An artificial intelligence-based preoperative evaluation system for maxillary posterior tooth implantation

Abstract

The application provides a maxillary posterior tooth area implantation preoperative evaluation system based on artificial intelligence, which comprises the following steps: constructing an oral cavity CBCT image library, then constructing a maxillary sinus basic model, an oral cavity maxillary posterior tooth area implantation decision design model and an oral cavity maxillary posterior tooth area windowing bone grafting decision design model, and constructing a maxillary sinus downstream task model according to the maxillary sinus basic model; inputting the oral cavity CBCT image data to be measured into the maxillary sinus downstream task model, and generating a first decision according to the qualitative and quantitative characteristics output by the oral cavity maxillary posterior tooth area implantation decision design model or generating a second decision according to the oral cavity maxillary posterior tooth area windowing bone grafting decision design model, so as to perform preoperative evaluation. The application realizes intelligent identification of different characteristics in the oral cavity CBCT image in the whole process, generates an optimal oral cavity maxillary posterior tooth area implantation decision, and performs preoperative evaluation of the oral cavity maxillary posterior tooth area implantation, thereby improving the reliability of the preoperative evaluation of the oral cavity maxillary posterior tooth area implantation.

Description

Technical Field

[0001] This invention relates to the field of medical image processing, and in particular to an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation. Background Technology

[0002] In my country, the incidence of missing teeth is high, reaching 80%-90% in people over 60 years old, with up to one-third of these cases involving the loss of upper posterior teeth. However, due to the influence of the maxillary sinus, implant restorations for missing upper posterior teeth often face the problem of insufficient bone volume in the implant area. Furthermore, the condition within the maxillary sinus frequently affects the decision-making process for implant placement in the maxillary posterior region. Cone-beam computed tomography (CBCT) is a commonly used imaging method in oral medicine, reflecting oral anatomical information such as teeth, maxillary sinus, and mandibular canal. In implant restorations for missing upper posterior teeth, dentists often need to conduct a comprehensive preoperative evaluation of the patient's CBCT scan to obtain quantitative indicators relevant to implant decisions, such as the height of remaining alveolar bone, width of the maxillary sinus, sinus angle, and thickness of the bone wall at the fenestration, as well as qualitative indicators such as sinus floor morphology, presence of mucosal thickening, presence of polyps, and patency of the sinus ostium. Treatment decisions for missing upper posterior teeth are then made based on these qualitative and quantitative characteristics. For example, according to the guidelines of the International Society for Implant Dentistry, direct implantation is recommended when the remaining alveolar bone height in the area of ​​missing upper posterior teeth is greater than 6mm. If the remaining alveolar bone height is less than 6mm, bone augmentation surgery is recommended first, with different bone augmentation techniques, including guided bone regeneration, intrasinus lift, and extrasinus lift, selected based on the specific bone volume. After determining the treatment decision for missing upper posterior teeth, a surgical plan needs to be developed accordingly: when direct implantation is recommended, the appropriate implant size should be selected and its three-dimensional position determined using preoperative implantation simulation software; when extrasinus lift is recommended, the location and size of the bone graft should be designed preoperatively. However, in clinical practice, the preoperative assessment steps for implantation of missing upper posterior teeth are complex and involve numerous indicators, and are highly dependent on the physician's personal experience, making it difficult to guarantee the consistency of the assessment. Therefore, the preoperative assessment process for implant surgery in the maxillary posterior region can be summarized as follows: 1) Assessment of anatomical indicators of the maxillary posterior region based on CBCT; 2) Making implant treatment decisions based on a comprehensive assessment of anatomical indicators of the maxillary posterior region according to evidence-based clinical pathways; 3) Planning surgical procedures based on CBCT, guided by the implant treatment decisions.

[0003] However, thanks to the development of artificial intelligence and deep learning technologies, intelligent analysis tools for medical images (such as CBCT) are gradually being applied in clinical specialties to achieve automatic, rapid, and accurate image analysis. However, the maxillary posterior region relies heavily on large volumes of CBCT images for evaluation, and the decision indicators exhibit complex data characteristics with multiple modalities and properties. Existing technologies face the following difficulties in achieving intelligent decision-making design for the entire maxillary posterior region implantation process: 1) Maxillary posterior region implant restoration involves complex qualitative and precise quantitative indicators. Traditional deep learning models typically employ supervised learning systems, are usually designed to solve specific tasks, and rely heavily on expert judgment and manually labeled data, which is time-consuming and labor-intensive, making it difficult to achieve accurate and intelligent prediction of numerous maxillary posterior region implantation decision-related indicators; 2) Maxillary posterior region implantation... Posterior tooth implantation decisions involve numerous authoritative literature and clinical guidelines, requiring a high level of knowledge from doctors and making them prone to decision-making errors. Existing tools can only present raw data and cannot automatically map preoperative assessment results to evidence-based decisions. 3) When designing surgical plans based on treatment decisions, doctors need to manually sketch the implant placement sites and the location and size of the external lift window based on their experience. Different doctors' judgments are highly subjective, and young doctors often make decision-making biases due to insufficient anatomical identification. Existing tools cannot directly achieve intelligent design of upper posterior tooth implantation simulation plans and maxillary sinus external lift window bone grafting plans based on CBCT. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation. It designs innovative solutions tailored to the characteristics of each step in the entire preoperative assessment process for maxillary posterior tooth implantation: 1) Training a self-supervised learning-based maxillary sinus baseline model to learn the general anatomical features of the maxillary posterior tooth region from a large amount of unlabeled CBCT data, enabling intelligent prediction of complex qualitative and precise quantitative indicators without relying on labeled data; 2) Constructing a decision rule engine for maxillary posterior tooth implantation, transforming expert knowledge bases and extensive clinical experience into quantifiable rule conditions and decision-making action mapping relationships, automatically mapping qualitative and quantitative indicators to corresponding implantation clinical decisions; 3) Constructing an intelligent simulation model for maxillary posterior tooth implantation and an intelligent simulation model for maxillary sinus lift fenestration and bone grafting based on a lightweight three-dimensional segmentation network, automatically, intelligently, and visually completing the personalized design and output of surgical plans based on implantation treatment decisions.

[0005] To achieve the above objectives, this invention provides an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation, comprising: an image library construction module, a first model construction module, a second model construction module, and an assessment module. The image library construction module is used to construct an oral CBCT image library based on several oral CBCT image data. The first model construction module is used to construct a maxillary sinus basic model, an oral maxillary posterior tooth implantation decision design model, and an oral maxillary posterior tooth fenestration bone grafting decision design model based on the oral CBCT image library. The second model construction module is used to construct a maxillary sinus downstream task model based on the weights of the maxillary sinus basic model. The assessment module is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model, and, based on the qualitative and quantitative features output, call the oral maxillary posterior tooth implantation decision design model to generate a first decision or call the oral maxillary posterior tooth fenestration bone grafting decision design model to generate a second decision, so as to perform preoperative assessment of maxillary posterior tooth implantation through the first decision or the second decision.

[0006] This invention proposes an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation. An image library construction module builds an oral CBCT image library using several oral CBCT images, providing a data foundation for the first model construction module to train the maxillary sinus basic model, the maxillary posterior tooth implantation decision design model, and the maxillary posterior tooth fenestration bone grafting decision design model, ensuring the reliability of the model construction. Then, a second model construction module constructs a downstream maxillary sinus task model with weights from the maxillary sinus basic model, addressing the instability issue that arises when the downstream maxillary sinus task model exhibits data class imbalance in downstream tasks. The assessment module then combines the maxillary sinus downstream... The task model automatically extracts qualitative and quantitative features from the oral CBCT image data to be tested, and adaptively calls the decision design model for implantation in the maxillary posterior teeth region or the decision design model for fenestration and bone grafting in the maxillary posterior teeth region based on the qualitative and quantitative features. It then generates the first or second decision accordingly to conduct preoperative assessment of implantation in the maxillary posterior teeth region. This achieves full-process automation from data feature identification to decision generation and preoperative assessment. The multi-task model architecture realizes intelligent quantitative and qualitative image analysis and decision design model association calling throughout the entire process. By replacing manual experience judgment with an indicator-driven model calling mechanism, the objectivity and reliability of implantation decision generation are significantly improved, and the reliability of preoperative implant assessment is enhanced.

[0007] Furthermore, the image library construction module is used to construct an oral CBCT image library based on several oral CBCT image data, including: a first data acquisition unit, a data filtering unit, a research site selection unit, and an image cropping unit; the first data acquisition unit is used to acquire several oral CBCT image data; the data filtering unit is used to filter out image data with motion artifacts and metal artifacts based on a preset image processing algorithm to obtain initial oral CBCT image data; the research site selection unit is used to adjust the head position in each initial oral CBCT image data based on a preset reference plane and select the research site; the image cropping unit is used to crop the cross section of the research site of each initial oral CBCT image data to obtain the oral CBCT image library.

[0008] In the above scheme, image data containing motion artifacts and metal artifacts are screened out to eliminate irrelevant data interference and improve the accuracy of the data. In addition, by adjusting the head position in the data to a preset reference plane, selecting the research site, and cropping the corresponding interface according to the research site, a standardized atlas library is provided for subsequent network and model training, thereby improving the reliability of preoperative assessment for implantation.

[0009] Furthermore, the first model construction module is used to construct a basic model of the maxillary sinus, a decision-making design model for implantation in the maxillary posterior teeth region, and a decision-making design model for fenestration and bone grafting in the maxillary posterior teeth region based on the oral CBCT image library. This includes: a target detection network construction unit, a pre-training dataset acquisition unit, a basic maxillary sinus model construction unit, a decision-making design model construction unit for implantation in the maxillary posterior teeth region, and a decision-making design model construction unit for fenestration and bone grafting in the maxillary posterior teeth region. The target detection network construction unit is used to select a predetermined first number of oral CBCT image data from the oral CBCT image library and perform region-of-interest (ROI) labeling to obtain the target detection network. The pre-training dataset acquisition unit is used to perform RIO labeling on all oral CBCT image data in the oral CBCT image library based on the target detection network to obtain the pre-training dataset. The basic maxillary sinus model construction unit... The first unit is used to construct a basic model of the maxillary sinus based on a pre-trained dataset. The second unit is used to select a second set of oral CBCT images from the oral CBCT image library, generate a simulated implantation plan for the maxillary posterior teeth region using a preset three-dimensional deep learning algorithm, and construct an implantation decision design model for the maxillary posterior teeth region based on the simulated implantation plan and the oral CBCT image data. The third unit is used to select a third set of oral CBCT images from the oral CBCT image library, generate a simulated fenestration bone grafting plan for the maxillary posterior teeth region using a preset three-dimensional deep learning algorithm, and construct a fenestration bone grafting decision design model for the maxillary posterior teeth region based on the simulated fenestration bone grafting plan and the oral CBCT image data.

[0010] In the above scheme, a portion of oral CBCT image data is first selected from the oral CBCT image database to mark regions of interest (ROIs), and a target detection network is constructed to reduce data computational redundancy. Then, the target detection network is used to identify ROIs from all oral CBCT image data in the database, filtering out cross-sectional data that do not contain ROIs to reduce the interference of useless data on subsequent model training and improve the reliability of subsequent maxillary sinus basic model construction and training. Finally, a portion of oral CBCT image data is selected from the oral CBCT image database and a 3D deep learning algorithm is used to generate the maxillary posterior tooth region... Implant placement simulation scheme and maxillary posterior tooth region fenestration bone grafting simulation scheme are combined with oral CBCT imaging data to construct maxillary posterior tooth region implant decision design model and maxillary posterior tooth region fenestration bone grafting decision design model. This establishes an objective mapping relationship between imaging features and clinical decisions, avoids subjective bias of human experience, and automatically outputs maxillary posterior tooth region implant decision design model and maxillary posterior tooth region fenestration bone grafting decision design model that conform to anatomical features and decision rules, thereby improving the reliability of preoperative implant assessment.

[0011] Furthermore, the object detection network construction unit is used to select a preset first number of oral CBCT image data from the oral CBCT image library for region of interest labeling to obtain the object detection network. This unit includes: a data selection subunit, a region of interest labeling subunit, a first dataset partitioning subunit, and a first model training subunit. The data selection subunit is used to select a preset first number of oral CBCT image data based on the oral CBCT image library. The region of interest labeling subunit is used to label the preset first number of oral CBCT image data for regions of interest based on a preset image processing algorithm, obtaining several labeled oral CBCT image data. The first dataset partitioning subunit is used to divide the several labeled oral CBCT image data into a first training set, a first validation set, and a first test set according to a preset ratio. The first model training subunit is used to train the initial object detection model using the training set, optimize the model parameters of the initial object model using the validation set, and adjust the hyperparameters of the optimized initial object model using the test set to obtain the object detection network.

[0012] In the above scheme, a portion of the oral CBCT image data is selected to mark the region of interest for training the target detection network. The target detection network is trained with a small amount of data so that it can learn to crop and divide the region of interest in the oral CBCT image data, thereby achieving automated region of interest division and improving the reliability of pre-implantation assessment.

[0013] Furthermore, the maxillary sinus basic model construction unit is used to construct a maxillary sinus basic model based on the pre-training dataset, including: a mesh construction subunit, an encoding subunit, a decoding subunit, a loss function calculation subunit, and a model update subunit; the mesh construction subunit is used to cut the pre-training dataset into uniform and non-overlapping meshes based on a preset self-supervised algorithm; the encoding subunit is used to select a preset ratio of meshes to set masks, and convert the unmasked meshes into vector sequences through an encoder; the decoding subunit is used to reconstruct the original image from the vector sequences and the masked meshes through a decoder, obtaining the reconstructed image; the loss function calculation subunit is used to calculate the corresponding loss function based on the reconstructed image and the original image; the model update subunit is used to update the preset ratio values, encoder weight settings, and mask setting modes, recalculate the corresponding loss function, and select a maxillary sinus basic model whose loss function meets the preset requirements.

[0014] In the above scheme, a self-supervised algorithm is used to divide the pre-training dataset into a uniform non-overlapping grid. Random masks are generated and the unmasked grids are converted into vector sequences to reconstruct the original image. The differences between the original image and the reconstructed image are then compared to calculate the loss function. By setting different conditions, the loss function under different conditions is obtained. The model corresponding to the optimal loss function is selected as the basic model of the maxillary sinus, ensuring that the model learns the features of CBCT images more fully, so as to build a more reliable basic model of the maxillary sinus and improve the reliability of preoperative assessment for implantation.

[0015] Furthermore, the oral posterior region implant decision design model construction unit is used to select a preset second number of oral CBCT image data from the oral CBCT image library, generate an oral posterior region implantation simulation plan through a preset three-dimensional deep learning algorithm, and construct an oral posterior region implantation decision design model based on the oral posterior region implantation simulation plan and oral CBCT image data, including: a first data preprocessing subunit, a first simulation plan acquisition subunit, a first decision dataset acquisition subunit, a second dataset partitioning subunit, and a second model training subunit; the first data preprocessing subunit is used to preprocess the preset second number of oral CBCT image data; the first simulation plan acquisition subunit is used to preprocess the oral CBCT image data through a preset three-dimensional deep learning algorithm. The data shows that the missing maxillary posterior teeth area is subjected to 3D reconstruction and implant simulation to obtain a simulated implantation plan for the maxillary posterior teeth region. The first decision dataset acquisition subunit is used to match the simulated implantation plan for the maxillary posterior teeth region with the corresponding oral CBCT image data to obtain the implantation decision dataset for the maxillary posterior teeth region. The second dataset partitioning subunit is used to divide the implantation decision dataset for the maxillary posterior teeth region into a second training set, a second validation set, and a second test set. The second model training subunit is used to train the preset initial image segmentation network model through the second training set, optimize the parameters of the preset initial image segmentation network model through the second validation set, and optimize the hyperparameters of the preset initial image segmentation network model through the second test set to obtain the implantation decision design model for the maxillary posterior teeth region.

[0016] In the above scheme, a portion of the selected oral CBCT image data is preprocessed. A three-dimensional deep learning algorithm is used to perform three-dimensional reconstruction and implant simulation of the maxillary posterior tooth missing area in the oral CBCT image data. The generated implant simulation plan for the maxillary posterior tooth region is matched with the corresponding oral CBCT image data to construct an implant decision dataset for the maxillary posterior tooth region. Then, a decision design model for implantation in the maxillary posterior tooth region is constructed by combining the dataset partitioning and the iterative optimization training mechanism of the three-dimensional image segmentation network. A three-dimensional deep learning algorithm is used to generate a simulated implantation plan for the maxillary posterior tooth region. Finally, the simulated implantation plan for the maxillary posterior tooth region and the corresponding oral CBCT image data are jointly used to construct an implant decision design model for the maxillary posterior tooth region. This establishes an objective mapping relationship between image features and clinical decisions, avoids subjective bias of human experience, and automatically outputs an implant decision design model for the maxillary posterior tooth region that conforms to anatomical features and decision rules, thereby improving the reliability of preoperative implant assessment.

[0017] Furthermore, the oral maxillary posterior region fenestration bone grafting decision design model construction unit is used to select a preset third number of oral CBCT image data from the oral CBCT image library, generate a simulated scheme for oral maxillary posterior region fenestration bone grafting through a preset three-dimensional deep learning algorithm, and construct an oral maxillary posterior region fenestration bone grafting decision design model based on the simulated scheme and the oral CBCT image data, including: a second data preprocessing subunit, a second simulation scheme acquisition subunit, a second decision dataset acquisition subunit, a third dataset partitioning subunit, and a third model training subunit; the second data preprocessing subunit is used to preprocess the preset third number of oral CBCT image data; the second simulation scheme acquisition subunit is used to preprocess the oral CBCT image data through a preset three-dimensional deep learning algorithm. The posterior tooth loss area is sequentially reconstructed in three dimensions, and the maxillary posterior tooth region is simulated for fenestration and bone grafting, resulting in a simulated scheme for maxillary posterior tooth region fenestration and bone grafting. The second decision dataset acquisition subunit is used to match the simulated scheme for maxillary posterior tooth region fenestration and bone grafting with corresponding oral CBCT image data, resulting in a decision dataset for maxillary posterior tooth region fenestration and bone grafting. The third dataset partitioning subunit is used to divide the decision dataset for maxillary posterior tooth region fenestration and bone grafting into a third training set, a third validation set, and a third test set. The third model training subunit is used to train a preset initial image segmentation network model using the third training set, optimize the parameters of the preset initial image segmentation network model using the third validation set, and optimize the hyperparameters of the preset initial image segmentation network model using the third test set, resulting in a decision design model for maxillary posterior tooth region fenestration and bone grafting.

[0018] In the above scheme, a portion of the selected oral CBCT image data is preprocessed. A 3D deep learning algorithm is used to perform 3D reconstruction of the maxillary posterior tooth missing area in the oral CBCT image data and to simulate the maxillary posterior tooth region fenestration and bone grafting area. The generated maxillary posterior tooth region fenestration and bone grafting simulation scheme is matched with the corresponding oral CBCT image data to construct a maxillary posterior tooth region fenestration and bone grafting decision dataset. Then, a maxillary posterior tooth region fenestration and bone grafting decision design model is constructed by combining dataset partitioning and a 3D image segmentation network iterative optimization training mechanism. A maxillary posterior tooth region fenestration and bone grafting simulation scheme is generated using a 3D deep learning algorithm. Finally, the maxillary posterior tooth region fenestration and bone grafting simulation scheme and the corresponding oral CBCT image data are jointly used to construct a maxillary posterior tooth region fenestration and bone grafting decision design model. This establishes an objective mapping relationship between image features and clinical decisions, avoids subjective biases from human experience, and automatically outputs a maxillary posterior tooth region fenestration and bone grafting decision design model that conforms to anatomical features and decision rules, thereby improving the reliability of pre-implantation assessment.

[0019] Furthermore, the second model construction module is used to construct a maxillary sinus downstream task model based on the weights of the maxillary sinus basic model. This module includes: a second data acquisition unit, a data annotation unit, a weight assignment unit, an annotated data partitioning unit, and a maxillary sinus downstream task model training unit. The second data acquisition unit selects a predetermined fourth number of oral CBCT image data from the pre-training dataset. The data annotation unit sets classification and measurement labels for the intended study areas of the predetermined fourth number of oral CBCT image data based on a predetermined downstream task annotation method, constructing an annotated dataset. The weight assignment unit assigns the encoder weights of the maxillary sinus basic model to the initial maxillary sinus downstream task model. The annotated data partitioning unit divides the annotated dataset into a fourth training set, a fourth validation set, and a fourth test set. The maxillary sinus downstream task model training unit trains the initial maxillary sinus downstream task model using the fourth training set, optimizes it using the fourth validation set, and performs performance testing on it using the fourth test set to obtain the maxillary sinus downstream task model.

[0020] In the above scheme, a labeled dataset is constructed by setting up downstream tasks such as classification, measurement and segmentation using a portion of oral CBCT image data. Combined with the encoder weights of the maxillary sinus basic model, it is ensured that the probability distribution of the source domain of the self-supervised training data is consistent with that of the target domain of the downstream task data. This solves the problem of unstable performance of the model when there is a data class imbalance in downstream tasks such as classification and segmentation, and improves the reliability of pre-implantation assessment.

[0021] Furthermore, the artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation proposed in this embodiment of the invention further includes: an initial rule engine construction module, a rule conversion module, a rule verification module, and a decision rule engine construction module; the initial rule engine construction module is used to construct an initial rule engine based on a preset historical decision rule dataset; the rule conversion module is used to convert the historical decision rule data into several rule conditions and several corresponding decision actions based on the initial rule engine; the rule verification module is used to perform logical consistency verification, coverage verification, and accuracy verification on the several rule conditions and several corresponding decision actions to obtain logical consistency verification results, coverage verification results, and accuracy verification results; the decision rule engine construction module is used to optimize the initial rule engine until the logical consistency verification results, coverage verification results, and accuracy verification results all meet the corresponding preset verification requirements to obtain the decision rule engine.

[0022] In the above scheme, a decision rule engine is constructed using a historical decision rule dataset. Clinical experience is transformed into a quantifiable mapping relationship between rule conditions and decision actions. After multi-dimensional verification and optimization of logical consistency, coverage, and accuracy, a decision rule library that meets clinical standards is formed. The decision rule engine systematically processes historical experience and eliminates rule conflicts and blind spots through an automated verification mechanism. This ensures that decision conditions are objectively matched with anatomical features and biomechanical indicators. When calling the corresponding decision design model based on qualitative and quantitative features, the reliability of historical experience can be inherited while avoiding subjective bias of human judgment. This achieves automated decision generation based on the fusion of standardized rules and imaging features, improving the reliability of preoperative assessment for implantation.

[0023] Furthermore, the evaluation module is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model, and, based on the output qualitative and quantitative features, calls the maxillary posterior tooth region implantation decision design model to generate a first decision or calls the maxillary posterior tooth region fenestration bone grafting decision design model to generate a second decision, so as to perform preoperative evaluation of maxillary posterior tooth region implantation through the first or second decision. This includes: a feature output unit, a semantic definition unit, a rule mapping unit, and a model calling and evaluation unit; the feature output unit is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model and output qualitative and quantitative features; the semantic definition unit... The definition unit is used to assign semantic definitions to qualitative and quantitative features based on a preset semantic model; the rule mapping unit is used to map the semantic definitions of qualitative and quantitative features to the corresponding rule conditions and decision actions based on the decision rule engine; the model invocation and evaluation unit is used to call the oral maxillary posterior tooth region implantation decision design model to generate a first decision or call the oral maxillary posterior tooth region fenestration bone grafting decision design model to generate a second decision based on the corresponding rule conditions and decision actions, so as to conduct preoperative evaluation of implantation in the maxillary posterior tooth region through the first decision or the second decision.

[0024] In the above scheme, the qualitative and quantitative features of the images to be tested are extracted using the downstream maxillary sinus task model. Semantic algorithms are used to give semantic definitions to the qualitative and quantitative features. Then, the decision rule engine maps the semantic definitions to the corresponding rule conditions and decision actions, automatically triggering the invocation of the maxillary posterior tooth region implantation decision design model. The interpretability of the qualitative and quantitative features of the images is analyzed by semantic algorithms. Combined with the matching mechanism of standardized decision rules in the decision rule engine, the whole chain from image feature analysis to decision model invocation is automated, eliminating the subjectivity of human experience intervention and ensuring the objective consistency between the decision logic and anatomical features and historical decision rule datasets. This generates a reliable implantation plan and improves the reliability of preoperative implantation assessment. Attached Figure Description

[0025] Figure 1 A schematic diagram of the module structure of a preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence, provided in a certain embodiment of the present invention;

[0026] Figure 2 A schematic diagram of a CBCT cross-sectional image of a preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence, provided for a certain embodiment of the present invention;

[0027] Figure 3 A schematic diagram illustrating the region of interest labeling principle of a target detection network in a preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence, provided for a certain embodiment of the present invention;

[0028] Figure 4 A schematic diagram of the maxillary sinus basic model image reconstruction process of an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation, provided for a certain embodiment of the present invention;

[0029] Figure 5 This is a schematic diagram of the preoperative assessment process of an artificial intelligence-based preoperative assessment system for implantation in the maxillary posterior tooth region, provided as an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1

[0032] See Figure 1 , Figure 1 This is a schematic flowchart illustrating the steps of a method for generating implant decisions in the maxillary posterior tooth region according to a certain embodiment of the present invention. Figure 1As shown in the figure, this invention proposes an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation, comprising: an image library construction module 201, a first model construction module 202, a second model construction module 203, and an assessment module 204; the image library construction module 201 is used to construct an oral CBCT image library based on several oral CBCT image data; the first model construction module 202 is used to construct a maxillary sinus basic model, an oral maxillary posterior tooth implantation decision design model, and an oral maxillary posterior tooth fenestration bone grafting decision design model based on the oral CBCT image library; the second model construction module 203 is used to construct a maxillary sinus downstream task model based on the weights of the maxillary sinus basic model; the assessment module 204 is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model, and, based on the qualitative and quantitative features output, call the oral maxillary posterior tooth implantation decision design model to generate a first decision or call the oral maxillary posterior tooth fenestration bone grafting decision design model to generate a second decision, so as to perform preoperative assessment of maxillary posterior tooth implantation through the first decision or the second decision.

[0033] In one specific implementation, the image library construction module 201 uses the NewTom VG oral CBCT imaging system to capture and collect oral CBCT image data with different shooting parameters in batches. The collected images typically contain motion artifacts and metal artifacts. Motion artifacts usually manifest as blurring, ghosting, or distortion in the image, while metal artifacts are usually caused by image distortion due to metal objects (such as dental fillings and metal frameworks) during scanning or imaging. Therefore, different image processing algorithms are needed to detect different artifacts in all oral CBCT image data. Image processing algorithms can be used to filter artifacts, and all processed oral CBCT image data are used to construct an oral CBCT image library. Then, the first model construction module 202 selects multiple sets of oral CBCT image data from the oral CBCT image library as training samples to train corresponding initial models to construct a maxillary sinus basic model, an oral maxillary posterior tooth region implantation decision design model, and an oral maxillary posterior tooth region fenestration bone grafting decision design model. The initial models include an initial target model and an initial image segmentation network model. The second model construction module 203 then... The encoder weights of the basic maxillary sinus model are assigned to the initial maxillary sinus downstream task model. The maxillary sinus downstream task model is further trained and optimized to construct the maxillary sinus downstream task model. Finally, the evaluation module 204 inputs the oral CBCT image data to be tested into the maxillary sinus downstream task model. The maxillary sinus downstream task model identifies the qualitative and quantitative features of the oral CBCT image data to be tested. Based on the qualitative and quantitative features of the oral CBCT image data to be tested, it calls the corresponding oral maxillary posterior tooth region implantation decision design model or oral maxillary posterior tooth region fenestration bone grafting decision design model. The oral maxillary posterior tooth region implantation decision design model generates a first decision based on the qualitative and quantitative features of the oral CBCT image data to be tested. The first decision can be used to guide the generation of oral maxillary posterior tooth region implantation surgery plan. The oral maxillary posterior tooth region fenestration bone grafting decision design model generates a second decision based on the qualitative and quantitative features of the oral CBCT image data to be tested. The second decision can be used to guide the generation of oral maxillary posterior tooth region fenestration bone grafting surgery plan. Thus, the preoperative evaluation of maxillary posterior tooth region implantation is performed through the first decision or the second decision. It is worth mentioning that the surgical plan generated by the preoperative assessment through the first and second decisions is not intended to execute or restrict the specific surgical procedure, but rather to play an auxiliary guiding role. For example, it may provide guidance on the selection of certain instruments for a certain situation. There are no restrictions on how the surgical operation is actually performed. The surgical procedure can be completed freely by using various methods and conditions according to the guidance.

[0034] This invention proposes an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation. An image library construction module builds an oral CBCT image library using several oral CBCT images, providing a data foundation for the first model construction module to train the maxillary sinus basic model, the maxillary posterior tooth implantation decision design model, and the maxillary posterior tooth fenestration bone grafting decision design model, ensuring the reliability of the model construction. Then, a second model construction module constructs a downstream maxillary sinus task model with weights from the maxillary sinus basic model, addressing the instability issue that arises when the downstream maxillary sinus task model exhibits data class imbalance in downstream tasks. The assessment module then combines the maxillary sinus downstream... The task model automatically extracts qualitative and quantitative features from the oral CBCT image data to be tested, and adaptively calls the decision design model for implantation in the maxillary posterior teeth region or the decision design model for fenestration and bone grafting in the maxillary posterior teeth region based on the qualitative and quantitative features. It then generates the first or second decision accordingly to conduct preoperative assessment of implantation in the maxillary posterior teeth region. This achieves full-process automation from data feature identification to decision generation and preoperative assessment. The multi-task model architecture realizes intelligent quantitative and qualitative image analysis and decision design model association calling throughout the entire process. By replacing manual experience judgment with an indicator-driven model calling mechanism, the objectivity and reliability of implantation decision generation are significantly improved, and the reliability of preoperative implant assessment is enhanced.

[0035] In a preferred embodiment, the image library construction module 201 is used to construct an oral CBCT image library based on several oral CBCT image data, including: a first data acquisition unit, a data filtering unit, a research site selection unit, and an image cropping unit; the first data acquisition unit is used to acquire several oral CBCT image data; the data filtering unit is used to filter out image data containing motion artifacts and metal artifacts based on a preset image processing algorithm to obtain initial oral CBCT image data; the research site selection unit is used to adjust the head position in each initial oral CBCT image data based on a preset reference plane and select the research site; the image cropping unit is used to crop the cross-section of the research site in each initial oral CBCT image data to obtain the oral CBCT image library.

[0036] In one preferred implementation, the first data acquisition unit uses the NewTom VG oral CBCT imaging system to capture and collect oral CBCT image data with different imaging parameters in batches. The collected images typically contain motion artifacts and metal artifacts. Motion artifacts usually manifest as blurring, ghosting, or distortion in the image, while metal artifacts are usually caused by image distortion due to metal objects (such as dental fillings and metal frameworks) during scanning or imaging. Therefore, different image processing algorithms are needed to detect different artifacts in all oral CBCT image data. For motion artifacts, the data filtering unit uses image processing algorithms (such as edge detection and registration algorithms) for detection. For metal artifacts, specific metal artifact correction algorithms or manual inspection are used for identification. After filtering out data containing motion artifacts and metal artifacts, the study site selection unit exports the initial oral CBCT image data as a DI (Digital Interface). The DICOM file is then imported into CBCT viewing software such as CoDiagnostiX. The head position is adjusted using the oral-maxillary plane as the horizontal reference plane (equivalent to the preset reference plane) and the midline as the vertical reference plane (equivalent to the preset reference plane). A cross-section at the cementoenamel junction is selected, and the geometric centers of all teeth on this cross-section are connected to draw a standard dental arch curve. The area to be studied (e.g., the maxillary sinus) is selected on the cross-section. The image cropping unit crops each coronal and / or sagittal section or horizontal section (e.g., each coronal section from the first premolar to the second molar from the mesial to the distal) of the area to be studied and includes it in the oral CBCT image library. It is worth mentioning that in this embodiment, the maxillary sinus is used as an example of the area to be studied, and it will not be described again below.

[0037] In the above scheme, image data containing motion artifacts and metal artifacts are screened out to eliminate irrelevant data interference and improve the accuracy of the data. In addition, by adjusting the head position in the data to a preset reference plane, selecting the research site, and cropping the corresponding interface according to the research site, a standardized atlas library is provided for subsequent network and model training, thereby improving the reliability of preoperative assessment for implantation.

[0038] In a preferred embodiment, the first model construction module 202 is used to construct a basic model of the maxillary sinus, a decision-making model for implantation in the maxillary posterior teeth region, and a decision-making model for fenestration and bone grafting in the maxillary posterior teeth region based on an oral CBCT image library. This module includes: a target detection network construction unit, a pre-training dataset acquisition unit, a basic maxillary sinus model construction unit, a decision-making model construction unit for implantation in the maxillary posterior teeth region, and a decision-making model construction unit for fenestration and bone grafting in the maxillary posterior teeth region. The target detection network construction unit is used to select a predetermined first number of oral CBCT image data from the oral CBCT image library and perform region-of-interest (ROI) labeling to obtain a target detection network. The pre-training dataset acquisition unit is used to perform RRO labeling on all oral CBCT image data in the oral CBCT image library based on the target detection network to obtain a pre-training dataset. The basic maxillary sinus model is constructed... The first unit is used to construct a basic model of the maxillary sinus based on a pre-trained dataset. The second unit for constructing a decision-making design model for implantation in the maxillary posterior teeth region is used to select a second preset number of oral CBCT images from the oral CBCT image library, generate a simulated implantation scheme for the maxillary posterior teeth region through a preset three-dimensional deep learning algorithm, and construct a decision-making design model for implantation in the maxillary posterior teeth region based on the simulated implantation scheme and the oral CBCT image data. The third unit for constructing a decision-making design model for fenestration bone grafting in the maxillary posterior teeth region is used to select a third preset number of oral CBCT images from the oral CBCT image library, generate a simulated fenestration bone grafting scheme for the maxillary posterior teeth region through a preset three-dimensional deep learning algorithm, and construct a decision-making design model for fenestration bone grafting in the maxillary posterior teeth region based on the simulated fenestration bone grafting scheme and the oral CBCT image data.

[0039] In one preferred implementation, the target detection network construction unit selects a portion of CBCT image data from the oral CBCT image library, such as CBCT cross-sectional images. The pre-training dataset acquisition unit uses specific image processing software such as Lableme to label the region of interest (maxillary sinus) in each selected CBCT cross-sectional image. The target detection network is then trained based on the labeled CBCT cross-sectional images. This target detection network can be used to independently identify and crop the region of interest in the CBCT cross-sectional images. Thus, the trained target detection network is used to identify the region of interest in all oral CBCT image data in the oral CBCT image library. In this embodiment, see [link to relevant documentation]. Figure 2 , Figure 2 A schematic diagram of a CBCT cross-sectional image of a preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence, provided in a certain embodiment of the present invention; [The image is then used to] obtain... Figure 2 All oral CBCT cross-sectional image data in the oral CBCT image library shown are available in the image database. Figure 3 , Figure 3This invention provides a schematic diagram illustrating the region of interest (ROI) labeling principle of a target detection network in a preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence, as shown in one embodiment of the invention. Figure 3 As shown, the system automatically identifies and divides regions of interest (ROIs) into all oral CBCT cross-sectional image data, automatically filters out irrelevant data that do not contain ROIs, and outputs all atlas data containing ROIs as a pre-training dataset. Figure 3 The red dashed box in the middle represents the region of interest; the maxillary sinus basic model building unit then uses a pre-trained dataset to build the maxillary sinus basic model through a preset self-supervised algorithm;

[0040] The oral posterior region implant decision-making design model construction unit selects a portion of CBCT image data from the oral CBCT image library, generates a simulated implantation plan for the oral posterior region using a preset 3D deep learning algorithm, and constructs an implant decision-making design model for the oral posterior region based on the simulated implantation plan and the oral CBCT image data. The oral posterior region fenestration bone graft decision-making design model construction unit then selects a portion of CBCT image data from the oral CBCT image library, generates a simulated fenestration bone graft plan for the oral posterior region using a preset 3D deep learning algorithm, and constructs an oral posterior region fenestration bone graft decision-making design model based on the simulated fenestration bone graft plan and the oral CBCT image data. In this implementation, the specific values ​​of the preset first quantity, preset second quantity, and preset third quantity are not limited. The preset first quantity, preset second quantity, and preset third quantity are used to explain that selecting a portion of CBCT image data from the oral CBCT image library is to select a small sample, and to distinguish the steps of selecting CBCT image data each time, which will not be elaborated further below.

[0041] In the above scheme, a portion of oral CBCT image data is first selected from the oral CBCT image database to mark regions of interest (ROIs), and a target detection network is constructed to reduce data computational redundancy. Then, the target detection network is used to identify ROIs from all oral CBCT image data in the database, filtering out cross-sectional data that do not contain ROIs to reduce the interference of useless data on subsequent model training and improve the reliability of subsequent maxillary sinus basic model construction and training. Finally, a portion of oral CBCT image data is selected from the oral CBCT image database and a 3D deep learning algorithm is used to generate the maxillary posterior tooth region... Implant placement simulation scheme and maxillary posterior tooth region fenestration bone grafting simulation scheme are combined with oral CBCT imaging data to construct maxillary posterior tooth region implant decision design model and maxillary posterior tooth region fenestration bone grafting decision design model. This establishes an objective mapping relationship between imaging features and clinical decisions, avoids subjective bias of human experience, and automatically outputs maxillary posterior tooth region implant decision design model and maxillary posterior tooth region fenestration bone grafting decision design model that conform to anatomical features and decision rules, thereby improving the reliability of preoperative implant assessment.

[0042] In a preferred embodiment, the target detection network construction unit is used to select a predetermined first number of oral CBCT image data from an oral CBCT image library for region of interest (ROI) labeling to obtain the target detection network. The unit includes: a data selection subunit, an ROI labeling subunit, a first dataset partitioning subunit, and a first model training subunit. The data selection subunit selects a predetermined first number of oral CBCT image data based on the oral CBCT image library. The ROI labeling subunit uses a predetermined image processing algorithm to label the predetermined first number of oral CBCT image data for ROI, obtaining several labeled oral CBCT image data. The first dataset partitioning subunit divides the several labeled oral CBCT image data into a first training set, a first validation set, and a first test set according to a predetermined ratio. The first model training subunit trains an initial target detection model using the training set, optimizes the model parameters of the initial target model using the validation set, and adjusts the hyperparameters of the optimized initial target model using the test set, thus obtaining the target detection network.

[0043] In one preferred implementation, a data selection subunit selects a portion of CBCT image data, such as CBCT cross-sectional images, from an oral CBCT image library. An interest labeling subunit uses specific image processing software such as Lableme to label the region of interest (maxillary sinus) in each selected image, automatically writing the labeling information to a designated file. The format of the labeled oral CBCT image data is then stored in a format recognizable by the target detection network. A first dataset partitioning subunit further divides the labeled oral CBCT image data into a first training set, a first validation set, and a first test set according to a preset ratio. A first model training subunit trains the initial target using the first training set. The model is optimized and adjusted using the first validation set and the first test set to obtain a target detection network that can independently identify and crop regions of interest in CBCT cross-sectional images. For the trained target detection network, appropriate evaluation metrics such as precision, recall, and area under the PR curve (AUPRC) can be used to evaluate the training effect of the network. When precision, recall, and AUPRC are all close to 1.0, it proves that the target detection network can accurately identify the maxillary sinus region in CBCT cross-sectional images. In this embodiment, YOLO V5 can be used as the target detection network for training, and five-fold cross-validation and data augmentation methods can be used to improve the model performance and generalization ability.

[0044] In the above scheme, a portion of the oral CBCT image data is selected to mark the region of interest for training the target detection network. The target detection network is trained with a small amount of data so that it can learn to crop and divide the region of interest in the oral CBCT image data, thereby achieving automated region of interest division and improving the reliability of pre-implantation assessment.

[0045] In a preferred embodiment, the maxillary sinus basic model construction unit is used to construct a maxillary sinus basic model based on a pre-training dataset, including: a mesh construction subunit, an encoding subunit, a decoding subunit, a loss function calculation subunit, and a model update subunit; the mesh construction subunit is used to cut the pre-training dataset into uniform and non-overlapping meshes based on a preset self-supervised algorithm; the encoding subunit is used to select a preset proportion of meshes to set masks, and convert the unmasked meshes into vector sequences through an encoder; the decoding subunit is used to reconstruct the original image from the vector sequences and the masked meshes through a decoder, obtaining a reconstructed image; the loss function calculation subunit is used to calculate the corresponding loss function based on the reconstructed image and the original image; the model update subunit is used to update the preset proportion values, the encoder weight settings, and the mask setting mode, recalculate the corresponding loss function, and select a maxillary sinus basic model whose loss function meets preset requirements.

[0046] One preferred implementation method is described in [reference]. Figure 4 , Figure 4 This invention provides a schematic diagram of the maxillary sinus basic model image reconstruction process for an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation, as shown in one embodiment of the present invention. Figure 4 As shown, the input image data is segmented into uniform and non-overlapping grids using a self-supervised algorithm, Masked Autoencoders network. Then, a mask is obtained by random sampling at a certain ratio, partially covering the grid. The encoder first converts the uncovered grids into a vector sequence, and then the decoder reconstructs the original image from the vector sequence and the masked grids. Specifically, the mean squared error (MSE) is typically used to calculate the pixel-space error between the reconstructed and original images. The loss function is calculated only for the prediction results of the mask patch. Furthermore, common anatomical features of oral CBCT image data are learned during the image masking and reconstruction process. Feature representation is learned by minimizing the difference between the original and reconstructed images. The specific formula for the MSE loss function is as follows:

[0047]

[0048] In the formula, f(x) is the predicted value, i.e. the reconstructed image, y is the true value, i.e. the original image, and n is the number of samples;

[0049] By comparing the model performance under different encoder weights, different mesh masking modes (such as random mesh masking, regular mesh masking, and patchy mesh masking) and different mesh masking ratios (such as 65%, 75%, or 85%), a maxillary sinus baseline model whose loss function meets the preset requirements is selected. The reconstructed image is then used to express features of the reconstructed image through the selected maxillary sinus baseline model to obtain the reconstructed image of the CBCT image data.

[0050] In the above scheme, a self-supervised algorithm is used to divide the pre-training dataset into a uniform non-overlapping grid. Random masks are generated and the unmasked grids are converted into vector sequences to reconstruct the original image. The differences between the original image and the reconstructed image are then compared to calculate the loss function. By setting different conditions, the loss function under different conditions is obtained. The model corresponding to the optimal loss function is selected as the basic model of the maxillary sinus, ensuring that the model learns the features of CBCT images more fully, so as to build a more reliable basic model of the maxillary sinus and improve the reliability of preoperative assessment for implantation.

[0051] In a preferred embodiment, the oral posterior region implant decision design model construction unit is used to select a preset second number of oral CBCT image data from an oral CBCT image library, generate an oral posterior region implantation simulation plan through a preset three-dimensional deep learning algorithm, and construct an oral posterior region implantation decision design model based on the oral posterior region implantation simulation plan and the oral CBCT image data. This model includes: a first data preprocessing subunit, a first simulation plan acquisition subunit, a first decision dataset acquisition subunit, a second dataset partitioning subunit, and a second model training subunit. The first data preprocessing subunit is used to preprocess the preset second number of oral CBCT image data; the first simulation plan acquisition subunit is used to preprocess the oral CBCT image data through a preset three-dimensional deep learning algorithm. The image data of the missing maxillary posterior teeth area is sequentially reconstructed in 3D and implants are simulated for implant placement to obtain a simulated implant placement plan for the maxillary posterior teeth region. The first decision dataset acquisition subunit is used to match the simulated implant placement plan for the maxillary posterior teeth region with the corresponding oral CBCT image data to obtain the implant decision dataset for the maxillary posterior teeth region. The second dataset partitioning subunit is used to divide the implant decision dataset for the maxillary posterior teeth region into a second training set, a second validation set, and a second test set. The second model training subunit is used to train the preset initial image segmentation network model through the second training set, optimize the parameters of the preset initial image segmentation network model through the second validation set, and optimize the hyperparameters of the preset initial image segmentation network model through the second test set to obtain the implant decision design model for the maxillary posterior teeth region.

[0052] One preferred implementation involves selecting a subset of oral CBCT image data from an oral CBCT image library. This data may include CBCT images showing missing maxillary posterior teeth. A first data preprocessing subunit performs format conversion to unify the format of the CBCT image data with missing maxillary posterior teeth, image normalization to unify the grayscale value range of all image data to a standard range, and noise removal to remove noise from the image data, making the image data clearer and highlighting important anatomical features. A first simulation scheme acquisition subunit then processes the preprocessed oral... The CBCT image data was used to perform 3D reconstruction and implant placement simulation of the maxillary posterior tooth loss area using professional software such as co-DiagnostiX, which can run 3D deep learning algorithms. The software provides a library of various implant brands and models. Based on the anatomical structure of the maxillary posterior tooth region, the appropriate implant type, length, and diameter were selected. The implant was then virtually placed in the ideal position and angle. The implant placement simulation results were output in 3D file format to obtain the maxillary posterior tooth implant placement simulation plan file. The first decision dataset was used to obtain sub-units for further processing of the maxillary posterior tooth implant placement simulation plan. The preprocessing of the implant simulation plan file primarily involves 3D model registration. For example, the STL format 3D model in the implant placement simulation plan file is registered with the corresponding CBCT image data. This can be achieved using open-source 3D medical image computing software such as 3DSlicer, whose registration function accurately registers the implant simulation plan to the coordinate system of the CBCT image data, ensuring an accurate spatial correspondence between the two. This allows for comprehensive consideration of both image information and simulated implant position information. The second step involves dividing the preprocessed maxillary posterior tooth region implant decision dataset into sub-units according to a preset ratio. The system consists of a second training set, a second validation set, and a second test set. The second model training subunit trains the initial image segmentation network model using the second training set, optimizes the model parameters of the initial image segmentation network model using the second validation set, and adjusts the hyperparameters of the optimized initial image segmentation network model using the second test set to obtain the implant decision design model for the maxillary posterior teeth region. Thus, the CBCT image data to be tested is input into the implant decision design model for the maxillary posterior teeth region, enabling intelligent design and output of implant decisions for the maxillary posterior teeth region in subsequent operations. In this embodiment, the preset initial image segmentation network model can be nn-Unet or 3D Unet, etc.

[0053] In the above scheme, a portion of the selected oral CBCT image data is preprocessed. A three-dimensional deep learning algorithm is used to perform three-dimensional reconstruction and implant simulation of the maxillary posterior tooth missing area in the oral CBCT image data. The generated implant simulation plan for the maxillary posterior tooth region is matched with the corresponding oral CBCT image data to construct an implant decision dataset for the maxillary posterior tooth region. Then, a decision design model for implantation in the maxillary posterior tooth region is constructed by combining the dataset partitioning and the iterative optimization training mechanism of the three-dimensional image segmentation network. A three-dimensional deep learning algorithm is used to generate a simulated implantation plan for the maxillary posterior tooth region. Finally, the simulated implantation plan for the maxillary posterior tooth region and the corresponding oral CBCT image data are jointly used to construct an implant decision design model for the maxillary posterior tooth region. This establishes an objective mapping relationship between image features and clinical decisions, avoids subjective bias of human experience, and automatically outputs an implant decision design model for the maxillary posterior tooth region that conforms to anatomical features and decision rules, thereby improving the reliability of preoperative implant assessment.

[0054] In a preferred embodiment, the oral posterior dentition fenestration bone grafting decision design model construction unit is used to select a predetermined third number of oral CBCT image data from the oral CBCT image library, generate a simulated scheme for oral posterior dentition fenestration bone grafting using a predetermined three-dimensional deep learning algorithm, and construct an oral posterior dentition fenestration bone grafting decision design model based on the simulated scheme and the oral CBCT image data. This model includes: a second data preprocessing subunit, a second simulation scheme acquisition subunit, a second decision dataset acquisition subunit, a third dataset partitioning subunit, and a third model training subunit. The second data preprocessing subunit is used to preprocess the predetermined third number of oral CBCT image data; the second simulation scheme acquisition subunit is used to preprocess the oral CBCT image data using a predetermined three-dimensional deep learning algorithm. The three-dimensional reconstruction of the maxillary posterior tooth loss area and the simulation of the maxillary posterior tooth fenestration bone grafting area were performed sequentially to obtain a simulated plan for maxillary posterior tooth fenestration bone grafting. The second decision dataset acquisition subunit was used to match the simulated plan for maxillary posterior tooth fenestration bone grafting with the corresponding oral CBCT image data to obtain a decision dataset for maxillary posterior tooth fenestration bone grafting. The third dataset partitioning subunit was used to divide the decision dataset for maxillary posterior tooth fenestration bone grafting into a third training set, a third validation set, and a third test set. The third model training subunit was used to train the preset initial image segmentation network model using the third training set, optimize the parameters of the preset initial image segmentation network model using the third validation set, and optimize the hyperparameters of the preset initial image segmentation network model using the third test set to obtain a decision design model for maxillary posterior tooth fenestration bone grafting.

[0055] One preferred implementation involves selecting a portion of CBCT image data from an oral CBCT image library. A second data preprocessing subunit and a second simulation scheme acquisition subunit sequentially perform three-dimensional reconstruction of the maxillary posterior tooth loss area in the oral CBCT image data. A second decision dataset acquisition subunit then performs bone wall segmentation and annotation for the maxillary sinus lift fenestration scheme on the selected three-dimensional reconstructed image data based on a preset image processing algorithm, obtaining an oral maxillary posterior tooth region fenestration bone graft decision dataset. A third dataset partitioning subunit divides the oral maxillary posterior tooth region fenestration bone graft decision dataset into a third training set, a third validation set, and a third test set according to a preset ratio. A third model training subunit then trains the model through the third training set... The training set is used to train the preset initial image segmentation network model. Then, the parameters of the preset initial image segmentation network model are optimized through the third validation set, and the hyperparameters of the preset initial image segmentation network model are optimized through the third test set. This yields the decision-making design model for fenestration bone grafting in the maxillary posterior tooth region. The decision-making design model for fenestration bone grafting in the maxillary posterior tooth region can realize the positioning of the maxillary sinus fenestration bone grafting scheme and bone wall segmentation in subsequent operations, and intelligently design and output the decision for fenestration bone grafting in the maxillary posterior tooth region. The generation process of the decision-making design model for fenestration bone grafting in the maxillary posterior tooth region is basically the same as that of the decision-making design model for implantation in the maxillary posterior tooth region. The only difference is the data used. The data processing process is basically the same, and will not be described in detail here.

[0056] In the above scheme, a portion of the selected oral CBCT image data is preprocessed. A 3D deep learning algorithm is used to perform 3D reconstruction of the maxillary posterior tooth missing area in the oral CBCT image data and to simulate the maxillary posterior tooth region fenestration and bone grafting area. The generated maxillary posterior tooth region fenestration and bone grafting simulation scheme is matched with the corresponding oral CBCT image data to construct a maxillary posterior tooth region fenestration and bone grafting decision dataset. Then, a maxillary posterior tooth region fenestration and bone grafting decision design model is constructed by combining dataset partitioning and a 3D image segmentation network iterative optimization training mechanism. A maxillary posterior tooth region fenestration and bone grafting simulation scheme is generated using a 3D deep learning algorithm. Finally, the maxillary posterior tooth region fenestration and bone grafting simulation scheme and the corresponding oral CBCT image data are jointly used to construct a maxillary posterior tooth region fenestration and bone grafting decision design model. This establishes an objective mapping relationship between image features and clinical decisions, avoids subjective biases from human experience, and automatically outputs a maxillary posterior tooth region fenestration and bone grafting decision design model that conforms to anatomical features and decision rules, thereby improving the reliability of pre-implantation assessment.

[0057] In a preferred embodiment, the second model construction module is used to construct a maxillary sinus downstream task model based on the weights of the maxillary sinus basic model. This module includes: a second data acquisition unit, a data annotation unit, a weight assignment unit, an annotated data partitioning unit, and a maxillary sinus downstream task model training unit. The second data acquisition unit selects a predetermined fourth number of oral CBCT image data from the pre-training dataset. The data annotation unit sets classification and measurement labels for the intended study areas of the predetermined fourth number of oral CBCT image data based on a predetermined downstream task annotation method, constructing an annotated dataset. The weight assignment unit assigns the encoder weights of the maxillary sinus basic model to the initial maxillary sinus downstream task model. The annotated data partitioning unit divides the annotated dataset into a fourth training set, a fourth validation set, and a fourth test set. The maxillary sinus downstream task model training unit trains the initial maxillary sinus downstream task model using the fourth training set, optimizes it using the fourth validation set, and performs performance testing on it using the fourth test set to obtain the final maxillary sinus downstream task model.

[0058] In one preferred implementation, the second data acquisition unit selects a portion of oral CBCT image data from the pre-training dataset. The data annotation unit sets classification labels (e.g., labels for risk factors of maxillary sinus lift surgery) using a preset downstream task annotation method. This preset downstream task annotation method is a method used to simulate professionals annotating the areas to be studied. It can annotate the areas to be studied in the oral CBCT image data according to predetermined rules to obtain the general features of the oral CBCT image data, thereby constructing an annotated dataset and providing a reliable data foundation for subsequent model training. The weight assignment unit retains the encoder part of the maxillary sinus basic model and assigns the encoder weights of the maxillary sinus basic model to the initial maxillary sinus downstream task model. The initial maxillary sinus downstream task model can... Using models such as Vit-Small, Vit-Base, and Vit-Large, the labeled data partitioning unit divides the acquired labeled dataset into a fourth training set, a fourth validation set, and a fourth test set according to a certain ratio. The maxillary sinus downstream task model training unit uses the fourth training set and the fourth validation set to train and optimize the initial maxillary sinus downstream task model respectively, and uses the fourth test set to test the performance of the initial maxillary sinus downstream task model, so as to obtain a maxillary sinus downstream task model that can perform tasks such as automatic recognition, qualitative classification, quantitative analysis, and image segmentation on oral CBCT image data. In this embodiment, the preset fourth quantity is not limited to a specific value, but is only used to distinguish the preset first quantity, preset second quantity, and preset third quantity.

[0059] To further ensure the accuracy of the maxillary sinus downstream task model, this embodiment of the invention also proposes a feedback verification step. Each output result of the maxillary sinus downstream task model is fed back to the model for verification, and the error range of the output result is determined to improve the reliability of the maxillary sinus downstream task model. Specifically, in one possible implementation, each output result of the maxillary sinus downstream task model is fed back to the target detection network and a preset standardized annotation model for verification, and the error range of the recognition result is determined. If the error range does not meet the requirements, it indicates that the maxillary sinus downstream task model has deviated from its task execution, and the model parameters need to be re-optimized and trained. If the error range meets the requirements, the output result of the current maxillary sinus downstream task model is considered accurate, and the output result is stored in the oral CBCT image library to provide a more reliable data foundation for subsequent model optimization and updates.

[0060] In the above scheme, a labeled dataset is constructed by setting up downstream tasks such as classification, measurement and segmentation using a portion of oral CBCT image data. Combined with the encoder weights of the maxillary sinus basic model, it is ensured that the probability distribution of the source domain of the self-supervised training data is consistent with that of the target domain of the downstream task data. This solves the problem of unstable performance of the model when there is a data class imbalance in downstream tasks such as classification and segmentation, and improves the reliability of pre-implantation assessment.

[0061] In a preferred embodiment of the present invention, a method for generating implantation decisions in the maxillary posterior tooth region further includes: an initial rule engine construction module, a rule conversion module, a rule verification module, and a decision rule engine construction module. The initial rule engine construction module is used to construct an initial rule engine based on a preset historical decision rule dataset. The rule conversion module is used to convert the historical decision rule data into several rule conditions and several corresponding decision actions based on the initial rule engine. The rule verification module is used to perform logical consistency verification, coverage verification, and accuracy verification on the several rule conditions and several corresponding decision actions, obtaining logical consistency verification results, coverage verification results, and accuracy verification results. The decision rule engine construction module is used to optimize the initial rule engine until the logical consistency verification results, coverage verification results, and accuracy verification results all meet the corresponding preset verification requirements, thus obtaining the decision rule engine.

[0062] One preferred implementation method involves an initial rule engine construction module that builds a clinical pathway rule engine for maxillary posterior tooth implantation (equivalent to an initial rule engine). Specifically, this engine can be constructed based on in-depth analysis of a large amount of authoritative literature, clinical guidelines, and historical expert experience in the field of oral implantology. This results in a decision tree driven by medical terminology and patient clinical data, defining the indications, contraindications, and surgical method selection rules for maxillary posterior tooth implantation. The rule transformation module, through the initial rule engine, transforms these decision rules into a series of rule conditions and corresponding actions. First, it meticulously analyzes and decomposes the clinical decision rules, clarifying the key factors and logical relationships involved in each decision point. For example, for maxillary posterior tooth implantation... The determination of indications for dental implantation involves numerous qualitative and quantitative features, including: determining alveolar bone height by setting thresholds within different height ranges; assessing the degree of thickening of the maxillary sinus mucosa through imaging measurements; and diagnosing the presence and severity of maxillary sinus polyps based on imaging findings. These features are rigorously defined using logical relationships such as AND, OR, and NOT. The rules are then precisely encoded using rule languages ​​such as Drools and CLIPS supported by the initial rule engine, transforming each condition and its corresponding logical relationship into a format recognizable and processable by the initial rule engine. Simultaneously, the corresponding decision actions are linked to the relevant rules. The conditions are then linked and bound together. Decision actions include whether implantation is suitable, whether bone augmentation is needed, what type of bone augmentation method should be used, and what implant length and diameter should be selected. This constructs a complete rule set, realizing the transformation of decision rules into executable rules for the initial rule engine. The rule verification module verifies and optimizes the initial rule engine to ensure the accuracy and completeness of the rules (equivalent to preset verification requirements), resulting in the decision rule engine. In the verification of the initial rule engine, logical consistency verification can be used, that is, reviewing each rule in the initial rule engine to check whether the logical relationships between rule conditions conform to clinical reality, ensuring that there are no logical contradictions or conflicts. For example, for cases with low remaining alveolar bone height and... In cases of significant thickening of the maxillary sinus mucosa, the logic in the rules regarding suitability for direct implantation is examined for its rationality and consistency with clinical practice (equivalent to a pre-set verification requirement). Coverage verification is employed by collecting a large number of historical oral implant clinical cases, covering information on patients with implantation in the maxillary posterior region under various conditions. This case data is input into the initial rule engine for simulation testing to check whether the initial rule engine can accurately match and process these cases, and whether it can provide targeted rule coverage for all possible clinical situations. The proportion of covered cases to the total number of cases is statistically analyzed. If it falls below a certain threshold, generally set at 90%, the rules need to be supplemented and improved (equivalent to a pre-set verification requirement).Accuracy verification can also be used, which involves comparing the decision results of the initial rule engine with the diagnostic decision results of similar or identical historical cases, calculating the consistency ratio. If the consistency is lower than the preset standard (generally set at 85%), the initial rule engine needs to be adjusted and optimized to identify the reasons for the discrepancy, which may be due to unreasonable rule conditions or inappropriate action selection (equivalent to the preset verification requirements). Optimization of the initial rule engine can be achieved through rule adjustment, parameter optimization, and rule expansion. Specifically, rule adjustment involves fine-tuning the thresholds in the rule conditions based on problems discovered during verification. For example, consulting multiple sources of dental implant professional guidelines, the threshold range for residual alveolar bone height can be appropriately widened or narrowed within a reasonable range to improve the adaptability and accuracy of the rules. It may also involve re-organizing the logical relationships of the rules to better align with clinical decision-making logic. Parameter optimization involves adjusting the initial rule engine... During operation, parameters such as the tolerance of fuzzy matching and rule priority are optimized and adjusted to improve the performance and decision-making effectiveness of the initial rule engine. For example, based on actual test results, rule priorities are reasonably set to ensure that when multiple applicable rules exist, the rule that best conforms to the clinical priority principle is executed first. Rule expansion involves timely incorporating new clinical research results, treatment techniques, and expert experience into the initial rule engine as oral implant technology continues to develop and clinical experience accumulates. This ensures that the initial rule engine keeps pace with the times and adapts to changing clinical needs. For example, when there is severe bone atrophy in the maxillary posterior region, the application rules of new bone augmentation techniques and implant surface treatment techniques are promptly incorporated into the initial rule engine. The decision rule engine construction module optimizes the initial rule engine to ensure that it meets the corresponding verification requirements in logical consistency verification, coverage verification, and accuracy verification, thus obtaining the decision rule engine.

[0063] In the above scheme, a decision rule engine is constructed using a historical decision rule dataset. Clinical experience is transformed into a quantifiable mapping relationship between rule conditions and decision actions. After multi-dimensional verification and optimization of logical consistency, coverage, and accuracy, a decision rule library that meets clinical standards is formed. The decision rule engine systematically processes historical experience and eliminates rule conflicts and blind spots through an automated verification mechanism. This ensures that decision conditions are objectively matched with anatomical features and biomechanical indicators. When calling the corresponding decision design model based on qualitative and quantitative features, the reliability of historical experience can be inherited while avoiding subjective bias of human judgment. This achieves automated decision generation based on the fusion of standardized rules and imaging features, improving the reliability of preoperative assessment for implantation.

[0064] In a preferred embodiment, the evaluation module is used to input the oral CBCT image data to be tested into the downstream maxillary sinus task model, and, based on the output qualitative and quantitative features, call the oral posterior maxillary implantation decision design model to generate a first decision or call the oral posterior maxillary fenestration bone grafting decision design model to generate a second decision, so as to perform preoperative evaluation of implantation in the maxillary posterior maxillary region through the first or second decision. This module includes: a feature output unit, a semantic definition unit, a rule mapping unit, and a model calling and evaluation unit; the feature output unit is used to input the oral CBCT image data to be tested into the downstream maxillary sinus task model and output qualitative and quantitative features; The semantic definition unit is used to assign semantic definitions to qualitative and quantitative features based on a preset semantic model; the rule mapping unit is used to map the semantic definitions of qualitative and quantitative features to the corresponding rule conditions and decision actions based on the decision rule engine; the model invocation and evaluation unit is used to generate a first decision by invoking the oral maxillary posterior tooth region implantation decision design model or generate a second decision by invoking the oral maxillary posterior tooth region fenestration bone grafting decision design model based on the corresponding rule conditions and decision actions, so as to conduct preoperative evaluation of implantation in the maxillary posterior tooth region through the first decision or the second decision.

[0065] One preferred implementation method is described in [reference]. Figure 5 , Figure 5 This invention provides a schematic diagram of the preoperative assessment process for an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation, as shown in one embodiment of the invention. Figure 5As shown, the feature output unit inputs the oral CBCT image data to be tested into the downstream task model of the maxillary sinus and outputs qualitative and quantitative features. The semantic definition unit defines the qualitative and quantitative features of the maxillary posterior tooth region before implant surgery using a preset semantic model, clarifying the clinical significance, normal range, and abnormality level of each feature. The rule mapping unit associates and maps the features with the rule conditions in the decision rule engine according to the semantic definition of each feature, determining the indicators and judgment conditions that need to be referenced for each rule condition. Specifically, it first clarifies the data structure design, such as the data structure of qualitative and quantitative features and the rule rules. The data structure of a decision rule engine includes data types, value ranges, and units. More specifically, the height of remaining alveolar bone might be a numerical feature in millimeters, while the degree of maxillary sinus mucosal thickening might be a categorical feature, qualitatively described as mild, moderate, or severe, or represented by corresponding numerical ranges. Rule conditions typically consist of multiple features and their logical relationships. Therefore, the decision rule engine needs to define a structure to represent these conditions, such as a condition tree or condition list. Each condition node contains references to indicators, comparison operators, and specific thresholds. Secondly, the engine maps features to rule conditions. This mapping process distinguishes between single-feature mapping and multi-feature mapping. For single-feature mappings, for simple rule conditions, the decision rule engine can directly reference the indicator name or identifier and set corresponding comparison conditions to complete the mapping. For example, if the rule condition is "remaining alveolar bone height less than 5 mm," the decision rule engine will associate the feature "remaining alveolar bone height" with the comparison condition "less than 5 mm." For multi-feature mappings, when rule conditions involve multiple features, the decision rule engine uses logical operators... Multiple indicators and their comparison conditions are combined. For example, if the rule condition is "remaining alveolar bone height is less than 5 mm and the degree of maxillary sinus mucosal thickening is moderate or severe", the decision rule engine will associate the two indicators "remaining alveolar bone height" and "degree of maxillary sinus mucosal thickening" with their respective comparison conditions, and combine them into a complete rule condition through "AND" and "OR" operators. This process is repeated to complete the mapping of all features to rule conditions. Finally, the model invocation and evaluation unit inputs the qualitative and quantitative features of the maxillary posterior tooth region preoperative implantation output by the maxillary sinus downstream task model into the decision rule engine.The decision rule engine reasones and judges based on preset rule conditions and logical relationships, and calls the corresponding intelligent design model. For example, it first judges whether the qualitative and quantitative features corresponding to the rule conditions and decision actions generated by the rule decision engine are suitable for implantation. If suitable, it calls the oral maxillary posterior tooth region implantation decision design model to generate the first decision. If unsuitable, it re-inputs the qualitative and quantitative features into the decision rule engine, remaps the semantic definition of the qualitative and quantitative features to the qualitative and quantitative feature corresponding rule conditions and decision actions, and obtains the updated remapped qualitative and quantitative feature corresponding rule conditions and decision actions. Then, it judges whether the remapped qualitative and quantitative feature corresponding rule conditions and decision actions are suitable for maxillary sinus lift surgery. If suitable, it calls the oral maxillary posterior tooth region fenestration bone grafting decision design model to generate the second decision. If unsuitable, it means that the current oral CBCT image data to be tested is not applicable to the oral maxillary posterior tooth region implantation decision design model and oral maxillary posterior tooth region fenestration bone grafting decision design model proposed in this invention, and the decision rule mapping needs to be paused to further explore other strategies. In this embodiment of the invention, the first or second decision includes one or more combinations of clinical decisions such as whether implantation is suitable, whether bone augmentation is needed, what type of bone augmentation method should be used, and what implant length and diameter should be selected. This embodiment of the invention also includes a user interface to facilitate doctors in viewing and understanding the decision suggestions, and to allow for adjustments and confirmation based on actual circumstances.

[0066] In the above scheme, the qualitative and quantitative features of the images to be tested are extracted using the downstream maxillary sinus task model. Semantic algorithms are used to give semantic definitions to the qualitative and quantitative features. Then, the decision rule engine maps the semantic definitions to the corresponding rule conditions and decision actions, automatically triggering the invocation of the maxillary posterior tooth region implantation decision design model. The interpretability of the qualitative and quantitative features of the images is analyzed by semantic algorithms. Combined with the matching mechanism of standardized decision rules in the decision rule engine, the whole chain from image feature analysis to decision model invocation is automated, eliminating the subjectivity of human experience intervention and ensuring the objective consistency between the decision logic and anatomical features and historical decision rule datasets. This generates a reliable implantation plan and improves the reliability of preoperative implantation assessment.

[0067] Therefore, this invention proposes an artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation, which automates the entire preoperative assessment process for maxillary posterior tooth implantation and designs innovative solutions for the characteristics and difficulties of each step: 1) Training a basic maxillary sinus model based on self-supervised learning, learning the general anatomical features of the maxillary posterior tooth region from a large amount of unlabeled CBCT data, and achieving intelligent prediction of complex qualitative and precise quantitative indicators before surgery without relying on labeled data; 2) Constructing a decision rule engine, transforming the expert knowledge base and a large amount of clinical experience into quantifiable rule conditions and decision action mapping relationships, and automatically mapping them to corresponding implantation clinical decisions based on qualitative and quantitative features; 3) Constructing an implantation decision design model and a fenestration bone grafting decision design model for the maxillary posterior tooth region based on a lightweight three-dimensional segmentation network, and completing the preoperative assessment automatically, intelligently, and visually based on the implantation treatment decision.

[0068] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0069] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the described specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0070] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

Claims

1. A preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence, characterized in that, include: The system includes an image library construction module, a first model construction module, a second model construction module, and an evaluation module. The image library construction module is used to construct an oral CBCT image library based on several oral CBCT image data, including: a first data acquisition unit, a data filtering unit, a research site selection unit, and an image cropping unit; the first data acquisition unit is used to acquire several oral CBCT image data; the data filtering unit is used to filter out image data containing motion artifacts and metal artifacts based on a preset image processing algorithm to obtain initial oral CBCT image data; the research site selection unit is used to adjust the head position in each initial oral CBCT image data based on a preset reference plane and select the research site; the image cropping unit is used to crop the cross-section of the research site in each initial oral CBCT image data to obtain the oral CBCT image library; The first model construction module is used to construct a basic maxillary sinus model, a decision-making model for implantation in the maxillary posterior teeth region, and a decision-making model for fenestration and bone grafting in the maxillary posterior teeth region based on the oral CBCT image library. It includes: a target detection network construction unit, a pre-training dataset acquisition unit, a basic maxillary sinus model construction unit, a decision-making model construction unit for implantation in the maxillary posterior teeth region, and a decision-making model construction unit for fenestration and bone grafting in the maxillary posterior teeth region. The target detection network construction unit is used to select a preset first number of oral CBCT image data from the oral CBCT image library and mark the regions of interest to obtain a target detection network. The pre-training dataset acquisition unit... The unit is used to mark the regions of interest (ROIs) of all oral CBCT image data in the oral CBCT image library based on the target detection network, thereby obtaining a pre-training dataset; the maxillary sinus basic model construction unit is used to construct a maxillary sinus basic model based on the pre-training dataset; the oral maxillary posterior tooth region implant decision design model construction unit is used to select a preset second number of oral CBCT image data from the oral CBCT image library, generate a simulated implantation plan for the maxillary posterior tooth region through a preset three-dimensional deep learning algorithm, and construct an oral maxillary posterior tooth region implantation model based on the simulated implantation plan and the oral CBCT image data. The decision-making design model for maxillary posterior tooth region fenestration bone grafting is constructed by selecting a preset third number of oral CBCT image data from the oral CBCT image library, generating a simulated scheme for maxillary posterior tooth region fenestration bone grafting through the preset three-dimensional deep learning algorithm, and constructing a decision-making design model for maxillary posterior tooth region fenestration bone grafting based on the simulated scheme and the oral CBCT image data. The maxillary sinus basic model construction unit includes: a mesh construction subunit, an encoding subunit, a decoding subunit, a loss function calculation subunit, and a model update subunit. The mesh construction subunit is used to perform self-monitoring based on a preset algorithm. The supervised algorithm divides the pre-training dataset into uniform and non-overlapping grids. The encoding subunit selects a grid with a preset ratio to set a mask, and the encoder converts the unmasked grids into a vector sequence. The decoding subunit reconstructs the original image using the vector sequence and the masked grids, obtaining a reconstructed image. The loss function calculation subunit calculates the corresponding loss function based on the reconstructed image and the original image. The model update subunit updates the preset ratio values, the encoder weight settings, and the mask setting mode, recalculates the corresponding loss function, and selects a maxillary sinus baseline model whose loss function meets preset requirements. The second model building module is used to construct a downstream task model of the maxillary sinus based on the weights of the maxillary sinus basic model; The evaluation module is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model, and call the oral maxillary posterior tooth region implantation decision design model to generate a first decision or call the oral maxillary posterior tooth region fenestration bone grafting decision design model to generate a second decision based on the output qualitative and quantitative features, so as to perform preoperative evaluation of maxillary posterior tooth region implantation through the first decision or the second decision.

2. The artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation as described in claim 1, characterized in that, The target detection network construction unit is used to select a preset first number of oral CBCT image data from the oral CBCT image library, mark the region of interest, and obtain the target detection network, including: Data selection sub-unit, interest tagging sub-unit, first dataset partitioning sub-unit, and first model training sub-unit; The data selection subunit is used to select a preset first number of oral CBCT image data based on the oral CBCT image library; The region of interest (ROI) tagging subunit is used to tag the preset first number of oral CBCT image data according to a preset image processing algorithm, so as to obtain a number of tagged oral CBCT image data. The first dataset partitioning subunit is used to divide the several labeled oral CBCT image data into a first training set, a first validation set, and a first test set according to a preset ratio; The first model training subunit is used to train the initial target detection model using the training set, optimize the model parameters of the initial target model using the validation set, and adjust the hyperparameters of the optimized initial target model using the test set to obtain the target detection network.

3. The artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation as described in claim 1, characterized in that, The oral posterior region implant decision-making design model construction unit is used to select a preset second number of oral CBCT image data from the oral CBCT image library, generate an oral posterior region implantation simulation plan through a preset three-dimensional deep learning algorithm, and construct an oral posterior region implantation decision-making design model based on the oral posterior region implantation simulation plan and the oral CBCT image data, including: The system comprises a first data preprocessing subunit, a first simulation scheme acquisition subunit, a first decision dataset acquisition subunit, a second dataset partitioning subunit, and a second model training subunit. The first data preprocessing subunit is used to preprocess the preset second number of oral CBCT image data; The first simulation scheme acquisition subunit is used to perform three-dimensional reconstruction and implant simulation on the maxillary posterior tooth missing area in the oral CBCT image data through a preset three-dimensional deep learning algorithm, so as to obtain a simulation scheme for implantation of maxillary posterior teeth in the oral cavity. The first decision dataset acquisition subunit is used to match the oral CBCT image data with the simulated implantation scheme in the maxillary posterior tooth region to obtain the implantation decision dataset in the maxillary posterior tooth region. The second dataset partitioning subunit is used to divide the oral maxillary posterior tooth region implantation decision dataset into a second training set, a second validation set, and a second test set; The second model training subunit is used to train the preset initial image segmentation network model using the second training set, then optimize the parameters of the preset initial image segmentation network model using the second validation set, and optimize the hyperparameters of the preset initial image segmentation network model using the second test set, so as to obtain the implantation decision design model for the maxillary posterior tooth region.

4. The artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation as described in claim 3, characterized in that, The oral maxillary posterior tooth region fenestration bone graft decision design model construction unit is used to select a preset third number of oral CBCT image data from the oral CBCT image library, generate an oral maxillary posterior tooth region fenestration bone graft simulation scheme through the preset three-dimensional deep learning algorithm, and construct an oral maxillary posterior tooth region fenestration bone graft decision design model based on the oral maxillary posterior tooth region fenestration bone graft simulation scheme and the oral CBCT image data, including: The system comprises a second data preprocessing subunit, a second simulation scheme acquisition subunit, a second decision dataset acquisition subunit, a third dataset partitioning subunit, and a third model training subunit. The second data preprocessing subunit is used to preprocess the preset third number of oral CBCT image data; The second simulation scheme acquisition subunit is used to perform three-dimensional reconstruction and simulate the fenestration and bone grafting area of ​​the maxillary posterior tooth region in the oral CBCT image data in sequence through the preset three-dimensional deep learning algorithm, so as to obtain the simulation scheme of fenestration and bone grafting in the maxillary posterior tooth region. The second decision dataset acquisition subunit is used to match the simulated scheme of fenestration bone grafting in the maxillary posterior tooth region with the corresponding oral CBCT image data to obtain the decision dataset of fenestration bone grafting in the maxillary posterior tooth region. The third dataset partitioning subunit is used to divide the oral maxillary posterior tooth region fenestration bone grafting decision dataset into a third training set, a third validation set, and a third test set. The third model training subunit is used to train the preset initial image segmentation network model using the third training set, then optimize the parameters of the preset initial image segmentation network model using the third validation set, and optimize the hyperparameters of the preset initial image segmentation network model using the third test set, to obtain the decision design model for fenestration bone grafting in the maxillary posterior tooth region.

5. The artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation as described in claim 1, characterized in that, The second model building module is used to construct a downstream task model of the maxillary sinus based on the weights of the maxillary sinus basic model, including: The second data acquisition unit, data annotation unit, weight assignment unit, annotated data partitioning unit, and maxillary sinus downstream task model training unit; The second data acquisition unit is used to select a preset fourth number of oral CBCT image data from the pre-trained dataset; The data annotation unit is used to set classification labels and measurement labels for the study sites of the preset fourth number of oral CBCT image data based on a preset downstream task annotation method, and to construct an annotation dataset. The weighting unit is used to assign the encoder weights of the maxillary sinus basic model to the initial maxillary sinus downstream task model. The labeled data partitioning unit is used to divide the labeled dataset into a fourth training set, a fourth validation set, and a fourth test set. The maxillary sinus downstream task model training unit is used to train the initial maxillary sinus downstream task model using the fourth training set, optimize the initial maxillary sinus downstream task model using the fourth validation set, and perform performance testing on the initial maxillary sinus downstream task model using the fourth test set to obtain the maxillary sinus downstream task model.

6. A preoperative assessment system for maxillary posterior tooth implantation based on artificial intelligence as described in any one of claims 1 to 5, characterized in that, Also includes: The system includes an initial rule engine building module, a rule transformation module, a rule verification module, and a decision rule engine building module. The initial rule engine building module is used to build an initial rule engine based on a preset historical decision rule dataset; The rule conversion module is used to convert historical decision rule data into several rule conditions and several corresponding decision actions based on the initial rule engine. The rule verification module is used to perform logical consistency verification, coverage verification, and accuracy verification on the several rule conditions and several corresponding decision actions, and obtain logical consistency verification results, coverage verification results, and accuracy verification results. The decision rule engine building module is used to optimize the initial rule engine until the logical consistency verification result, coverage verification result, and accuracy verification result all meet the corresponding preset verification requirements, thus obtaining the decision rule engine.

7. The artificial intelligence-based preoperative assessment system for maxillary posterior tooth implantation as described in claim 6, characterized in that, The evaluation module is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model, and, based on the output qualitative and quantitative features, call the oral maxillary posterior tooth region implantation decision design model to generate a first decision or call the oral maxillary posterior tooth region fenestration bone grafting decision design model to generate a second decision, so as to perform preoperative evaluation of maxillary posterior tooth region implantation through the first decision or the second decision, including: Feature output unit, semantic definition unit, rule mapping unit, and model invocation and evaluation unit; The feature output unit is used to input the oral CBCT image data to be tested into the maxillary sinus downstream task model and output qualitative and quantitative features. The semantic definition unit is used to assign semantic definitions to the qualitative and quantitative features based on a preset semantic model; The rule mapping unit is used to map the semantic definitions of the qualitative and quantitative features into rule conditions and decision actions corresponding to the qualitative and quantitative features, based on the decision rule engine. The model invocation and evaluation unit is used to generate a first decision by invoking the oral maxillary posterior tooth region implantation decision design model or generate a second decision by invoking the oral maxillary posterior tooth region fenestration bone grafting decision design model based on the qualitative and quantitative feature corresponding rule conditions and the qualitative and quantitative feature corresponding decision actions, so as to conduct preoperative evaluation of maxillary posterior tooth region implantation through the first decision or the second decision.

Images