A method and system for three-dimensional reconstruction of oral cavity based on low-dose projection data

Abstract

The application discloses a kind of oral cavity three-dimensional image reconstruction method and system based on low-dose projection data, belong to oral cavity medical image processing technical field, comprising: obtaining the individualized tooth digital twin model and low-dose projection data of target tooth area;The individualized tooth digital twin model is projected forward simulation, and with low-dose projection data is calculated residual error pixel by pixel, and it is projected backward mapping, generates structure mapping residual field;According to the structure mapping residual field, the low-dose projection data is locally sampled and back projection, forms iterative projection correction field, according to the low-dose projection data is corrected according to the iterative projection correction field.This scheme can focus on key areas by the dynamic local sampling mechanism guided by residual field, combined with fast local back projection, while maintaining overall computing efficiency, significantly improve the reconstruction accuracy of local features.

Description

Technical Field

[0001] This invention belongs to the field of oral medical image processing technology, specifically relating to a method and system for three-dimensional oral image reconstruction based on low-dose projection data. Background Technology

[0002] With the rapid development of oral medicine and digital diagnosis and treatment technologies, three-dimensional oral imaging reconstruction has become an indispensable basic technology in modern dental diagnosis and treatment. Three-dimensional oral images can intuitively and accurately reflect the teeth, periodontal tissues and adjacent jawbone structures, providing accurate data support for oral disease diagnosis, treatment planning, surgical simulation and personalized restoration. In this technological environment, low-dose projection imaging technology has gradually attracted attention. It can obtain two-dimensional projection data of the tooth area while reducing the patient's radiation exposure, providing original information for three-dimensional image reconstruction.

[0003] However, in the process of implementing the technical solutions of the present application, it was found that the above-mentioned technologies have at least the following technical problems: In the prior art, some methods use low-dose projection data combined with conventional reconstruction algorithms, such as filtered back projection and iterative reconstruction, which can reconstruct the overall morphology of teeth to a certain extent and suppress noise. The advantage of such methods is that the algorithms are mature, the computational efficiency is relatively high, and they can provide usable three-dimensional oral structure models under conventional low-dose conditions. The above solutions are difficult to accurately capture the detailed changes in the microstructure of individual teeth, and in the time series reconstruction of tooth wear or occlusal evolution, it is often difficult to fully optimize the target individualized tooth information and low-dose projection data, resulting in inaccurate microstructure features and insufficient reconstruction quality of wear-sensitive areas and highly complex structural areas. Summary of the Invention

[0004] This invention addresses the problem that existing algorithms struggle to dynamically correlate and optimize individualized tooth models with low-dose projection data, resulting in insufficient reconstruction of structural details in highly complex regions. It provides a method for reconstructing three-dimensional oral images based on low-dose projection data, constructing a dynamic interaction mechanism between a digital twin model and measured data. The method quantifies and locates model errors through a structural mapping residual field, performs data-driven local correction using an iterative projection correction field, and achieves coupled modeling of anatomical structure and functional changes by combining a wear evolution probability field. Finally, it achieves fine-grained fusion of multiple reconstruction results through residual decomposition and confidence matrices.

[0005] According to one aspect of this specification, a method for three-dimensional oral cavity image reconstruction based on low-dose projection data is provided, comprising:

[0006] Acquire individualized digital twin models and low-dose projection data of the target tooth region;

[0007] A forward projection simulation is performed on the individualized digital twin model of teeth, and the residual is calculated with the low-dose projection data. The residual results are then back-projected and mapped to generate a structure-mapping residual field.

[0008] Based on the structure mapping residual field, the low-dose projection data is locally sampled and back-projected to form an iterative projection correction field, and the low-dose projection data is corrected based on the iterative projection correction field.

[0009] A wear evolution probability field is generated by coupling a personalized digital twin model of teeth with corrected low-dose projection data.

[0010] Candidate reconstruction results are generated based on the corrected low-dose projection data, individualized digital twin models of teeth, and wear evolution probability fields.

[0011] Based on the candidate reconstruction results, a consistency confidence matrix is ​​obtained through residual decomposition and probability weighting. The individualized tooth digital twin model is then optimized for global consistency by combining the structural mapping residual field and the wear evolution probability field, thereby generating a three-dimensional reconstruction result.

[0012] Further, the structure mapping residual field is generated, including:

[0013] Perform forward projection simulation on individualized digital twin models of teeth to generate simulated projection sequences;

[0014] The residuals between the simulated projection sequence and the low-dose projection data are calculated pixel by pixel to generate a projection residual sequence.

[0015] The projected residual sequence is mapped back to the three-dimensional volume domain by reverse weighted back projection, thus generating the three-dimensional volume domain residual.

[0016] Based on the individualized digital twin model of teeth, the three-dimensional volume domain residuals are decomposed into multiple scales, and the spatial morphology-preserving filter is applied to the decomposition results at each scale to generate the initial structure mapping residual field.

[0017] The initial structure mapping residual field is verified, and a structure mapping residual field is generated.

[0018] Furthermore, the initial structure mapping residual field is verified, and a structure mapping residual field is generated, including:

[0019] The simulated projection sequence is corrected based on the residual field mapped from the initial structure.

[0020] The residuals of the corrected simulated projection sequence and the low-dose projection data are calculated pixel by pixel to generate the corrected projection residual sequence, and then the difference between the corrected projection residual sequence and the projection residual sequence is processed.

[0021] Determine the difference processing result and the preset discrimination threshold: When the difference processing result is greater than the preset discrimination threshold, use the initial structure mapping residual field as the structure mapping residual field.

[0022] Furthermore, determining the magnitude of the difference processing result and the preset discrimination threshold also includes:

[0023] When the difference processing result is not greater than the preset discrimination threshold, the initial structure mapping residual field is regenerated according to the corrected projection residual sequence until the preset discrimination threshold is met, and the corresponding initial structure mapping residual field is used as the structure mapping residual field.

[0024] Furthermore, the low-dose projection data is corrected based on the iterative projection correction field, including:

[0025] Based on the structure mapping residual field, low-dose projection data are randomly sampled in the projection domain to obtain each local subdomain.

[0026] The response distribution of each local subdomain in the volume domain is obtained by fast local backprojection, and the optimal local projection correction coefficient is obtained by combining the structure mapping residual field.

[0027] In the projection domain, the local projection correction coefficients are interpolated and fused according to the angle and spatial position to form an iterative projection correction field;

[0028] The iterative projection correction field is applied to the low-dose projection data to generate corrected low-dose projection data.

[0029] Furthermore, a wear evolution probability field is generated, including:

[0030] Based on the corrected low-dose projection data and individualized digital twin models of teeth, an evolution-sensitive nucleus was constructed;

[0031] The positive part of the structure mapping residual field is used as the excitation source. The excitation source is mapped to the evolution sensitive kernel through weighted convolution to form the initial probabilistic potential.

[0032] The initial probabilistic potential energy is corrected by angular coupling based on the iterative projection correction field to generate the wear evolution probability field.

[0033] Further, the initial probabilistic potential energy is angularly coupled and corrected based on the iterative projection correction field to generate a wear evolution probability field, including:

[0034] The initial probabilistic potential energy is initially angularly coupled and corrected based on the iterative projection correction field to generate the initial wear evolution probability field.

[0035] The initial wear evolution probability field was forward-projected, and the forward-projection results were compared with the corrected low-dose projection data.

[0036] The iterative projection correction field is iteratively adjusted based on the difference comparison results until the difference comparison results converge, and the current initial wear evolution probability field is taken as the wear evolution probability field.

[0037] Furthermore, based on the candidate reconstruction results, a consistency confidence matrix is ​​obtained through residual decomposition and probability weighting, including:

[0038] For each candidate reconstruction result, a corresponding simulated projection sequence is generated, and the projection residuals of the corrected low-dose projection data are calculated respectively.

[0039] The projection residual is further decomposed based on the structural mapping residual field, and the decomposition result is recalibrated using the wear evolution probability field to obtain the consistency confidence matrix of the voxels.

[0040] Furthermore, the individualized dental digital twin model is globally consistent by combining the structural mapping residual field and the wear evolution probability field to generate a three-dimensional reconstruction result, including:

[0041] The corrected low-dose projection data and the individualized digital twin model of teeth are jointly mapped in the projection domain and volume domain;

[0042] The joint mapping results are redistributed by the structural mapping residual field, and the voxel density of the joint mapping results is adjusted by combining the iterative projection correction field. The joint mapping results are dynamically adjusted according to the wear evolution probability field and the consistency confidence matrix, and the three-dimensional reconstruction results are output.

[0043] According to one aspect of this specification, a three-dimensional oral cavity image reconstruction system based on low-dose projection data is provided, comprising:

[0044] The data acquisition module is used to acquire individualized digital twin models of teeth and low-dose projection data of the target tooth region;

[0045] The structure mapping residual field generation module is used to perform forward projection simulation on the individualized digital twin model of teeth, calculate the residual with the low-dose projection data, and perform reverse projection mapping on the residual result to generate the structure mapping residual field.

[0046] The correction module is used to perform local sampling and backprojection on low-dose projection data based on the structure mapping residual field to form an iterative projection correction field, and to correct the low-dose projection data based on the iterative projection correction field.

[0047] The wear evolution probability field generation module is used to couple the individualized digital twin model of teeth with the corrected low-dose projection data to generate the wear evolution probability field.

[0048] The candidate reconstruction result generation module is used to generate candidate reconstruction results based on the corrected low-dose projection data, the individualized digital twin model of teeth, and the wear evolution probability field.

[0049] The 3D reconstruction result generation module is used to obtain a consistency confidence matrix based on the candidate reconstruction results through residual decomposition and probability weighting, and to perform global consistency optimization on the individualized tooth digital twin model by combining the structure mapping residual field and the wear evolution probability field, so as to generate the 3D reconstruction result.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1. This invention uses a residual field-guided dynamic local sampling mechanism to adaptively focus on key areas. Combined with fast local back-projection, it significantly improves the accuracy of local feature reconstruction while maintaining overall computational efficiency. Furthermore, this scheme introduces an angle-space coupling interpolation strategy, which effectively maintains the data continuity in high-curvature areas such as the occlusal surface of teeth. It solves the problem of artifacts easily generated in narrow areas such as the interproximal surfaces of teeth in traditional methods, thereby improving the clarity of the enamel and dentin boundary reconstruction.

[0052] 2. This invention constructs an evolution-sensitive kernel to quantify the sensitivity of structural differences and utilizes the excitation effect of the residual field and the angle coupling correction mechanism to realize dynamic probabilistic modeling of the wear evolution process of tooth microstructure. This overcomes the problem of insufficient identification accuracy of traditional methods in the wear area of ​​complex occlusal surfaces and provides a reliable basis for wear feature distribution for subsequent three-dimensional reconstruction.

[0053] 3. This invention integrates multi-dimensional data through dual-domain joint mapping and constructs a dynamic optimization framework using residual fields and probability fields. It achieves sub-voxel-level precision reconstruction in complex geometric regions such as cusp bevels and pits and fissures, solving the problem of insufficient spatial matching accuracy between low-dose projection data and digital models, and significantly improving the geometric fidelity of dental hard tissue microstructure reconstruction. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart of a method for reconstructing three-dimensional oral images based on low-dose projection data is provided for an embodiment of the present invention.

[0056] Figure 2This is a schematic diagram of a module of a three-dimensional oral imaging reconstruction system based on low-dose projection data, provided in an embodiment of the present invention. Detailed Implementation

[0057] Specifically, in existing technologies, oral 3D image reconstruction technology relies on traditional filtered back projection or iterative reconstruction algorithms to process low-dose projection data. Although conventional methods can suppress noise and reconstruct the overall tooth morphology, they cannot effectively capture changes in the microstructure of individual tooth surfaces. Especially when dealing with long-term wear or occlusal surface morphology evolution, existing algorithms have difficulty dynamically linking and optimizing individualized tooth models with low-dose projection data, resulting in insufficient restoration of structural details in highly complex areas in the reconstruction results.

[0058] Specifically, to address existing problems, research revealed a gap in the joint optimization of projection data and individualized models, leading to the loss of microstructure features. Analysis showed that the residual distribution between low-dose projection data and the digital twin model contains key structural difference information. Therefore, a dynamic residual mapping mechanism is proposed, which uses iterative correction and probability field fusion to achieve collaborative optimization of multi-source data, thereby improving the accuracy of microstructure reconstruction under low-dose conditions.

[0059] 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.

[0060] Specifically, such as Figure 1As shown, this invention provides a method for reconstructing three-dimensional oral images based on low-dose projection data, comprising: acquiring an individualized digital twin model of a target tooth region and low-dose projection data; performing forward projection simulation on the individualized digital twin model of the tooth, calculating the residual pixel by pixel with the low-dose projection data, performing back projection mapping on it, and generating a structure mapping residual field; performing local sampling and back projection on the low-dose projection data according to the structure mapping residual field to form an iterative projection correction field, and correcting the low-dose projection data according to the iterative projection correction field; coupling the individualized digital twin model of the tooth and the corrected low-dose projection data to generate a wear evolution probability field; generating candidate reconstruction results based on the corrected low-dose projection data, the individualized digital twin model of the tooth, and the wear evolution probability field; forming a consistency confidence matrix by refactoring the residuals and weighting the probability of probability on the candidate reconstruction results, and performing global consistency optimization on the individualized digital twin model of the tooth in combination with the structure mapping residual field and the wear evolution probability field, and outputting the three-dimensional reconstruction results. Among them, the individualized digital twin model of teeth refers to a personalized three-dimensional tooth model established through target historical image data, which can be generated by cone-beam CT scanning combined with surface reconstruction algorithms; the structural mapping residual field refers to the three-dimensional residual distribution formed by back-mapping the difference between forward projection simulation and measured projection, which can be generated by weighted back-projection algorithms; the iterative projection correction field refers to the correction parameter matrix formed by local sampling back-projection guided by the residual field, which can be constructed by Monte Carlo random sampling combined with interpolation algorithms; the wear evolution probability field refers to the probability distribution map reflecting the wear trend of the tooth surface, which can be generated by sensitive kernel convolution and angle coupling correction; the consistency confidence matrix refers to the confidence distribution of candidate reconstruction results at the voxel level, which can be obtained by residual decomposition and probability recalibration calculation.

[0061] Specifically, this invention constructs a dynamic interaction mechanism between a digital twin model and measured data. It achieves quantitative localization of model errors through a structural mapping residual field, completes data-driven local correction using an iterative projection correction field, and realizes coupled modeling of anatomical structure and functional changes by combining a wear evolution probability field. Furthermore, it achieves fine-grained fusion of multiple reconstruction results through residual decomposition and confidence matrices. Through the above technical solutions, this application can effectively improve the reconstruction accuracy of tooth surface microstructures under low-dose conditions, accurately restore the morphological details of the wear area on the occlusal surface, and solve the problem of insufficient capture of individualized features by traditional methods through dynamic interaction optimization between the digital twin model and measured data. Simultaneously, it utilizes probability field modeling to achieve a quantitative expression of tooth morphological evolution trends, providing more accurate three-dimensional data support for clinical diagnosis and treatment.

[0062] Specifically, a forward projection simulation is performed on the individualized digital twin model of teeth, and the residual is calculated pixel by pixel with the low-dose projection data. Then, a backward projection mapping is performed to generate a structure mapping residual field, which includes the following steps:

[0063] A forward projection simulation was performed on an individualized digital twin model of teeth to obtain a simulated projection sequence;

[0064] The residuals between the simulated projection sequence and the low-dose projection data are calculated pixel by pixel to generate a projection residual sequence.

[0065] The projection residual sequence is mapped back to the three-dimensional volume domain through reverse weighted backprojection to form the three-dimensional volume domain residual;

[0066] Based on the individualized digital twin model of teeth, the three-dimensional volume domain residuals are decomposed into multiple scales, and the spatial morphology-preserving filter is applied to the decomposition results at each scale to generate the initial structure mapping residual field.

[0067] The initial structure mapping residual field is verified, and a structure mapping residual field is generated.

[0068] Specifically, forward projection simulation refers to projecting a 3D digital twin model onto a 2D plane at different angles to generate simulated projection data, which can be achieved using X-ray projection algorithms or ray tracing algorithms. The projection residual sequence refers to the set of differences between the simulated projection data and the real low-dose projection data at each pixel, which can be generated through pixel-by-pixel subtraction. Back-weighted back-projection refers to mapping the 2D projection residuals back to 3D space according to weights, which can be achieved using distance-weighted back-projection algorithms. Multi-scale decomposition refers to decomposing the 3D residuals into components with different spatial frequencies, which can be achieved using wavelet transform or Gaussian pyramid algorithms. Spatial morphology-preserving filtering refers to smoothing the decomposed residual components while preserving their edges, which can be achieved using anisotropic diffusion filtering or bilateral filtering algorithms.

[0069] Specifically, a simulated projection sequence with the same angle as the real low-dose projection data is generated through forward projection simulation. The projection residual sequence is calculated by pixel-by-pixel comparison. Then, the two-dimensional residual is converted into a three-dimensional volume domain residual using reverse weighted back projection. The weight allocation can be optimized based on the ray path length or tissue density difference. Next, the three-dimensional residual is decomposed into three scale components, such as high frequency, mid frequency and low frequency. Different morphology-preserving filters with different parameters are used for processing: the high frequency component is treated with weak smoothing intensity to preserve the details of enamel microcracks, the mid frequency component is treated with medium smoothing intensity to optimize dentin texture, and the low frequency component is treated with strong smoothing intensity to remove global artifacts. Finally, the effectiveness of the initial residual field is verified through a verification step. For example, the difference between the residual of the simulated projection and the real data is corrected by iteration to ensure that the residual field can accurately reflect the structural deviation.

[0070] Specifically, by using multi-scale decomposition and morphology-preserving filtering, the high-frequency residuals generated by tooth occlusal surface wear and the low-frequency residuals formed by artifacts in the jawbone region can be effectively distinguished. Iterative correction ensures the dynamic consistency between the residual field and the real projection data. Through the above technical solution, this application can accurately separate tooth microstructure deviations and imaging noise, and suppress back-projection artifacts in the jawbone region while preserving the fine wear features of the enamel surface. This improves the recognition accuracy of the wear evolution region of the occlusal surface in subsequent three-dimensional reconstruction. Through multi-scale residual processing and closed-loop verification mechanism, the problem of detail loss and error accumulation caused by single-scale back-projection in traditional methods is effectively solved.

[0071] Specifically, the initial structure mapping residual field is verified, and a structure mapping residual field is generated, which includes the following steps:

[0072] The simulated projection sequence is corrected based on the residual field mapped from the initial structure;

[0073] The residuals of the corrected simulated projection sequence and the low-dose projection data are calculated pixel by pixel to generate the corrected projection residual sequence, and then the difference between the corrected and low-dose projection residual sequences is processed.

[0074] Determine whether the difference processing result is greater than the preset discrimination threshold. If so, use the initial structure mapping residual field as the structure mapping residual field.

[0075] Otherwise, the initial structure mapping residual field is regenerated based on the corrected projection residual sequence until the preset discrimination threshold is met, and the corresponding initial structure mapping residual field is used as the structure mapping residual field.

[0076] Specifically, the initial structure mapping residual field refers to the three-dimensional volume domain residual distribution after multi-scale decomposition and spatial morphology preservation filtering, which can be implemented using wavelet transform combined with morphological filtering algorithms. Difference processing refers to the calculation of the difference between the two projection residual sequences, which can be implemented using a pixel-by-pixel absolute value difference algorithm. The discrimination threshold is a preset criterion for residual convergence, which can be dynamically set according to the noise level of the projection equipment.

[0077] Specifically, after generating the initial structure mapping residual field, the residual difference is recalculated by correcting the simulated projection sequence. When the difference between the corrected residual sequence and the original residual sequence does not reach the preset threshold, it indicates that the initial residual field has failed to effectively characterize the data difference, and the residual field needs to be regenerated. This process gradually optimizes the accuracy of the residual field through iterative iteration until the change in the corrected residual reaches the convergence criterion, ensuring that the final obtained structure mapping residual field can accurately reflect the true difference distribution between the low-dose projection data and the digital twin model.

[0078] Specifically, by introducing a residual sequence difference comparison and iterative optimization mechanism, residual field distortion caused by projection simulation errors or noise interference can be effectively identified and eliminated, significantly improving the residual field's ability to represent real structural differences. Through the above technical solution, this application solves the problem of inaccurate structural difference representation between low-dose projection data and digital twin models. The iterative verification mechanism ensures that the residual field accurately reflects real structural differences, providing reliable guidance for subsequent projection data correction and three-dimensional reconstruction, and effectively improving the detail reproduction of tooth microstructure reconstruction.

[0079] Specifically, the low-dose projection data is locally sampled and back-projected based on the structure mapping residual field to form an iterative projection correction field. The low-dose projection data is then corrected based on the iterative projection correction field, which includes the following steps:

[0080] Based on the structure mapping residual field, low-dose projection data are randomly sampled in the projection domain to obtain each local subdomain.

[0081] The response distribution of each local subdomain within the volume domain is obtained through fast local backprojection. The optimal local projection correction coefficient is then calculated by combining the structure mapping residual field. The specific calculation formula is as follows:

[0082] (1)

[0083] In the formula, Indicates the first Local projection correction coefficients for each local subdomain, Indicates the first After fast local backprojection, individual voxels within a local subdomain are obtained. The response value, Indicates the first The structure mapping residual field of a local subdomain in a voxel The value at that location, Indicates the scaling factor. Indicates the first A set of voxels in a local subdomain This represents the coefficient used to minimize the error.

[0084] In the projection domain, the local projection correction coefficients are interpolated and fused according to the angle and spatial position to form an iterative projection correction field. The specific calculation formula is as follows:

[0085] (2)

[0086] In the formula, Indicates the iterative projection correction field at the projection angle and the spatial position of the projection plane Correction coefficients on This represents the total number of local subdomains. Indicates the first Weighting factors for each local subdomain, Indicates the first A local subdomain at the projection angle and the spatial position of the projection plane Local projection correction coefficients on.

[0087] The iterative projection correction field is applied to the low-dose projection data to generate corrected low-dose projection data.

[0088] Specifically, the structure-mapped residual field refers to the three-dimensional residual distribution field generated by back-mapping the residuals of forward projection simulation and low-dose projection data. This can be achieved using reverse weighted back-projection combined with multi-scale decomposition. Local sampling refers to dynamically selecting regions in the projection domain based on the residual field weights, which can be implemented using a Monte Carlo random sampling strategy. Fast local back-projection refers to rapid three-dimensional reconstruction of subdomain data, which can be implemented using compressed sensing algorithms or local convolutional neural networks. Local projection correction coefficients are quantification parameters of the degree of matching between subdomain data and the residual field, which can be optimized using least squares or gradient descent methods. Interpolation fusion refers to transforming discrete subdomain correction coefficients into a continuous projection field, which can be achieved using radial basis function interpolation or bicubic spline interpolation.

[0089] Specifically, based on the spatial distribution characteristics of the structure mapping residual field, random sampling is performed on regions with significant residuals in the projection domain. For example, sampling is performed according to the probability density distribution in regions where the residual intensity exceeds a preset threshold, generating multiple local subdomains. Then, fast local backprojection is performed on each subdomain. For example, a GPU-accelerated local filtering backprojection algorithm is used to reconstruct the three-dimensional voxel response corresponding to the subdomain in milliseconds. Combining the residual values ​​at the corresponding positions in the structure mapping residual field, the optimal correction coefficient for each subdomain is calculated through an optimization algorithm. For example, the sum of squares of the residuals of the corrected projection data is minimized. The correction coefficients of each subdomain are spatially interpolated in the projection domain. For example, an anisotropic diffusion algorithm is used to perform bidirectional interpolation in the angular and spatial dimensions to form an iterative projection correction field covering the entire projection domain. Finally, the correction field is superimposed pixel by pixel with the original low-dose projection data to complete the data correction.

[0090] Specifically, the dynamic local sampling mechanism guided by the residual field can adaptively focus on key areas. Combined with fast local backprojection, it significantly reduces computational resource consumption. The introduction of an angle-space coupling interpolation strategy effectively maintains the data continuity of high-curvature areas such as the occlusal surface of teeth. Through the above technical solutions, this application can achieve accurate data correction for complex microstructure areas such as wear areas of the occlusal surface of teeth and root bifurcation sites. While maintaining overall computational efficiency, it significantly improves the accuracy of local feature reconstruction. Through the dynamic generation mechanism of iterative projection correction field, it effectively suppresses the transmission of noise and artifacts in low-dose projection data, providing high-quality input data for subsequent 3D reconstruction. The synergistic effect of local sampling and global interpolation solves the problem of artifacts easily generated in narrow areas such as the interproximal surfaces of teeth in traditional methods, thus improving the clarity of enamel and dentin boundary reconstruction.

[0091] Specifically, the individualized digital twin model of teeth and the corrected low-dose projection data are coupled to generate a wear evolution probability field, which includes the following steps:

[0092] Evolution-sensitive nuclei were constructed based on the revised low-dose projection data and individualized digital twin models of teeth;

[0093] The positive part of the structure-mapped residual field is used as the excitation source and mapped onto the evolution-sensitive kernel through weighted convolution to form an initial probabilistic potential. The specific calculation formula is as follows:

[0094] (3)

[0095] In the formula, This represents the initial probabilistic potential energy, located in the volume or projective domain. The probability intensity, This represents the positive portion of the structure-mapped residual field, located at... The incentive value, The spatial window representing the convolution operation. This represents an evolution-sensitive kernel, defining its location. Position Response weights This represents the weighting coefficient.

[0096] The initial probabilistic potential energy is corrected by angular coupling based on the iterative projection correction field to generate the wear evolution probability field.

[0097] Specifically, the evolution sensitivity kernel refers to a three-dimensional spatial distribution model used to quantify the sensitivity of tooth structure to differences between projected data and digital models. This can be implemented using a gradient field construction method based on voxel intensity differences. The excitation source refers to the portion of the structure mapping residual field that reflects the positive difference between projected and simulated data. This can be implemented using a method for extracting the positive value region of the residual field. Weighted convolution refers to performing spatial domain convolution operations between the excitation source and the evolution sensitivity kernel and introducing weight coefficients. This can be implemented using a dynamic weight adjustment method based on the projection angle distribution. Angle coupling correction refers to optimizing the spatial distribution of the initial probabilistic potential energy using correction coefficients for different projection angles in the iterative projection correction field.

[0098] Specifically, in constructing the evolution-sensitive kernel, the gradient distribution is first calculated by comparing the voxel intensity differences between the corrected low-dose projection data and the individualized dental digital twin model in three-dimensional space, forming a three-dimensional kernel function that reflects the sensitivity to structural differences. Then, the positive value part in the structural mapping residual field is used as the active excitation region. Through the dynamic weight adjustment of the convolution kernel, the intensity distribution of the excitation source is mapped to the spatial position corresponding to the evolution-sensitive kernel, generating the preliminary probabilistic potential energy reflecting the potential wear region. Furthermore, by combining the correction coefficients of each projection angle recorded in the iterative projection correction field, spatial interpolation compensation is performed on the distribution of the preliminary probabilistic potential energy in different projection directions, and finally, a wear evolution probability field with angular consistency is generated.

[0099] Specifically, by constructing an evolution-sensitive kernel to quantify structural difference sensitivity and utilizing the excitation effect of the residual field and the angle coupling correction mechanism, dynamic probabilistic modeling of the wear evolution process of tooth microstructures is achieved, overcoming the problem of insufficient identification accuracy of traditional methods in complex occlusal surface wear areas. Through the above technical solution, this application can effectively integrate the complementary information of low-dose projection data and individualized digital models to accurately capture the wear probability distribution of microstructure areas such as occlusal surfaces and proximal surfaces of teeth, solving the technical problem of inaccurate modeling of the dynamic wear evolution process of teeth in existing technologies, and providing a reliable basis for wear feature distribution for subsequent three-dimensional reconstruction.

[0100] Specifically, the initial probabilistic potential energy is angularly coupled and corrected based on the iterative projection correction field to generate the wear evolution probability field, which includes the following steps:

[0101] The initial probabilistic potential energy is angularly coupled and corrected based on the iterative projection correction field to generate the initial wear evolution probability field.

[0102] The initial wear evolution probability field was forward-projected, and the forward-projection results were compared with the corrected low-dose projection data.

[0103] The iterative projection correction field is adjusted iteratively based on the difference comparison results until the difference comparison results converge, and the initial wear evolution probability field at this point is taken as the wear evolution probability field.

[0104] Specifically, angle coupling correction refers to spatially calibrating the probability distribution using projection angle information, which can be achieved using angle-related spatial interpolation methods. The initial wear evolution probability field refers to the initial three-dimensional probability distribution generated based on structural residuals and the projection correction field, which can be achieved using probability density function mapping methods. Difference comparison refers to verifying the accuracy of the probability field by comparing the similarity between the forward projection data and the actual collected data, which can be achieved using pixel-level mean square error calculation methods. Iterative adjustment refers to dynamically optimizing the projection correction parameters through a feedback mechanism, which can be achieved using gradient descent optimization algorithms.

[0105] Specifically, after generating the initial wear evolution probability field, it is converted into a two-dimensional projection sequence through a forward projection operation. The projection sequence is then compared with the corrected low-dose projection data pixel by pixel to generate a difference distribution map. When the mean square error of the difference distribution map exceeds a preset threshold, the angle coupling parameters in the iterative projection correction field are adjusted in reverse according to the difference distribution. The adjusted correction field is then applied to the initial probabilistic potential energy again to generate an updated wear evolution probability field. This process is repeated until the difference comparison results reach the convergence condition. The final output probability field accurately reflects the spatial distribution characteristics of the tooth wear area.

[0106] Specifically, by introducing an angle coupling correction and iterative optimization mechanism, dynamic matching of probability field and multi-angle projection data is achieved. Through the above technical solution, this application effectively solves the problem of angle sensitivity deviation when coupling low-dose projection data with digital twin model, and significantly improves the spatial accuracy of wear evolution probability field. This method can automatically identify and correct systematic errors in probability distribution, ensuring that the three-dimensional representation of tooth wear area in the final reconstruction result is highly consistent with the real anatomical structure.

[0107] Specifically, candidate reconstruction results are generated based on the corrected low-dose projection data, individualized digital twin models of teeth, and wear evolution probability fields, including the following steps:

[0108] Three candidate reconstruction results are generated, including the candidate reconstruction result obtained by backprojecting the corrected low-dose projection data, the candidate reconstruction result obtained by using an individualized digital twin model of teeth as a prior, and the candidate reconstruction result obtained by weighting the wear evolution probability field.

[0109] Specifically, candidate reconstruction results refer to the preliminary reconstruction results of the 3D structure generated through different data sources or methods. This can be achieved by using a filtered back-projection algorithm to back-project the corrected low-dose projection data. Candidate reconstruction results obtained from a priori personalized dental digital twin model refer to generating candidate results using a model-driven reconstruction algorithm with pre-established personalized 3D dental models as constraints. This can be achieved using a registration and fusion method based on morphological similarity. Candidate reconstruction results obtained by weighting the wear evolution probability field refer to generating candidate results by dynamically adjusting the weights of 3D voxels based on the wear probability distribution in different regions of the probability field. This can be achieved using a voxel weighting algorithm based on a probability density function.

[0110] Specifically, the corrected low-dose projection data is back-projected to generate candidate reconstruction results, which preserve the geometric features of the original projection data; the individualized digital twin model of the teeth is matched with the current projection data through a non-rigid registration algorithm to generate candidate reconstruction results, which incorporate the personalized anatomical features of the patient's teeth; the wear evolution probability field dynamically weights the three-dimensional voxels through a spatial probability distribution function to generate candidate reconstruction results, which reflect the potential wear patterns of the tooth structure; the three candidate results are constructed from three dimensions: data-driven, model-driven, and probability-driven, respectively, forming a complementary reconstruction perspective.

[0111] Specifically, by using a multi-source candidate result generation mechanism, the spatial constraints of individualized models and the temporal dynamic characteristics of wear evolution are integrated while preserving the geometric accuracy of the original data, effectively overcoming the limitations of single reconstruction methods. Through the above technical solution, this application can improve the reconstruction accuracy of the microstructure of the occlusal surface of teeth and the enamel margin, reduce the probability of artifact generation caused by projection data noise or model deviation, and enhance the characterization ability of the long-term wear evolution process of teeth through the synergistic effect of multi-dimensional candidate results, providing more comprehensive three-dimensional anatomical information support for clinical practice.

[0112] Specifically, the candidate reconstruction results are processed by residual decomposition and probability weighting to form a consistency confidence matrix, which includes the following steps:

[0113] For each candidate reconstruction result, generate its corresponding simulated projection sequence, and calculate its projection residual with the corrected low-dose projection data respectively.

[0114] The projection residuals are further decomposed according to the structure mapping residual field, and the decomposition results are recalibrated using the wear evolution probability field to obtain the consistency confidence matrix of the voxels.

[0115] Specifically, residual decomposition refers to decomposing the projected residuals according to the multi-scale characteristics of the structure-mapped residual field, which can be achieved using wavelet transform or frequency domain filtering methods. Probability weighting refers to dynamically assigning weights to the decomposed residual components using the wear evolution probability field, which can be achieved by combining convolution operations with probability density functions. The consistency confidence matrix refers to the voxel-level confidence distribution that integrates the residual decomposition and probability weighting, which can be generated using tensor fusion methods.

[0116] Specifically, after generating candidate reconstruction results, a forward projection simulation is performed for each candidate result to obtain the corresponding simulated projection sequence. The simulated projection sequence is then compared pixel by pixel with the corrected low-dose projection data to generate projection residuals. Subsequently, based on the multi-scale residual distribution pattern recorded in the structure mapping residual field, the projection residuals are spatially decomposed, for example, the residuals are decomposed into high-frequency detail components and low-frequency structural components. The decomposed residual components are recalibrated through the wear evolution probability field, for example, applying higher weights to the residual components of high-probability wear regions and suppressing low-probability regions. Finally, a voxel-level consistency confidence matrix is ​​generated through weighted fusion. This matrix can quantify the confidence distribution of different candidate reconstruction results in three-dimensional space.

[0117] Specifically, through the dual optimization of residual decomposition and probability weighting, the weight allocation can be dynamically adjusted according to the multi-scale characteristics of the structure mapping residual field. At the same time, the confidence of highly sensitive areas is enhanced by combining the wear evolution probability field, thereby achieving more accurate reconstruction result fusion in microstructural areas such as the occlusal surface and enamel margin. Through the above technical solution, this application can effectively solve the problem of ignoring the correlation between local structural features and wear evolution when fusing candidate reconstruction results in the prior art. By establishing a voxel-level consistency confidence matrix, the accuracy of tooth microstructure reconstruction is significantly improved. In particular, in the optimization of density distribution in complex geometric areas such as cusp bevels and proximal surfaces, the problem of detail blurring or artifacts caused by traditional methods can be avoided.

[0118] Specifically, the individualized digital twin model of teeth is globally consistent by combining the structural mapping residual field and the wear evolution probability field to output the three-dimensional reconstruction result. The specific steps include:

[0119] The corrected low-dose projection data and the individualized digital twin model of teeth are jointly mapped in the projection domain and volume domain;

[0120] The joint mapping results are redistributed by the structural mapping residual field, and the voxel density of the joint mapping results is adjusted by combining the iterative projection correction field. The joint mapping results are dynamically adjusted according to the wear evolution probability field and the consistency confidence matrix, and the three-dimensional reconstruction results are output.

[0121] Specifically, joint mapping refers to the bidirectional matching of the correspondence between two-dimensional projection data and three-dimensional digital models in the projection plane and the three-dimensional spatial coordinate system. This can be achieved using projection matrix transformation and voxel space interpolation algorithms. The structure mapping residual field refers to the three-dimensional error distribution field formed by forward projection simulation and reverse residual mapping. This can be generated using multi-scale decomposition and shape-preserving filtering algorithms. Dynamic adjustment refers to the real-time weight adjustment of three-dimensional voxels based on the probability field and confidence matrix. This can be achieved using an adaptive optimization algorithm based on the probability density function.

[0122] Specifically, in the projection domain, the corrected low-dose projection data is mapped to the three-dimensional voxel coordinate system through geometric transformation. At the same time, in the volume domain, the spatial topology of the digital twin model is converted into projection plane parameters. During the joint mapping process, the structural mapping residual field identifies high-error regions through spatial gradient analysis, guiding the voxel density to be locally enhanced on the occlusal surface of the crown or the bifurcation area of ​​the root. The iterative projection correction field acts on the projection domain to nonlinearly compensate for the ray attenuation coefficient at the enamel-dentin junction. The wear evolution probability field and the consistency confidence matrix jointly construct a dynamic optimization function, and iterative voxel intensity correction is implemented in the cusp wear area and the interdental space, ultimately achieving global consistency reconstruction of the three-dimensional structure.

[0123] Specifically, by integrating multi-dimensional data through dual-domain joint mapping and constructing a dynamic optimization framework using residual and probability fields, sub-voxel-level precision reconstruction is achieved in complex geometric regions such as cusp bevels and pits and fissures. Through the above technical solution, this application solves the problem of insufficient spatial matching accuracy between low-dose projection data and digital models, significantly improving the geometric fidelity of dental hard tissue microstructure reconstruction. In particular, it effectively suppresses step-like artifacts in the occlusal wear area and root canal morphology reconstruction, ensuring the clinical usability of three-dimensional images in orthodontic force simulation and implant positioning applications.

[0124] The implementation of the various embodiments of the present invention is based on programmed processing by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a three-dimensional oral imaging reconstruction system based on low-dose projection data. This system is used to execute a three-dimensional oral imaging reconstruction method based on low-dose projection data from the above method embodiments.

[0125] like Figure 2As shown, the system includes: a data acquisition module for acquiring an individualized digital twin model of the target tooth region and low-dose projection data; a structure mapping residual field generation module for performing forward projection simulation on the individualized digital twin model of the tooth, calculating residuals with the low-dose projection data, performing back projection mapping on the residual results, and generating a structure mapping residual field; a correction module for performing local sampling and back projection on the low-dose projection data according to the structure mapping residual field to form an iterative projection correction field, and correcting the low-dose projection data according to the iterative projection correction field; a wear evolution probability field generation module for coupling the individualized digital twin model of the tooth and the corrected low-dose projection data to generate a wear evolution probability field; a candidate reconstruction result generation module for generating candidate reconstruction results based on the corrected low-dose projection data, the individualized digital twin model of the tooth, and the wear evolution probability field; and a three-dimensional reconstruction result generation module for obtaining a consistency confidence matrix by residual re-decomposition and probability weighting based on the candidate reconstruction results, and performing global consistency optimization on the individualized digital twin model of the tooth in combination with the structure mapping residual field and the wear evolution probability field to generate a three-dimensional reconstruction result.

[0126] The oral cavity three-dimensional image reconstruction system based on low-dose projection data provided in this invention addresses the problem that existing algorithms struggle to dynamically correlate and optimize individualized tooth models with low-dose projection data, resulting in insufficient reconstruction of structural details in highly complex regions. This system employs several modules to construct a dynamic interaction mechanism between the digital twin model and measured data. It achieves quantitative localization of model errors through a structural mapping residual field, completes data-driven local correction using an iterative projection correction field, and combines a wear evolution probability field to achieve coupled modeling of anatomical structure and functional changes. Finally, it achieves fine-grained fusion of multiple reconstruction results through residual decomposition and confidence matrices.

[0127] Finally, it should be noted that the above specific embodiments are merely illustrative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and many variations are possible. Any simple modifications, equivalent changes, and alterations made to the above specific embodiments based on the technical essence of the present invention should be considered within the protection scope of the present invention.

Claims

1. A method for reconstructing three-dimensional oral images based on low-dose projection data, characterized in that, include: Acquire individualized digital twin models and low-dose projection data of the target tooth region; A forward projection simulation is performed on the individualized digital twin model of teeth, and the residual is calculated with the low-dose projection data. The residual results are then back-projected and mapped to generate a structure-mapping residual field. Based on the structure mapping residual field, the low-dose projection data is locally sampled and back-projected to form an iterative projection correction field, and the low-dose projection data is corrected based on the iterative projection correction field. A wear evolution probability field is generated by coupling a personalized digital twin model of teeth with corrected low-dose projection data. Generating a wear evolution probability field includes: constructing an evolution-sensitive kernel based on corrected low-dose projection data and a personalized digital twin model of teeth; using the positive value portion of the structure mapping residual field as an excitation source, mapping the excitation source to the evolution-sensitive kernel through weighted convolution to form a preliminary probabilistic potential energy; and performing angular coupling correction on the preliminary probabilistic potential energy based on the iterative projection correction field to generate the wear evolution probability field; wherein, performing angular coupling correction on the preliminary probabilistic potential energy based on the iterative projection correction field to generate the wear evolution probability field includes: The initial probabilistic potential energy is angularly coupled and corrected based on the iterative projection correction field to generate the initial wear evolution probability field. The initial wear evolution probability field was forward-projected, and the forward-projection results were compared with the corrected low-dose projection data. The iterative projection correction field is iteratively adjusted according to the difference comparison results until the difference comparison results converge, and the current initial wear evolution probability field is taken as the wear evolution probability field. Candidate reconstruction results are generated based on the corrected low-dose projection data, individualized digital twin models of teeth, and wear evolution probability fields. Based on the candidate reconstruction results, a consistency confidence matrix is ​​obtained through residual decomposition and probability weighting. This matrix is ​​then combined with the structure mapping residual field and the wear evolution probability field to perform global consistency optimization on the individualized dental digital twin model, generating a three-dimensional reconstruction result. The consistency confidence matrix obtained through residual decomposition and probability weighting based on the candidate reconstruction results includes: For each candidate reconstruction result, a corresponding simulated projection sequence is generated, and the projection residuals of the corrected low-dose projection data are calculated respectively. The projection residual is further decomposed based on the structural mapping residual field, and the decomposition result is recalibrated using the wear evolution probability field to obtain the consistency confidence matrix of the voxels.

2. The method for reconstructing three-dimensional oral images based on low-dose projection data according to claim 1, characterized in that, Generate the structure-mapped residual field, including: Perform forward projection simulation on individualized digital twin models of teeth to generate simulated projection sequences; The residuals between the simulated projection sequence and the low-dose projection data are calculated pixel by pixel to generate a projection residual sequence. The projected residual sequence is mapped back to the three-dimensional volume domain by reverse weighted back projection, thus generating the three-dimensional volume domain residual. Based on the individualized digital twin model of teeth, the three-dimensional volume domain residuals are decomposed into multiple scales, and the spatial morphology-preserving filter is applied to the decomposition results at each scale to generate the initial structure mapping residual field. The initial structure mapping residual field is verified, and a structure mapping residual field is generated.

3. The method for reconstructing three-dimensional oral images based on low-dose projection data according to claim 2, characterized in that, The initial structure mapping residual field is validated, and a structure mapping residual field is generated, including: The simulated projection sequence is corrected based on the residual field mapped from the initial structure. The residuals of the corrected simulated projection sequence and the low-dose projection data are calculated pixel by pixel to generate the corrected projection residual sequence, and then the difference between the corrected projection residual sequence and the projection residual sequence is processed. Determine the difference processing result and the preset discrimination threshold: When the difference processing result is greater than the preset discrimination threshold, use the initial structure mapping residual field as the structure mapping residual field.

4. The method for reconstructing three-dimensional oral images based on low-dose projection data according to claim 3, characterized in that, The determination of the difference processing result and the preset discrimination threshold also includes: When the difference processing result is not greater than the preset discrimination threshold, the initial structure mapping residual field is regenerated according to the corrected projection residual sequence until the preset discrimination threshold is met, and the corresponding initial structure mapping residual field is used as the structure mapping residual field.

5. The method for reconstructing three-dimensional oral images based on low-dose projection data according to claim 1, characterized in that, The low-dose projection data is corrected based on the iterative projection correction field, including: Based on the structure mapping residual field, low-dose projection data are randomly sampled in the projection domain to obtain each local subdomain. The response distribution of each local subdomain in the volume domain is obtained by fast local backprojection, and the optimal local projection correction coefficient is obtained by combining the structure mapping residual field. In the projection domain, the local projection correction coefficients are interpolated and fused according to the angle and spatial position to form an iterative projection correction field; The iterative projection correction field is applied to the low-dose projection data to generate corrected low-dose projection data.

6. The method for reconstructing three-dimensional oral images based on low-dose projection data according to claim 1, characterized in that, The individualized digital twin model of teeth is globally consistent by combining the structural mapping residual field and the wear evolution probability field to generate a three-dimensional reconstruction result, including: The corrected low-dose projection data and the individualized digital twin model of teeth are jointly mapped in the projection domain and volume domain; The joint mapping results are redistributed by the structural mapping residual field, and the voxel density of the joint mapping results is adjusted by combining the iterative projection correction field. The joint mapping results are dynamically adjusted according to the wear evolution probability field and the consistency confidence matrix, and the three-dimensional reconstruction results are output.

7. A three-dimensional oral imaging reconstruction system based on low-dose projection data, characterized in that, A method for reconstructing oral cavity three-dimensional images based on low-dose projection data according to any one of claims 1-6 includes: The data acquisition module is used to acquire individualized digital twin models of teeth and low-dose projection data of the target tooth region; The structure mapping residual field generation module is used to perform forward projection simulation on the individualized digital twin model of teeth, calculate the residual with the low-dose projection data, and perform reverse projection mapping on the residual result to generate the structure mapping residual field. The correction module is used to perform local sampling and backprojection on low-dose projection data based on the structure mapping residual field to form an iterative projection correction field, and to correct the low-dose projection data based on the iterative projection correction field. The wear evolution probability field generation module is used to couple the individualized digital twin model of teeth with the corrected low-dose projection data to generate the wear evolution probability field. The candidate reconstruction result generation module is used to generate candidate reconstruction results based on the corrected low-dose projection data, the individualized digital twin model of teeth, and the wear evolution probability field. The 3D reconstruction result generation module is used to obtain a consistency confidence matrix based on the candidate reconstruction results through residual decomposition and probability weighting, and to perform global consistency optimization on the individualized tooth digital twin model by combining the structure mapping residual field and the wear evolution probability field, so as to generate the 3D reconstruction result.

Images