Artificial intelligence-based planting abutment automatic design method and system, terminal and storage medium

Abstract

The invention relates to the technical field of data analysis, and discloses an artificial intelligence-based implant abutment automatic design method and system, a terminal and a storage medium, and the method comprises the steps: taking an oral cavity scanning model of a target object, implant region description and corresponding abutment parameters as training data; the relationship between the oral space structure of the target object and the implant characteristics is learned through the encoder and the decoder, automatic prediction and personalized design of abutment parameters are realized, and a proper implant upper abutment model is selected. According to the method, a personalized abutment design scheme can be generated according to personalized features such as gingival thickness and implant position of the target object, and an adaptive implant upper abutment model is selected, so that the design efficiency and precision are improved, and manual measurement errors are reduced.

Description

Technical Field

[0001] This invention relates to the field of implantation abutment analysis technology, and in particular to an artificial intelligence-based automatic design method, system, terminal, and computer-readable storage medium for implantation abutments. Background Technology

[0002] In implant restoration, abutment design is a crucial step affecting the function and biological stability of the restoration. A suitable abutment requires comprehensive consideration of clinical parameters such as implant size, gingival morphology, and occlusal space. Inappropriate design may lead to uneven stress distribution or soft tissue inflammation, affecting the long-term stability of the implant.

[0003] Traditional methods typically rely on manual measurements to determine the geometric features of the restorative space, including key parameters such as gingival thickness (pergingival margin), implant location (diameter), and gingival-occlusal distance (height). The manual abutment pre-selection process is usually quite cumbersome, requiring multiple steps including oral scanning, model printing, space measurement, abutment pre-selection, and installation verification. This process is time-consuming and complex, not only inefficient but also prone to inaccurate abutment matching due to human error, thus increasing the risk of complications.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide an artificial intelligence-based automatic design method, system, terminal, and computer-readable storage medium for planting bases, aiming to solve the problems in the prior art where errors easily occur in manual judgment during the planting and restoration process, resulting in inaccurate base matching and low efficiency in scheme design.

[0006] To achieve the above objectives, the present invention provides an automatic design method for implantation abutments based on artificial intelligence, the method comprising the following steps: The training oral cavity model is divided into multiple local patches. After random occlusion processing of all the local patches, feature reconstruction is performed using an encoder and a decoder to obtain the reconstructed point set of the training oral cavity model. Based on the reconstructed point set, the chamfer distance loss and the first mean square error loss of the trained oral cavity model are constructed, and the encoder and the decoder are optimized using the chamfer distance loss and the first mean square error loss; The deep mesh features of the trained oral cavity model are obtained using the optimized encoder. The deep mesh features and the text features of the implantation area are input into the feedforward neural network for prediction, and the enhanced mesh features are output. Based on the enhanced mesh features and true results of the trained oral model, a smoothing loss function and a second mean squared error loss are constructed. The feedforward neural network is then optimized using these two loss functions. The optimized feedforward neural network is used to predict the mesh features and text features of the oral model under test, outputting the target prediction result. Using the predicted results of the target's transgingival height, cuff diameter, and abutment height, the similarity between the target repair parameters and the parameters of each candidate abutment is calculated. Finally, the abutment model with the highest similarity is selected from the candidate abutments, and the final abutment model recommendation result is output.

[0007] Optionally, the AI-based automatic implant abutment design method, wherein dividing the training oral model into multiple local patches, randomly occluding all the local patches, and then using an encoder and decoder to reconstruct features to obtain the reconstructed point set of the training oral model, specifically includes: A training oral cavity model of training samples is obtained, and multiple adjacent surfaces in the training oral cavity model are divided into individual patches, and the patches are divided into multiple local patches according to a preset standard. Random occlusion processing is performed on all the aforementioned local patches to obtain visible features and occlusion features; The visible features are input into the encoder for feature extraction, and the output is the mesh feature: ; ; in, Represents grid features, Indicates encoding processing, Indicates visible features, This indicates that it is data. Indicates the mask rate. This represents the total amount of input. Indicates the feature dimension; The visible features and the occlusion features are input into the decoder. The decoder sets the occlusion features to null and then concatenates them with the visible features to output the reconstructed mesh features. ; ; in, Indicates the reconstructed mesh features, Indicates decoding processing. This indicates splicing / joining. Indicates the occlusion feature after being set to empty; The mesh features are input into the decoder, which predicts the position of the occlusion features in the reconstructed mesh features based on the mesh features, and outputs the reconstructed point set of the trained oral cavity model.

[0008] Optionally, the AI-based automatic design method for planting platforms, wherein the random occlusion processing of all the local patches to obtain visible features and occlusion features specifically includes: Using a multilayer perceptron, each of the local patches is mapped to a feature embedding, and an input set is constructed based on all the feature embeddings; Randomly mask multiple feature embeddings in the input set, and define all unmasked feature embeddings as visible features and all masked features as masked features.

[0009] Optionally, the AI-based automatic implant abutment design method, wherein constructing the chamfer distance loss and the first mean square error loss of the trained oral model based on the reconstructed point set, and optimizing the encoder and the decoder using the chamfer distance loss and the first mean square error loss, specifically includes: Calculate the Euclidean distance between the predicted features and the occlusion features in the reconstructed point set, and construct the chamfer distance loss based on the Euclidean distance: ; in, Indicates chamfer distance loss. This represents the set of points that are obscured within the reconstructed point set. Represents the set of original points that are occluded. express The point in the middle, express The point in the middle; A first mean squared error loss is constructed based on the feature values ​​of the predicted feature and the feature values ​​of the occlusion feature: ; in, This represents the first mean square error loss. This represents the number of feature points in the predicted feature. The feature value representing the occlusion feature, The feature values ​​represent the predicted features. Represents the vertices of the patch that are occluded from 1 to 45; A comprehensive loss function is constructed based on the chamfer distance loss and the first mean square error loss, and the encoder and decoder are optimized and trained using the comprehensive loss function: ; in, Represents the comprehensive loss function. This represents the loss weights for facial features.

[0010] Optionally, the artificial intelligence-based automatic implant abutment design method, wherein the step of using an optimized encoder to obtain the deep mesh features of the trained oral cavity model, inputting the deep mesh features and the text features of the implant area into a feedforward neural network for prediction, and outputting enhanced mesh features, specifically includes: The visible and occlusion features are input into the optimized encoder for deep feature extraction, and the deep mesh features are output: ; ; in, Indicates deep mesh features, This represents the sum of visible and occluded features; Obtain the descriptive text of the implantation region of the trained oral cavity model, input the descriptive text into the contrastive language-image pre-trained model for feature extraction, and output the text features: ; ; in, Representing text features, This indicates that a contrastive language-image pre-trained model is used for feature extraction. express, Indicates the dimension of text features; The deep mesh features are subjected to max pooling to obtain mesh embedding features. These mesh embedding features and the text features are then input into a feedforward neural network, which fuses them to obtain fused features. ; ; ; in, Indicates fusion features, This indicates the processing of the feedforward neural network. Represents grid embedding features, Indicates the feature dimension of the mesh embedding; The feedforward neural network uses the text features to enhance the fused features, resulting in enhanced grid features: ; ; in, Indicates enhanced mesh features, This indicates an enhancement of the grid feature dimension.

[0011] Optionally, the artificial intelligence-based automatic implant abutment design method, wherein constructing a smoothing loss function and a second mean square error loss based on the enhanced mesh features and real results of the trained oral model, and optimizing the feedforward neural network using the smoothing loss function and the second mean square error loss, specifically includes: Obtain the true results of the trained oral cavity model, and construct a smoothing loss function and a second mean squared error loss for each sample based on the enhanced mesh features and the true results: ; in, Indicates the first A smoothing loss function, Indicating the enhanced mesh feature, the first The first predicted value and the actual result The difference between the true values, Represents a smoothing function. The coefficients of a linear function are represented. Describe the domain of a linear function; ; in, express The second mean square error loss, The symbol represents the mean squared error loss that distinguishes the different roles played in the pre-training phase. This represents the total number of samples for enhanced mesh features or true results; Based on each of the smoothing loss functions and each of the second mean squared error losses, a total smoothing loss function and a total second mean squared error loss are constructed respectively. A joint loss function is then constructed based on the total smoothing loss function and the total second mean squared error loss. ; in, Denotes the joint loss function. Represents the total smoothing loss function. This represents the total second mean square error loss; The feedforward neural network is optimized using the joint loss function to obtain the target feedforward neural network.

[0012] Optionally, the artificial intelligence-based automatic implant abutment design method, wherein the step of using an optimized feedforward neural network to predict the grid features and text features of the test oral cavity model and outputting the target prediction result specifically includes: The test text features of the oral cavity model to be tested are obtained, and the test oral cavity model is processed using the optimized encoder and encoder to output the test deep mesh features. The test deep mesh features and the test text features are input into the target feedforward neural network, and the target prediction result of the test oral cavity model is output.

[0013] Furthermore, to achieve the above objectives, the present invention also provides an automated design system for implantation bases based on artificial intelligence, wherein the automated design system for implantation bases based on artificial intelligence includes: The feature reconstruction module is used to divide the training oral cavity model into multiple local patches, perform random occlusion processing on all the local patches, and then use the encoder and decoder to perform feature reconstruction to obtain the reconstructed point set of the training oral cavity model. The first optimization module is used to construct the chamfer distance loss and the first mean square error loss of the trained oral cavity model based on the reconstructed point set, and to optimize the encoder and the decoder using the chamfer distance loss and the first mean square error loss. The feature enhancement module is used to obtain the deep mesh features of the trained oral cavity model using the optimized encoder, input the deep mesh features and the text features of the implantation area into the feedforward neural network for prediction, and output the enhanced mesh features. The second optimization module constructs a smoothing loss function and a second mean squared error loss based on the enhanced mesh features and true results of the trained oral model. It then optimizes the feedforward neural network using these two loss functions, and uses the optimized feedforward neural network to predict the mesh features and text features of the oral model under test, outputting the target prediction result. Using the predicted results of the target's transgingival height, cuff diameter, and abutment height, it calculates the similarity between the target repair parameters and the parameters of each candidate abutment. The similarity can be calculated using Euclidean distance, weighted distance, similarity score, or other alternative metrics; the system is not limited to a specific algorithm. Finally, it selects the abutment model with the highest similarity from the candidate abutments and outputs the final recommendation result.

[0014] In addition, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and an artificial intelligence-based automatic design program for implantation bases stored in the memory and executable on the processor, wherein when the artificial intelligence-based automatic design program for implantation bases is executed by the processor, it implements the steps of the artificial intelligence-based automatic design method for implantation bases as described above.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an artificial intelligence-based automatic design program for implantation bases, and when the artificial intelligence-based automatic design program for implantation bases is executed by a processor, it implements the steps of the artificial intelligence-based automatic design method for implantation bases as described above.

[0016] In this invention, the training oral cavity model is divided into multiple local patches. After random occlusion processing of all the local patches, feature reconstruction is performed using an encoder and a decoder to obtain the reconstructed point set of the training oral cavity model. Based on the reconstructed point set, a chamfer distance loss and a first mean square error loss are constructed for the training oral cavity model, and the encoder and decoder are optimized using the chamfer distance loss and the first mean square error loss. The optimized encoder is used to obtain the deep mesh features of the training oral cavity model, and the deep mesh features and text features of the implantation area are input into a feedforward neural network for prediction, outputting enhanced mesh features. Based on the enhanced mesh features of the training oral cavity model and the true results, a smoothing loss function and a second mean square error loss are constructed. The smoothing loss function and the second mean square error loss are used to optimize the feedforward neural network, and the optimized feedforward neural network is used to predict the mesh features and text features of the oral cavity model to be tested, outputting the target prediction result and selecting a suitable implant abutment model. This invention can generate personalized abutment schemes based on individual characteristics such as gingival thickness and implant position of the target object, and select suitable implant superabutment models, thereby improving design efficiency and accuracy and reducing manual measurement errors. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the artificial intelligence-based automatic design method for planting platforms of the present invention; Figure 2 This is a flowchart illustrating the process of manually determining abutment parameters in a preferred embodiment of the artificial intelligence-based automatic design method for planting abutments of the present invention. Figure 3 This is a general network framework diagram of a preferred embodiment of the artificial intelligence-based automatic design method for planting platforms of the present invention; Figure 4 This is a structural diagram of a preferred embodiment of the artificial intelligence-based automatic design system for planting platforms of the present invention; Figure 5 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] The preferred embodiment of the present invention describes an automated design method for implantation abutments based on artificial intelligence, such as... Figure 1 As shown, the AI-based automatic design method for planting platforms includes the following steps: Step S10: Divide the training oral cavity model into multiple local patches, perform random occlusion processing on all the local patches, and then use the encoder and decoder to reconstruct features to obtain the reconstructed point set of the training oral cavity model.

[0020] Measuring the restorative space is a crucial step in implant abutment pre-selection and design. Traditional methods typically rely on manual measurements to determine the geometric characteristics of the restorative space, including key parameters such as gingival thickness (perforation), implant location (diameter), and gingival-occlusal distance (height). The manual abutment pre-selection process is usually quite cumbersome (e.g., Figure 2 As shown in the image, the process requires multiple steps, including oral cavity scanning, model printing, spatial measurement, abutment pre-selection, and installation verification. This process is time-consuming and complex, resulting in low efficiency and a high risk of complications due to human error in abutment matching.

[0021] In recent years, although computer-aided design and manufacturing (CAD-CAM) technologies have been gradually introduced, manual positioning and parameter setting are still required in the measurement and design stages, and data deviations may occur during the measurement process. Therefore, this invention proposes an artificial intelligence-based method for automatic base station design, including a pre-training stage and a fine-tuning stage (such as...). Figure 3 As shown in the figure, key parameters for abutment design can be predicted by simply inputting the patient's oral cavity scan model and implant information, without the need for complicated and time-consuming operations, thus simplifying the abutment design process.

[0022] Specifically, a training oral cavity model of training samples is obtained; multiple adjacent faces in the training oral cavity model are divided into individual patches; and each patch is further divided into multiple local patches according to a preset standard; random occlusion processing is performed on all local patches to obtain visible features and occlusion features; the visible features are input into an encoder for feature extraction, and mesh features are output. ; ; in, Represents grid features, Indicates encoding processing, Indicates visible features, This indicates that it is data. Indicates the mask rate. This represents the total amount of input. The feature dimension is represented; the visible feature and the occlusion feature are input into the decoder. The decoder sets the occlusion feature to null and then concatenates it with the visible feature to output the reconstructed mesh feature. ; ; in, Indicates the reconstructed mesh features, Indicates decoding processing. This indicates splicing / joining. The occlusion feature is represented after being empty; the grid feature is input into the decoder, and the decoder predicts the position of the occlusion feature in the reconstructed grid feature based on the grid feature, and outputs the reconstructed point set of the trained oral cavity model.

[0023] In the pre-training phase of the neural network, an oral cavity scanning model of the target object is used for pre-training. Random masking and reconstruction of the oral cavity scanning model enhances the understanding of oral structure and the ability to capture key features. Specifically, the oral cavity scanning model, composed of a grid, is divided into several local patches by using a pre-defined algorithm, with each of the 64 adjacent faces forming a patch. The features of each patch are formed by the ordered concatenation of the 13-dimensional feature vectors of the constituent faces. These features are then mapped using a multilayer perceptron to obtain feature embeddings, resulting in input features (i.e., a set of tokens, where a token is the smallest semantic unit when the model processes text information). Some tokens are randomly masked according to a masking rate, resulting in the remaining visible tokens (i.e., the aforementioned visible features). This random masking strategy enhances the model's ability to recover local structures from context, improves the understanding of oral geometric features and generalization performance, while reducing training computation.

[0024] Furthermore, using a multilayer perceptron, each local patch is mapped to a feature embedding, and an input set is constructed based on all the feature embeddings; multiple feature embeddings in the input set are randomly masked, and all unmasked feature embeddings are defined as visible features, and all masked features are defined as masked features.

[0025] The visible features are input into the encoder to extract their deep semantic and geometric features, resulting in encoded mesh features. This captures the global topological structure and spatial relationships of the input oral cavity scanning model, providing contextual information for subsequent feature recovery. The features output by the encoder are then decoded to reconstruct the features of the masked region. The decoder infers the spatial and geometric features of the masked region using contextual information, enabling the model to reconstruct the complete oral cavity mesh structure.

[0026] In this context, it is understandable that the decoder's concatenation of visible and occluded features actually involves concatenating the indices of the visible and occluded features, thereby ensuring that subsequent models understand which features need to be predicted.

[0027] Step S20: Construct the chamfer distance loss and the first mean square error loss of the trained oral cavity model based on the reconstructed point set, and optimize the encoder and the decoder using the chamfer distance loss and the first mean square error loss.

[0028] The constructed chamfer distance loss and first mean square error loss can be used to evaluate the reconstruction accuracy of the oral scan model in the pre-training stage. By combining geometric space error and feature reconstruction error, the network's structural restoration ability and stability can be improved.

[0029] Specifically, the Euclidean distance between the predicted features and the occlusion features in the reconstructed point set is calculated, and a chamfer distance loss is constructed based on the Euclidean distance: ; in, Indicates chamfer distance loss. This represents the set of points that are obscured within the reconstructed point set. Represents the set of original points that are occluded. express The point in the middle, express The points in the middle; construct a first mean squared error loss based on the feature values ​​of the predicted feature and the feature values ​​of the occlusion feature: ; in, This represents the first mean square error loss. This represents the number of feature points in the predicted feature. The feature value representing the occlusion feature, The feature values ​​represent the predicted features. The vertices of patches 1 to 45 that are occluded are represented; a comprehensive loss function is constructed based on the chamfer distance loss and the first mean square error loss, and the encoder and decoder are optimized and trained using the comprehensive loss function: ; in, Represents the comprehensive loss function. This represents the loss weights for facial features.

[0030] In order to measure the consistency between the reconstructed model and the original model in geometric space, in the embodiments disclosed in this invention, the chamfer distance is used as a geometric error index. By calculating the Euclidean distance between the occluded point set in the reconstructed point set and the original point set before occlusion, the standard measure of the difference in position between the two points in three-dimensional space is obtained, thereby measuring the accuracy of the model in restoring the shape in three-dimensional space, which can effectively reflect the model's ability to reconstruct the oral cavity geometry.

[0031] Furthermore, in order to constrain the numerical consistency of feature reconstruction, a first mean squared error loss is introduced. This loss value is mainly obtained by comparing the outputs of the encoder and decoder, and is used to constrain the consistency between the reconstructed features and the original features in the numerical domain, thereby improving the stability of network feature representation.

[0032] Furthermore, by combining the first mean squared error loss and the chamfer distance loss, and simultaneously optimizing the encoder and decoder, collaborative learning of features and geometry is achieved. This enables the network to learn the spatial geometric structure and feature representation rules of the oral cavity scanning model during the pre-training stage, thereby providing a more robust and high-fidelity feature representation for the abutment parameter prediction in the subsequent fine-tuning stage.

[0033] Step S30: Use the optimized encoder to obtain the deep mesh features of the trained oral cavity model, input the deep mesh features and the text features of the implantation area into the feedforward neural network for prediction, and output the enhanced mesh features.

[0034] Specifically, the visible features and the occlusion features are input into the optimized encoder for deep feature extraction, and the deep mesh features are output: ; ; in, Indicates deep mesh features, This represents the sum of visible and occlusion features; the descriptive text of the implantation region of the trained oral cavity model is obtained, and the descriptive text is input into a contrastive language-image pre-trained model for feature extraction, outputting text features: ; ; in, Representing text features, This indicates that a contrastive language-image pre-trained model is used for feature extraction. express, The deep grid features are represented by a dimension; max pooling is performed on the deep grid features to obtain grid embedding features, and the grid embedding features and the text features are input into a feedforward neural network. The feedforward neural network fuses the grid embedding features and the text features to obtain fused features. ; ; ; in, Indicates fusion features, This indicates the processing of the feedforward neural network. Represents grid embedding features, The dimension of the grid embedding feature is represented; the feedforward neural network uses the text features to enhance the fused features, resulting in enhanced grid features: ; ; in, Indicates enhanced mesh features, This indicates an enhancement of the grid feature dimension.

[0035] Having acquired the optimized encoder and decoder, the textual description of the implantation area can be fused with the grid features of the oral cavity scanning model across modalities. By introducing a semantic guidance mechanism, the model can accurately focus on the spatial location and structural features of the target implantation area, thereby improving positioning accuracy and personalized design capabilities.

[0036] In the fine-tuning phase, the encoder trained in the pre-training phase is used as the feature extractor. Data processing is similar to pre-training, but without random masking. All tokens of the oral cavity scanning model (i.e., all visible features and occluded features before masking) are input into the encoder to extract deep semantic and geometric features, resulting in the input deep mesh features. The features output by the pre-trained encoder in the fine-tuning phase provide a high-fidelity, stable oral cavity mesh representation, preserving local and global geometric information. These features are used to fuse with text descriptions, guiding the network to focus on the implant area, thereby improving the accuracy of abutment key parameter prediction.

[0037] Furthermore, the descriptive text of the planting area is input into the Contrastive Language–Image Pre-training (CLIP) model to extract the textual features of the planting area, thereby mapping the natural language description to the vector space and achieving feature alignment and fusion with the grid features.

[0038] Furthermore, textual features and grid features are fused at the feature level to construct cross-modal associations. A deep fusion of textual semantics and spatial geometric features is achieved through nonlinear mapping, thereby enhancing the network's understanding of the semantics of the planting area. Then, based on the fused features, textual features are again used to guide the spatial features, outputting the final enhanced grid features. This process superimposes and maps textual semantic information with the fused visual features, enabling the model to focus on planting areas related to the descriptive semantics in the three-dimensional grid space.

[0039] Step S40: Construct a smoothing loss function and a second mean square error loss based on the enhanced grid features and true results of the trained oral cavity model. Optimize the feedforward neural network using the smoothing loss function and the second mean square error loss. Then, use the optimized feedforward neural network to predict the grid features and text features of the oral cavity model to be tested, and output the target prediction result.

[0040] Specifically, the true results of the trained oral cavity model are obtained, and a smoothing loss function and a second mean squared error loss are constructed for each sample based on the enhanced mesh features and the true results: ; in, Indicates the first A smoothing loss function, Indicating the enhanced mesh feature, the first The first predicted value and the actual result The difference between the true values, Represents a smoothing function. The coefficients of a linear function are represented. Describe the domain of a linear function; ; in, express The second mean square error loss, The symbol represents the mean squared error loss that distinguishes the different roles played in the pre-training phase. The total number of samples representing the enhanced grid features or the true result; based on each of the smoothing loss functions and each of the second mean squared error losses, a total smoothing loss function and a total second mean squared error loss are constructed respectively; a joint loss function is constructed based on the total smoothing loss function and the total second mean squared error loss. ; in, Denotes the joint loss function. Represents the total smoothing loss function. The second mean squared error loss represents the total loss; the feedforward neural network is optimized using the joint loss function to obtain the target feedforward neural network.

[0041] To supervise the prediction results of key base plate parameters during the model fine-tuning phase, this invention introduces a measurement method combining a smoothing loss function and a second mean squared error loss function. This enhances the model's robustness and stability while improving prediction accuracy. The smoothing loss function maintains quadratic continuity within a small error range, balancing stability and outlier resistance during training. It employs a quadratic function form to ensure smooth optimization when errors are small, while using a linear function to mitigate the impact of outliers when errors are large, effectively improving training stability and generalization performance. The second mean squared error loss function precisely constrains the overall prediction error, ensuring the model output closely approximates the true parameters numerically. To balance error accuracy and convergence smoothness, this invention uses a joint loss function of the smoothing loss function and the second mean squared error loss function to optimize the feedforward neural network. This reduces overall error while effectively suppressing interference from outlier data points, achieving high-precision prediction and stable training of base plate parameters. Through joint optimization of multiple loss functions, the model can automatically output high-precision personalized base plate parameters, overcoming the human bias and measurement errors present in traditional CAD-CAM design.

[0042] Further, the test text features of the oral cavity model to be tested are obtained, and the test oral cavity model is processed using the optimized encoder and encoder to output the test deep mesh features; the test deep mesh features and the test text features are input into the target feedforward neural network to output the target prediction result of the test oral cavity model.

[0043] Furthermore, after obtaining the target prediction results, a design scheme can be constructed based on the predicted abutment parameters, and similarity calculations can be performed with the parameters of the candidate abutments to obtain the matching degree between each candidate abutment and the target parameters. The abutment model with the highest matching degree is then selected as the recommended result. The similarity calculation of this invention can employ Euclidean distance, weighted distance, neural network similarity measurement, or other equivalent methods. This invention utilizes the predicted results of the target's perforation height, cuff diameter, and abutment height, and uses Euclidean distance, weighted distance, similarity score, or other alternative metrics to calculate the difference between the target repair parameters and the parameters of each candidate abutment to calculate the similarity. Finally, the abutment model with the highest similarity is selected from the candidate abutments, and the final recommended result is output.

[0044] This invention is the first to apply artificial intelligence models to the automated design process of implant abutments. Through intelligent analysis based on patient oral scan data and implant area description information, it achieves automatic prediction and generation of abutment parameters, significantly simplifying the complex process that traditionally relied on manual measurement and design. This solution supports the mass production and standardization of implant abutments, greatly improving design efficiency and consistency, and providing an automated design path for dental implant restoration.

[0045] This invention can generate personalized abutment solutions based on individual characteristics such as gingival thickness and implant location of the target patient, thereby improving design efficiency and accuracy and reducing errors from manual measurement.

[0046] Furthermore, such as Figure 4 As shown, based on the above-mentioned AI-based automatic design method for planting bases, the present invention also provides an AI-based automatic design system for planting bases, wherein the AI-based automatic design system for planting bases includes: The feature reconstruction module 51 is used to divide the training oral cavity model into multiple local patches, perform random occlusion processing on all the local patches, and then use the encoder and decoder to perform feature reconstruction to obtain the reconstructed point set of the training oral cavity model. The first optimization module 52 is used to construct the chamfer distance loss and the first mean square error loss of the trained oral cavity model based on the reconstructed point set, and to optimize the encoder and the decoder using the chamfer distance loss and the first mean square error loss. The feature enhancement module 53 is used to obtain the deep mesh features of the trained oral cavity model using the optimized encoder, input the deep mesh features and the text features of the implantation area into the feedforward neural network for prediction, and output the enhanced mesh features. The second optimization module 54 is used to construct a smoothing loss function and a second mean square error loss based on the enhanced grid features and true results of the trained oral cavity model, optimize the feedforward neural network using the smoothing loss function and the second mean square error loss, and use the optimized feedforward neural network to predict the grid features and text features of the oral cavity model to be tested, and output the target prediction result.

[0047] Furthermore, such as Figure 5 As shown, based on the above-mentioned artificial intelligence-based automatic design method and system for planting platforms, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 5 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0048] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores an AI-based automatic planting platform design program 40, which can be executed by the processor 10 to implement the AI-based automatic planting platform design method of this application.

[0049] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the artificial intelligence-based automatic design method for planting bases.

[0050] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0051] In one embodiment, when the processor 10 executes the AI-based implantation abutment automatic design program 40 in the memory 20, it implements the steps of the AI-based implantation abutment automatic design method as described above.

[0052] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an artificial intelligence-based automatic design program for implantation bases, which, when executed by a processor, implements the steps of the artificial intelligence-based automatic design method for implantation bases as described above.

[0053] In summary, this invention provides an automatic design method and related equipment for implant abutments based on artificial intelligence. The method includes: dividing a training oral model into multiple local patches; randomly occluding all the local patches; reconstructing features using an encoder and a decoder to obtain a reconstructed point set of the training oral model; constructing a chamfer distance loss and a first mean square error loss for the training oral model based on the reconstructed point set; optimizing the encoder and decoder using the chamfer distance loss and the first mean square error loss; acquiring deep mesh features of the training oral model using the optimized encoder; inputting the deep mesh features and text features of the implant region into a feedforward neural network for prediction; outputting enhanced mesh features; constructing a smoothing loss function and a second mean square error loss based on the enhanced mesh features and the true results of the training oral model; optimizing the feedforward neural network using the smoothing loss function and the second mean square error loss; predicting the mesh features and text features of the oral model to be tested using the optimized feedforward neural network; outputting the target prediction result and selecting a suitable implant abutment model. This invention can generate personalized abutment schemes based on individual characteristics such as gingival thickness and implant position of the target object, and select suitable implant superabutment models, thereby improving design efficiency and accuracy and reducing manual measurement errors.

[0054] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0055] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0056] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. An automated design method for planting platforms based on artificial intelligence, characterized in that, The AI-based automatic design method for implantation platforms includes: The training oral cavity model is divided into multiple local patches. After random occlusion processing of all the local patches, feature reconstruction is performed using an encoder and a decoder to obtain the reconstructed point set of the training oral cavity model. Based on the reconstructed point set, the chamfer distance loss and the first mean square error loss of the trained oral cavity model are constructed, and the encoder and the decoder are optimized using the chamfer distance loss and the first mean square error loss; The deep mesh features of the trained oral cavity model are obtained using the optimized encoder. The deep mesh features and the text features of the implantation area are input into the feedforward neural network for prediction, and the enhanced mesh features are output. Based on the enhanced grid features and true results of the trained oral cavity model, a smoothing loss function and a second mean square error loss are constructed. The feedforward neural network is optimized using the smoothing loss function and the second mean square error loss. The optimized feedforward neural network is then used to predict the grid features and text features of the oral cavity model to be tested, and the target prediction result is output.

2. The artificial intelligence-based automatic design method for planting platforms according to claim 1, characterized in that, The process of dividing the trained oral cavity model into multiple local patches, randomly occluding all the local patches, and then using an encoder and decoder to reconstruct features to obtain the reconstructed point set of the trained oral cavity model specifically includes: A training oral cavity model of training samples is obtained, and multiple adjacent surfaces in the training oral cavity model are divided into individual patches, and the patches are divided into multiple local patches according to a preset standard. Random occlusion processing is performed on all the aforementioned local patches to obtain visible features and occlusion features; The visible features are input into the encoder for feature extraction, and the output is the mesh feature: ； ； in, Represents grid features, Indicates encoding processing, Indicates visible features, This indicates that it is data. Indicates the mask rate. This represents the total amount of input. Indicates feature dimension; The visible features and the occlusion features are input into the decoder. The decoder sets the occlusion features to null and then concatenates them with the visible features to output the reconstructed mesh features. ； ； in, Indicates the reconstructed mesh features, Indicates decoding processing. This indicates splicing / joining. This indicates the occlusion feature after being set to empty; The mesh features are input into the decoder, which predicts the position of the occlusion features in the reconstructed mesh features based on the mesh features, and outputs the reconstructed point set of the trained oral cavity model.

3. The artificial intelligence-based automatic design method for planting platforms according to claim 2, characterized in that, The random occlusion process performed on all the local patches to obtain visible and occlusion features specifically includes: Using a multilayer perceptron, each of the local patches is mapped to a feature embedding, and an input set is constructed based on all the feature embeddings; Randomly mask multiple feature embeddings in the input set, and define all unmasked feature embeddings as visible features and all masked features as masked features.

4. The artificial intelligence-based automatic design method for planting platforms according to claim 2, characterized in that, The step of constructing the chamfer distance loss and the first mean square error loss of the trained oral cavity model based on the reconstructed point set, and optimizing the encoder and the decoder using the chamfer distance loss and the first mean square error loss, specifically includes: Calculate the Euclidean distance between the predicted features and the occlusion features in the reconstructed point set, and construct the chamfer distance loss based on the Euclidean distance: ； in, Indicates chamfer distance loss. This represents the set of points that are obscured within the reconstructed point set. This represents the set of original points that are occluded. express The point in the middle, express The point in the middle; A first mean squared error loss is constructed based on the feature values ​​of the predicted feature and the feature values ​​of the occlusion feature: ； in, This represents the first mean square error loss. This represents the number of feature points in the predicted feature. The feature value representing the occlusion feature, The feature values ​​represent the predicted features. Represents the vertices of the patch that are occluded from 1 to 45; A comprehensive loss function is constructed based on the chamfer distance loss and the first mean square error loss, and the encoder and decoder are optimized and trained using the comprehensive loss function: ； in, Represents the comprehensive loss function. This represents the loss weights for facial features.

5. The artificial intelligence-based automatic design method for planting platforms according to claim 2, characterized in that, The process involves using an optimized encoder to obtain deep mesh features of the trained oral cavity model, inputting these deep mesh features and text features of the implantation area into a feedforward neural network for prediction, and outputting enhanced mesh features. Specifically, this includes: The visible and occlusion features are input into the optimized encoder for deep feature extraction, and the deep mesh features are output: ； ； in, Indicates deep mesh features, This represents the sum of visible and occluded features; Obtain the descriptive text of the implantation region of the trained oral cavity model, input the descriptive text into the contrastive language-image pre-trained model for feature extraction, and output the text features: ； ； in, Representing text features, This indicates that a contrastive language-image pre-trained model is used for feature extraction. express, Indicates the dimension of text features; The deep mesh features are subjected to max pooling to obtain mesh embedding features. These mesh embedding features and the text features are then input into a feedforward neural network, which fuses them to obtain fused features. ； ； ； in, Indicates fusion features, This indicates the processing of the feedforward neural network. Represents grid embedding features, Indicates the feature dimension of the mesh embedding; The feedforward neural network uses the text features to enhance the fused features, resulting in enhanced grid features: ； ； in, Indicates enhanced mesh features, This indicates an enhancement of the grid feature dimension.

6. The artificial intelligence-based automatic design method for planting platforms according to claim 2, characterized in that, The step of constructing a smoothing loss function and a second mean square error loss based on the enhanced mesh features and true results of the trained oral cavity model, and optimizing the feedforward neural network using the smoothing loss function and the second mean square error loss, specifically includes: Obtain the true results of the trained oral cavity model, and construct a smoothing loss function and a second mean squared error loss for each sample based on the enhanced mesh features and the true results: ； in, Indicates the first A smoothing loss function, Indicating the enhanced mesh feature, the first The first predicted value and the actual result The difference between the true values, Represents a smoothing function. The coefficients of a linear function are represented. Describe the domain of a linear function; ； in, express The second mean square error loss, The symbol represents the mean squared error loss that distinguishes the different roles played in the pre-training phase. This represents the total number of samples for enhanced mesh features or true results; Based on each of the smoothing loss functions and each of the second mean squared error losses, a total smoothing loss function and a total second mean squared error loss are constructed respectively. A joint loss function is then constructed based on the total smoothing loss function and the total second mean squared error loss. ； in, Denotes the joint loss function. Represents the total smoothing loss function. This represents the total second mean square error loss; The feedforward neural network is optimized using the joint loss function to obtain the target feedforward neural network.

7. The artificial intelligence-based automatic design method for planting platforms according to claim 6, characterized in that, The process of using an optimized feedforward neural network to predict the grid features and text features of the oral cavity model to be tested, and outputting the target prediction result, specifically includes: The test text features of the oral cavity model to be tested are obtained, and the test oral cavity model is processed using the optimized encoder and encoder to output the test deep mesh features. The test deep mesh features and the test text features are input into the target feedforward neural network, and the target prediction result of the test oral cavity model is output.

8. An automated design system for planting platforms based on artificial intelligence, characterized in that, The AI-based automated planting platform design system is applied to the AI-based automated planting platform design method as described in any one of claims 1-7, wherein the AI-based automated planting platform design system comprises: The feature reconstruction module is used to divide the training oral cavity model into multiple local patches, perform random occlusion processing on all the local patches, and then use the encoder and decoder to perform feature reconstruction to obtain the reconstructed point set of the training oral cavity model. The first optimization module is used to construct the chamfer distance loss and the first mean square error loss of the trained oral cavity model based on the reconstructed point set, and to optimize the encoder and the decoder using the chamfer distance loss and the first mean square error loss. The feature enhancement module is used to obtain the deep mesh features of the trained oral cavity model using the optimized encoder, input the deep mesh features and the text features of the implantation area into the feedforward neural network for prediction, and output the enhanced mesh features. The second optimization module is used to construct a smoothing loss function and a second mean square error loss based on the enhanced grid features and true results of the trained oral cavity model. The smoothing loss function and the second mean square error loss are used to optimize the feedforward neural network. The optimized feedforward neural network is then used to predict the grid features and text features of the oral cavity model to be tested, and the target prediction result is output.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and an AI-based automatic implantation platform design program stored in the memory and executable on the processor. When the AI-based automatic implantation platform design program is executed by the processor, it implements the steps of the AI-based automatic implantation platform design method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an AI-based automatic design program for implantation abutments, which, when executed by a processor, implements the steps of the AI-based automatic design method for implantation abutments as described in any one of claims 1-7.

Images