activated high-weight rule paths; if so, the high-risk output will be maintained. If the AI ​​score is below 0.65, but the medical rule hits a high-risk path, the medical rule output will be used, and the result will be marked as "rule-priority intervention". The fusion judgment output includes: risk level (high, medium, low); numerical score (AI score P); hit rule path number and content; dominant variable contribution (SHAP weights show the n features with the largest contribution); and a risk description field (automatically generated structured text, such as: "This case is assessed as high risk due to the impacted mandibular third molar with an angle of 46° and a contact area of ​​5.2 mm², and distal caries of the adjacent tooth"), supporting doctors in result tracking and risk interpretation.

[0055] In an optional embodiment, the method may further include: Obtain human input and / or real-time updated information and optimize the model.

[0056] Specifically, all scoring logs (including structural variables, judgment levels, and final decision paths) are stored in a structured database for subsequent model tuning and risk rule retraining. Doctors can annotate the output with "acceptance / rejection" on the front-end interface, creating a manually corrected dataset. When the cumulative number of feedback cases reaches a specified threshold (≥50 cases per category), the model retraining process will be triggered, creating a fine-tuned sub-model while retaining the original model parameters. Manual rule editing is also supported, allowing doctors to register newly observed high-risk structural patterns as rule entries. All updates are recorded with version numbers, and the system can trace the historical scoring changes of each model version for the same case, ensuring safe iteration.

[0057] In an optional embodiment, the method may further include: Based on the target tooth extraction risk warning information and the basic database, generate clinical demonstration information and surgical procedure recommendation information; The basic database includes a literature database, a multimedia database, a technique knowledge base, and local technique data.

[0058] Specifically, standardized text descriptions are generated based on tooth extraction risk warnings, detailing the impaction type, contact pattern, lesion condition, and source of risk. These descriptions are simultaneously displayed on a constructed visual 3D model, automatically annotating the tilt direction, adjacent tooth contact surfaces, and key anatomical sites in the nerve pathway proximity zone, supporting dynamic rotation and custom cross-sectional observation. Furthermore, it connects to cloud-based literature databases, multimedia libraries, and surgical procedure knowledge bases, as well as local surgical materials via API interfaces. It automatically retrieves highly relevant literature based on risk tags, providing brief abstracts and external links. The surgical procedure knowledge base creates a mapping rule table between impaction types and surgical plans, allowing for the push of key surgical terms and links to external video materials. If teaching videos, surgical diagrams, or operation flowcharts have been uploaded locally, the corresponding content can be automatically retrieved as reference materials.

[0059] In an optional embodiment, the method may further include: A search index is built based on uploaded custom procedure tags and / or local procedure data.

[0060] In an optional embodiment, the method may further include: New conditions can be added or risk assessment weights can be fine-tuned through manual operation.

[0061] It should be noted that the concepts of "sample" related terms and "target" related terms involved in the above embodiments correspond one-to-one. The difference is that "sample" related terms are used in the model training process, while "target" related terms are used in the model application process.

[0062] As can be seen from the above technical solutions of the embodiments of this application, the embodiments of this application integrate CBCT image analysis, intraoral scan information processing and structured medical record information, automatically identify the spatial relationship and pathological condition of the mandibular third molar, and output tooth extraction risk warning information, providing relevant surgical references, which can effectively improve the scientificity and accuracy of surgical planning, reduce the incidence of surgical complications, ensure patient treatment safety and improve clinical treatment efficiency.

[0063] Corresponding to the mandibular third molar extraction risk assessment method provided in the above embodiments, this application also provides a mandibular third molar extraction risk assessment system. Since the mandibular third molar extraction risk assessment system provided in this application corresponds to the mandibular third molar extraction risk assessment method provided in the above embodiments, the implementation method of the aforementioned mandibular third molar extraction risk assessment method is also applicable to the mandibular third molar extraction risk assessment system provided in this embodiment, and will not be described in detail in this embodiment.

[0064] Please see Figure 3 The diagram shown is a structural block diagram of a mandibular third molar extraction risk assessment system provided in an embodiment of this application; the system includes: 01: Information acquisition module, used to acquire target acquisition information corresponding to the target object; target acquisition information includes target intraoral scan information, target CBCT image information, and target medical record information; 02: Information processing module, used to input the target intraoral scan information and target CBCT image information into the target information processing model, so that the target information processing model can process the target intraoral scan information and target CBCT image information to obtain target risk assessment information; 03: Risk Assessment Module. This module inputs target risk assessment information and target medical record information into the target extraction risk assessment model. The model then comprehensively quantifies the proximity of the target subject's mandibular third molar to the mandibular second molar and the inferior alveolar nerve canal based on the target risk assessment information. It also cross-validates the target subject's mandibular second molar lesion status and extraction risk-related medical history based on the target risk assessment information and target medical record information. Furthermore, the module enables the target extraction risk assessment model to determine the target extraction risk level based on a fusion of medical rules and AI scoring algorithms, and outputs target extraction risk warning information. This warning information includes the target extraction risk level corresponding to the target subject and extraction-related information that meets preset rules.

[0065] In an optional embodiment, the system supports dynamic model updates and user interaction feedback mechanisms. All scoring logs (including structural variables, judgment levels, and final decision paths) are stored in a structured database for subsequent model tuning and risk rule retraining. Doctors can annotate the system output with "approval / rejection" on the front-end interface, forming a manual correction dataset. When the cumulative number of feedback cases reaches a specified threshold (≥50 cases per category), the system will trigger a model retraining process, forming a fine-tuned sub-model while retaining the original model parameters. Manual rule editing is also supported, allowing doctors to register newly observed high-risk structural patterns as rule entries. All updates are recorded with version numbers, and the system can trace the historical score changes of each model version for the same case to ensure safe iteration.

[0066] In an optional embodiment, the system integrates an embedded clinical demonstration and surgical strategy support module to provide doctors with diagnostic interpretation, evidence-based support, and surgical procedure recommendations. The module first generates a standardized text description based on extraction risk warning information, detailing the impaction type, contact pattern, lesion condition, and risk source. Simultaneously, the text description is displayed on a constructed visual 3D model, automatically annotating the tilt direction, adjacent tooth contact surfaces, and key anatomical sites in the nerve pathway proximity area, supporting dynamic rotation and sectional observation. It also connects to cloud-based literature databases, multimedia databases, and surgical procedure knowledge bases, as well as local surgical procedure data via API interfaces. Based on risk tags, it automatically retrieves highly relevant literature and provides brief abstracts and external links. The surgical procedure knowledge base uses a mapping rule table to establish impaction types and surgical plans. For example, in cases of horizontal impaction with nerve proximity, the "crown splitting + root resection + bone window" procedure is recommended. Key surgical terms can be pushed, and external video materials can be linked. If teaching videos, surgical procedure diagrams, or operation flowcharts have been uploaded locally, the corresponding content can be automatically retrieved as reference materials.

[0067] In one optional embodiment, the system possesses high customizability and external integration capabilities. It supports physicians in defining custom surgical procedure tags, uploading local surgical procedure data, and creating search indexes; the system's rule base can be expanded with new conditions or risk assessment weights fine-tuned through the administrator interface; the system's surgical procedure recommendation and discrimination models can interface with external CDSS, PACS, or intelligent imaging workstations for seamless integration. The system also provides a keyword-based surgical procedure search entry, allowing physicians to input surgical procedure keywords based on case characteristics. The system will then connect to external video platforms or medical literature engines, outputting results and displaying them in an embedded manner.

[0068] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0069] The mandibular third molar extraction risk assessment system of this application integrates CBCT image analysis, intraoral scan information processing, and structured medical record information. It automatically identifies the spatial relationship and pathological condition of the mandibular third molar and outputs extraction risk warning information, providing relevant surgical references. It can effectively improve the scientificity and accuracy of surgical planning, reduce the incidence of surgical complications, ensure patient safety, and improve clinical treatment efficiency.

[0070] This application also provides an electronic device, including a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the processor loads and executes the at least one instruction or at least one program to implement the mandibular third molar extraction risk assessment method provided in the above method embodiments.

[0071] Memory can be used to store software programs and modules. The processor executes these stored software programs and modules to perform various functional applications and achieve advanced autonomous driving. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for functions, etc.; the data storage area can store data created based on device usage, etc. Furthermore, memory can include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, memory can also include a memory controller to provide the processor with access to the memory.

[0072] The method embodiments provided in this application can be executed in a computer terminal, server or similar computing device, that is, the above-mentioned electronic device may include a computer terminal, server or similar computing device. Figure 4 This is a hardware structure block diagram of an electronic device for running a method for assessing the risk of mandibular third molar extraction, as provided in an embodiment of this application. Figure 4 As shown, the internal structure of this electronic device may include, but is not limited to, a processor, a network interface, and a memory. The processor, network interface, and memory within the electronic device can be connected via a bus or other means, as illustrated in the embodiments of this specification. Figure 4 Taking the example of a connection between China and Israel via a bus.

[0073] The processor (or CPU, Central Processing Unit) is the computing and control core of the electronic device. The network interface may optionally include standard wired interfaces or wireless interfaces (such as Wi-Fi, mobile communication interfaces, etc.). Memory is the storage device in the electronic device used to store programs and data. It is understood that the memory here can be a high-speed RAM storage device or a non-volatile memory device, such as at least one disk storage device; optionally, it can also be at least one storage device located remotely from the aforementioned processor. The memory provides storage space containing the operating system of the electronic device, which may include, but is not limited to: Windows (an operating system), Linux (an operating system), Android (a mobile operating system), iOS (a mobile operating system), etc., and this application does not limit this; furthermore, the storage space also contains one or more instructions suitable for loading and execution by the processor, which may be one or more computer programs (including program code). In the embodiments of this specification, the processor loads and executes one or more instructions stored in the memory to implement the mandibular third molar extraction risk assessment method provided in the above method embodiments.

[0074] This application also provides a computer-readable storage medium storing at least one instruction or at least one program, wherein the at least one instruction or at least one program is loaded and executed by a processor to implement the mandibular third molar extraction risk assessment method provided in the method embodiment.

[0075] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0076] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than those shown in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multi-sample image classification and parallel processing are also possible or may be advantageous.

[0077] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0078] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. The above are merely preferred embodiments of this application and are not intended to limit the application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for assessing the risk of mandibular third molar extraction, characterized in that, The method includes: Acquire target acquisition information corresponding to the target object; the target acquisition information includes target intraoral scan information, target CBCT image information, and target medical record information; The target intraoral scan information and the target CBCT image information are input into the target information processing model so that the target information processing model can process the target intraoral scan information and the target CBCT image information to obtain target risk assessment information; The target risk assessment information and the target medical record information are input into the target extraction risk assessment model, so that the target extraction risk assessment model comprehensively quantifies the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target risk assessment information, and cross-validates the target object's mandibular second molar lesion status and extraction risk-related past history based on the target risk assessment information and the target medical record information; and so that the target extraction risk assessment model determines the target extraction risk level based on the fusion of medical rules and AI scoring algorithms and outputs target extraction risk warning information; the target extraction risk warning information includes the target extraction risk level corresponding to the target object and extraction-related information corresponding to the target object that meets preset rules; The process of the target tooth extraction risk assessment model determining the target tooth extraction risk level and outputting target tooth extraction risk warning information based on the fusion of medical rules and AI scoring algorithms specifically includes: The target tooth extraction risk assessment model extracts the rule judgment parameters of the target object, and independently matches each parameter in the rule judgment parameters of the target object in parallel based on the medical rules. After traversing all the medical rules, the target rule matching result is obtained. At the same time, the target tooth extraction risk assessment model extracts the AI ​​judgment parameters of the target object, inputs the AI ​​judgment parameters of the target object into the AI ​​scoring algorithm, outputs the target tooth extraction risk score, and obtains the target AI score result based on the target tooth extraction risk score and the AI ​​scoring classification rules. The target tooth extraction risk assessment model integrates the target rule matching result and the target AI scoring result to determine the target tooth extraction risk level and outputs the target tooth extraction risk warning information; The rule-based judgment parameters include the impaction angle of the mandibular third molar, the horizontal position of the mandibular third molar, the vertical position of the mandibular third molar, the contact state between the mandibular third molar and adjacent teeth, the distance between the mandibular third molar and the inferior alveolar nerve canal, the lesion state of the mandibular second molar, and the history related to extraction risk. The AI ​​judgment parameters include the impaction angle of the mandibular third molar, the vertical depth between the crown and the ulnar surface of the mandibular third molar, the shortest distance from the root of the mandibular third molar to the inferior alveolar nerve, the contact area between the mandibular third molar and adjacent teeth, the CT value change of the distal proximal surface of the mandibular third molar, the restoration type label, and the history related to extraction risk.

2. The method for assessing the risk of mandibular third molar extraction according to claim 1, characterized in that, The step of inputting the target intraoral scan information and the target CBCT image information into the target information processing model, so that the target information processing model processes the target intraoral scan information and the target CBCT image information to obtain target risk assessment information, specifically includes: The target intraoral scan information and the target CBCT image information are input into the target information processing model, so that the target information processing model performs rigid registration on the target intraoral scan information and the target CBCT image information to establish a unified spatial coordinate system; and so that the target information processing model performs voxel-by-voxel automatic segmentation on the target CBCT image information based on a pre-trained three-dimensional deep convolutional neural network to extract the target structure corresponding to the target object, and performs tooth and soft tissue segmentation on the target intraoral scan information based on a pre-trained deep hierarchical neural network to extract the soft tissue morphology of the free gingiva, attached gingiva and oral mucosa corresponding to the target object, thereby obtaining the target risk assessment information.

3. The method for assessing the risk of mandibular third molar extraction according to claim 2, characterized in that, The three-dimensional deep convolutional neural network adopts an improved 3DnnUNet, which is based on the native nnUNet framework, embeds an attention mechanism, and adds loss functions at different levels of the decoder; the deep hierarchical neural network adopts PointNet++.

4. The method for assessing the risk of mandibular third molar extraction according to claim 2, characterized in that, The target risk assessment information includes target acquisition information and a target visualization 3D model; the target visualization 3D model carries target 3D axial classification information, target horizontal displacement grading information, and target vertical position grading information; the target 3D axial classification information indicates the impaction direction type of the target object's mandibular third molar, obtained by combining the angle between the long axis of the target object's mandibular third molar and the normal vector of the standard occlusal plane with the tooth projection direction; the target horizontal displacement grading information indicates the horizontal displacement level of the target object's mandibular third molar, obtained by combining the horizontal overlap between the crown of the target object's mandibular third molar and the anterior edge of the mandibular ramus with the Pell & Gregory classification standard; the target vertical position grading indicates the depth level of the target object's mandibular third molar, obtained by combining the vertical distance from the apex of the crown of the target object's mandibular third molar to the occlusal surface of the adjacent tooth with the Pell & Gregory classification standard.

5. The method for assessing the risk of mandibular third molar extraction according to claim 1, characterized in that, The process involves inputting the target risk assessment information and the target medical record information into a target extraction risk assessment model. This model comprehensively quantifies the proximity of the target subject's mandibular third molar to the mandibular second molar and the inferior alveolar nerve canal based on the target risk assessment information. Furthermore, it cross-validates the target subject's mandibular second molar lesion status and extraction risk-related medical history based on the target risk assessment information and the target medical record information. Specifically, this includes: The target risk assessment information and the target medical record information are input into the target extraction risk assessment model so that the target extraction risk assessment model obtains the target parameters of the target object based on the target risk assessment information. The target extraction risk assessment model comprehensively quantifies the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target parameters. The target extraction risk assessment model also cross-validates the lesion status of the target object's mandibular second molar based on the target CBCT image information and the target intraoral scan information, and determines the target object's past history related to extraction risk based on the target medical record information. The target parameters include the shortest distance between the crown of the mandibular third molar and the distal surface of the mandibular second molar, the estimated contact area between the crown of the mandibular third molar and the mandibular second molar, the axial angle of the mandibular third molar, the surface contact volume between the proximal root of the mandibular third molar and the root surface of the mandibular second molar, the three-dimensional shortest path distance between the root apex of the mandibular third molar and the inferior alveolar nerve canal, the overlapping projection area of ​​the shadow of the mandibular third molar and the shadow of the inferior alveolar nerve canal, the angle difference between the long axis of the mandibular third molar and the direction of the inferior alveolar nerve canal, and the volume ratio of the overlapping area in three-dimensional space. The pathological state of the mandibular second molar of the target object is determined based on the proximal caries, residual tooth structure, external root resorption, restoration status, and the health status of the mandibular soft tissue of the target object. The proximal caries and residual tooth structure of the mandibular second molar of the target object are determined based on the target parameters. The low-density radiolucent area of ​​the mandibular second molar obtained from CBCT imaging information is used to determine the external resorption of the mandibular second molar root, which is determined by the root contour defect of the mandibular second molar obtained from the target CBCT imaging information. The restoration status of the target object is determined by whether the target object has a restoration or filling and the restoration method, which is determined by the target CBCT and intraoral scan information. The mandibular soft tissue health status of the target object is assessed by the gingival margin, free gingiva, attached gingiva, and oral mucosa of the mandibular third and second molars, which are determined by the target intraoral scan information. The target object's past history of tooth extraction risk is determined based on the target medical record information. The target medical record information includes the target object's oral medical history and non-oral medical history that meets the oral medical relevance criteria.

6. A method for assessing the risk of mandibular third molar extraction according to any one of claims 1-5, characterized in that, The AI ​​scoring algorithm uses the gradient boosting tree algorithm, with its training objective function being binary cross-entropy and its optimization metrics being AUC and / or recall. An early stopping strategy is used to prevent overfitting.

7. The method for assessing the risk of mandibular third molar extraction according to claim 6, characterized in that, The method further includes: Based on the target tooth extraction risk warning information and the basic database, clinical demonstration information and surgical procedure recommendation information are generated. The basic database includes a literature database, a multimedia database, a technique knowledge base, and local technique data.

8. A risk assessment system for mandibular third molar extraction, implemented using the risk assessment method for mandibular third molar extraction as described in any one of claims 1 to 7, characterized in that, The system includes: The information acquisition module is used to acquire target acquisition information corresponding to the target object; the target acquisition information includes target intraoral scan information, target CBCT image information, and target medical record information; The information processing module is used to input the target intraoral scanning information and the target CBCT image information into the target information processing model, so that the target information processing model can process the target intraoral scanning information and the target CBCT image information to obtain target risk assessment information; The risk assessment module is used to input the target risk assessment information and the target medical record information into the target tooth extraction risk assessment model, so that the target tooth extraction risk assessment model comprehensively quantifies the proximity of the target object's mandibular third molar, mandibular second molar, and inferior alveolar nerve canal based on the target risk assessment information, and cross-validates the target object's mandibular second molar lesion status and extraction risk-related past history based on the target risk assessment information and the target medical record information; and so that the target tooth extraction risk assessment model determines the target tooth extraction risk level based on the fusion of medical rules and AI scoring algorithms and outputs target tooth extraction risk warning information; the target tooth extraction risk warning information includes the target tooth extraction risk level corresponding to the target object and extraction-related information corresponding to the target object that meets preset rules.

9. An electronic device, characterized in that, The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement a method for assessing the risk of extraction of the mandibular third molar as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing at least one instruction or at least one program, said at least one instruction or said at least one program being loaded and executed by a processor to implement a method for assessing the risk of extraction of a mandibular third molar as described in any one of claims 1 to 7.

Images