A method, device, system and storage medium for evaluating the edge fit of a fixed denture

Abstract

The application relates to the technical field of oral cavity evaluation, and discloses a fixed denture edge tightness evaluation method, device, system and storage medium.The fixed denture edge tightness evaluation method comprises the following steps: S1, acquiring multi-modal image data of a contact area of a denture and an abutment; S2, performing image denoising on the multi-modal image data; S3, performing edge detection and gap analysis on the image data after image denoising processing to obtain a gap distribution diagram; S4, performing evaluation on the gap distribution diagram according to a scoring standard to obtain a tightness distribution diagram; and S5, superimposing the tightness distribution diagram into a real environment through augmented reality technology.The application realizes high-precision and high-efficiency fixed denture edge tightness evaluation through the fusion of multi-dimensional data, a dynamic scoring model and augmented reality interaction technology, has the functions of multi-material dynamic adaptation and complex environment anti-interference capability, and provides three-dimensional quantitative decision support for clinics.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oral cavity evaluation, in particular to a fixed denture edge tightness evaluation method, device, system and storage medium. BACKGROUND

[0002] As the core treatment means of oral cavity repair, the edge tightness of fixed denture directly determines the long-term stability of the prosthesis and the health of the patient. With the wide application of high-precision repair materials (such as zirconia ceramic and composite resin) and the increase of complex cases (such as multi-unit bridge and implant-supported repair), the traditional evaluation method faces serious technical bottlenecks in three-dimensional gap quantification, material adaptability and dynamic environment anti-interference. The oral wet environment, tissue deformation characteristics and biomechanical differences of multi-material interface further require the evaluation technology to have multi-dimensional data fusion and intelligent analysis capability to accurately guide the prosthesis trimming and clinical decision-making.

[0003] In current clinical practice, the evaluation of fixed denture edge tightness mainly relies on visual inspection, probe palpation and dye leakage method. Some improved technologies combine two-dimensional optical imaging (such as oral endoscope) or X-ray tomography to measure the projected gap of the contact area between the prosthesis and the abutment by manual measurement. Advanced schemes try to introduce three-dimensional scanners to obtain prosthesis surface topography data, or use ultrasonic flaw detection technology to detect interface micro-gaps, but data processing still relies on manual threshold setting and experience interpretation.

[0004] The existing technology also has some shortcomings; visual and dyeing methods are limited by human eye resolution and liquid surface tension effect, and cannot quantify the distribution characteristics of micron-level gaps; two-dimensional imaging technology causes systematic underestimation of the curvature change of three-dimensional contact surface and the gap in hidden areas due to projection distortion and data dimension reduction; single-mode detection methods (such as pure optical or pure mechanical) lack comprehensive analysis capability for multi-physical field coupling effects such as material thermal expansion and elastic deformation, resulting in deviation between the evaluation results and the real biomechanical state. In addition, the existing methods have insufficient signal stability in dynamic wet environment, and interference sources such as saliva flow and metal reflection can easily cause data acquisition distortion, while the time-consuming nature and subjectivity of manual interpretation process further limit the clinical evaluation efficiency and repeatability. SUMMARY

[0005] In view of the shortcomings of the prior art, the present application provides a fixed denture edge tightness evaluation method, device, system and storage medium, which solves the problems of strong subjectivity, low precision, poor efficiency and insufficient material adaptability of the prior art in evaluating fixed denture edge tightness in complex oral environment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for evaluating the marginal fit of a fixed denture, characterized by comprising the following steps:

[0007] S1. Acquire multimodal image data of the contact area between the denture and the abutment tooth;

[0008] S2. Perform image denoising on the multimodal image data;

[0009] S3. Perform edge detection and gap analysis on the image data after image denoising to obtain a gap distribution map;

[0010] S4. Evaluate the gap distribution map using the scoring criteria to obtain the fit distribution map; three levels of evaluation intervals can be established: perfect fit <50μm, minor gap 50–100μm is clinically acceptable, and >100μm requires re-making;

[0011] S5. Overlay the conformity distribution map onto the real environment using augmented reality technology.

[0012] S1 includes:

[0013] S11. Acquire two-dimensional high-resolution images of the contact area between the denture and the abutment tooth using a high-definition optical probe;

[0014] S12. Collect three-dimensional depth data of the contact area between the denture and the abutment tooth using a depth camera;

[0015] S13. Use an ultrasonic sensor to collect minute cracks or irregular surfaces in the contact area between the denture and the abutment tooth.

[0016] S14. Enhancement processing of two-dimensional high-resolution images, three-dimensional depth data, and micro-cracks or irregular surfaces in saliva and blood environments using infrared scanning sensors.

[0017] S15. Multimodal image data is obtained by fusing and enhancing two-dimensional high-resolution images, three-dimensional depth data, and micro-cracks or irregular surfaces through a cross-modal feature pyramid network.

[0018] S2 includes:

[0019] S21. A generative adversarial network is trained using the cycle consistency algorithm. The generative adversarial network includes a generator network for generating high-quality images that conform to the actual situation of dental prostheses and a discriminator network for judging whether the images are real.

[0020] S22. Denoising the multimodal data using a trained generative adversarial network.

[0021] S3 includes:

[0022] S31. Use the UNet model to perform edge detection on the denoised multimodal data;

[0023] S32. Based on the results of edge detection, analyze whether there are overhangs, missing parts or irregular shapes at the edges of the restoration;

[0024] S33. Using the variance analysis algorithm, the distribution of edge gaps in the contact area is statistically analyzed to obtain a gap distribution map.

[0025] S4 includes:

[0026] A quantitative scoring model is constructed, which calculates the comprehensive score of fit using the following formula:

[0027]

[0028] Where, d i w represents the gap value at the i-th measurement point. i is the weighting coefficient for the corresponding region; CAS% is the contact area percentage; β is the material attenuation coefficient; α is the gap weighting factor; γ is the contact area compensation coefficient; n is the total number of valid measurement points.

[0029] S5 includes:

[0030] The specific steps to achieve 3D spatial overlay using AR glasses, smart tablets, or mobile phones are as follows:

[0031] Establish a unified coordinate system using the cross-platform OpenXR framework

[0032] Equipment Difference Handling:

[0033] AR glasses: Construct a point cloud map of the oral environment using SLAM technology;

[0034] Smart tablets and mobile phones: An improved PnP algorithm is used to solve the camera pose.

[0035] Coordinate transformation matrix:

[0036] T device =R·T scan +t;

[0037] Where R is the rotation matrix; t is the translation vector; T scan For the scanning system coordinate system; T device For AR device coordinates;

[0038] Dynamic rendering: Rendering gap heatmaps using the Vulkan engine;

[0039] Real-time data synchronization:

[0040] Transmission protocol: WebRTC low-latency transmission;

[0041] Data compression: Lightweight feature encoding based on EfficientNet-Lite;

[0042] Security check: SHA-256 hash check.

[0043] The present invention also provides an evaluation device for the marginal fit of fixed dentures, comprising:

[0044] The first processing module is used to acquire and enhance two-dimensional high-resolution images, three-dimensional depth data, and data of micro-cracks or irregular surfaces in saliva and blood environments through high-definition optical probes, depth cameras, ultrasonic sensors, and infrared scanning sensors, and to fuse the acquired data into multimodal image data through a cross-modal feature pyramid network.

[0045] The second processing module is used to perform image denoising on the multimodal image data;

[0046] The third processing module is used to perform edge detection and gap analysis on the image data after image denoising to obtain a gap distribution map.

[0047] The fourth processing module is used to evaluate the gap distribution map according to the scoring criteria to obtain the fabric tightness distribution map;

[0048] The fifth processing module is used to overlay the conformity distribution map onto the real environment using augmented reality technology.

[0049] The present invention also provides a system for evaluating the marginal fit of fixed dentures, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the above-described method for evaluating the marginal fit of fixed dentures is performed.

[0050] The present invention also provides a storage medium storing a computer program that, when running, executes the above-described method for evaluating the marginal fit of a fixed denture.

[0051] This invention provides a method, apparatus, system, and storage medium for evaluating the marginal fit of fixed dentures. It offers the following advantages:

[0052] 1. This invention integrates 3D scanning, depth imaging, and infrared scanning technologies to simultaneously capture the surface morphology, edge gaps, and thermal expansion characteristics of the restoration, overcoming the limitations of traditional single-modal detection. The multimodal data complementarity mechanism effectively covers visible light blind spots and scenarios lacking depth information, significantly improving the dimensional integrity and spatial resolution of edge fit assessment.

[0053] 2. This invention employs a deep learning-driven edge detection algorithm and a dynamic scoring model to transform clinical experience into quantifiable evaluation standards, avoiding the risk of misjudgment caused by differences in operator experience during manual visual inspection. The automated process achieves end-to-end standardization from data collection to result output, ensuring consistency of evaluation results across different medical institutions.

[0054] 3. This invention establishes a dynamically adjustable quantitative scoring system by introducing biomechanical parameters such as material elastic modulus and surface roughness, addressing the limitations of traditional fixed threshold methods in evaluating heterogeneous materials such as metal, all-ceramic, and resin restorations. The model uses a nonlinear weighting mechanism to accurately match the clinical fit requirements of different materials.

[0055] 4. This invention utilizes AR overlay display technology based on cross-device spatial registration to fuse a heat map of gap distribution with the real oral environment in real time, assisting dentists in intuitively locating high-risk areas. The interactive visualization solution supports gesture control and multi-view observation, significantly improving the positioning accuracy and decision-making efficiency of prosthesis adjustment operations.

[0056] 5. This invention effectively suppresses the impact of common intraoperative interference sources such as saliva, blood, and metallic reflections on data quality by combining generative adversarial network image denoising with wavelet threshold shrinking ultrasound signal processing. The dual denoising mechanism, through time-frequency domain joint optimization, ensures stable output of highly reliable evaluation results even in complex oral environments. Attached Figure Description

[0057] Figure 1 This is one of the schematic diagrams of the method flow of the present invention;

[0058] Figure 2 This is a second schematic diagram of the method flow of the present invention;

[0059] Figure 3 This is the third schematic diagram of the method flow of the present invention;

[0060] Figure 4 This is the fourth schematic diagram of the method flow of the present invention;

[0061] Figure 5 This is a diagram of the device architecture of the present invention;

[0062] Figure 6 This is a system structure diagram of the present invention;

[0063] Figure 7 This is a control diagram of the storage medium of the present invention. Detailed Implementation

[0064] The technical solutions in 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.

[0065] Please see the appendix Figure 1 - Appendix Figure 4 This invention provides a method for evaluating the marginal fit of fixed dentures, comprising the following steps:

[0066] S1. Acquire multimodal image data of the contact area between the denture and the abutment tooth;

[0067] In this embodiment, the acquisition of multimodal image data of the contact area between the denture and the abutment tooth is achieved through multi-sensor collaborative acquisition and cross-modal data fusion, specifically including the following technical contents:

[0068] Multi-angle optical imaging of the contact area between the restoration and the abutment tooth is performed using a high-resolution optical probe. Preferably, the optical probe employs a ring array light source layout, emitting visible light signals and receiving reflected light signals to form a two-dimensional high-resolution image.

[0069] During optical imaging, polarization filters suppress interference from reflected light on the surface of the metal restoration, and dynamic exposure adjustment technology is employed to adapt to the differences in optical properties of different materials. For transparent or translucent restoration materials, the contrast of the edge areas is enhanced by adjusting the incident light angle and intensity.

[0070] A depth sensing device based on the time-of-flight principle is used to transmit modulated light signals to the target area and receive reflected signals. By calculating the phase delay of the light signals, point cloud data containing three-dimensional spatial coordinates is generated.

[0071] Point cloud data is processed by a surface reconstruction algorithm and transformed into a continuous three-dimensional mesh model, accurately representing the three-dimensional geometry of the contact surface between the restoration and the abutment tooth. Preferably, the surface reconstruction algorithm employs the Poisson reconstruction method, which solves for implicit functions to fit the point cloud distribution, preserving sub-millimeter level details.

[0072] A high-frequency ultrasonic sensor is deployed to emit pulsed sound waves into the contact area and receive the echo signals. By analyzing the time-domain characteristics and frequency-domain attenuation properties of the echo signals, microscopic cracks and irregularities at the edges of the restoration are detected.

[0073] Preferably, the ultrasonic signal processing uses Hilbert transform to extract the envelope, and combines it with the time-of-flight ranging formula to calculate the defect depth.

[0074]

[0075] Where c is the speed of sound propagation in the oral cavity medium; Δt is the time difference between transmitting and receiving signals; and d is the depth of the defect.

[0076] In a moist environment containing saliva or blood, a near-infrared light source is used to illuminate the target area, utilizing the penetrating properties of infrared light to suppress liquid interference. An infrared sensor collects data on the distribution of reflected light intensity to generate an enhanced surface texture image.

[0077] Preferably, the operating wavelength of the infrared sensor is selected within a range where the absorption rate of biological tissue is low, reducing the masking effect of background noise on the effective signal. For blood-covered areas, hemoglobin absorption characteristics are separated through multispectral analysis to improve the discernibility of edge information.

[0078] High-definition optical probes capture images from multiple angles to ensure accurate capture of contact surface details. Two-dimensional images provide information about the contact surface between the prosthesis and the abutment tooth, especially in the absence of significant marginal protrusions or irregularities, accurately demonstrating the fit of the prosthesis. A depth camera captures three-dimensional data, supplementing the spatial information that two-dimensional images cannot represent. This allows the system to more accurately identify the three-dimensional structure of the prosthesis and abutment tooth contact surface, particularly in complex cases or those with irregular edges, clearly showing the three-dimensional spatial relationship of the contact area. Depth data allows for the calculation of spatial distances between different points, further determining the gap in the contact area. Infrared scanning sensors capture details through temperature differences, especially in the presence of interfering substances such as saliva and blood, avoiding image distortion or data loss.

[0079] Deep fusion of multi-source data is achieved through a cross-modal feature pyramid network (CM-FPN), specifically including the following processing steps:

[0080] Optical images are processed by convolutional neural networks to extract texture features and capture grayscale gradients and morphological details in edge regions.

[0081] Three-dimensional point cloud data is used to extract spatial geometric features through graph convolutional networks to quantify the curvature changes and gap distribution of the contact surface;

[0082] Defect features and anti-interference texture features were extracted from ultrasonic and infrared data through one-dimensional convolution and frequency domain transformation, respectively.

[0083] The Scale Invariant Feature Transform (SIFT) algorithm is used to achieve sub-pixel level registration of multimodal data, eliminating spatial offset caused by sensor position differences.

[0084] By progressively upsampling and weighting the feature maps of the pyramid structure, feature information from different modalities is fused to generate a multimodal fusion image with consistent resolution.

[0085] During the fusion process, the original data is enhanced as follows:

[0086] The training dataset is expanded by random affine transformations (translation, rotation, scaling) to improve the model's robustness to clinical operational biases.

[0087] The dynamic range of optical and infrared images is optimized by employing histogram equalization and contrast-limited adaptive histogram equalization (CLAHE) algorithms.

[0088] Wavelet threshold denoising is applied to the ultrasonic signal to suppress the interference of environmental noise on defect detection.

[0089] S2. Denoise the multimodal image data;

[0090] Image denoising of multimodal image data is achieved through generative adversarial networks, specifically including the following techniques:

[0091] Generative Adversarial Networks (GANs) consist of a generator network and a discriminator network, achieving noise suppression and preservation of realistic features through adversarial training strategies. Preferably, the generator adopts an encoder-decoder structure, where the encoder extracts deep features from multimodal data through convolutional layers, and the decoder reconstructs the denoised image through transposed convolutional layers.

[0092] The discriminator network employs a fully convolutional architecture, receiving the generator's output image and a real, noise-free image, and outputting the authenticity judgment result for local regions. Preferably, the discriminator uses a sliding window mechanism to independently judge the authenticity of each sub-region in the image, improving the ability to identify local details.

[0093] The process of training a Generative Adversarial Network (GAN) includes:

[0094] The input raw image is matched and learned with the corresponding correctly labeled image, and then optimized. To achieve the recognition of S and F(s), where F is the forward generator; S is the labeled image domain discriminator; S is the original image domain; Label the map area; For mathematical expectation operators; s ~ p data (s) represents the original image sample s from the real data distribution p. data Sampling in (s); t~p data (t) represents the labeled sample t from the real data distribution p. data Sampling in (t).

[0095] The generator G network extracts the target region from the correctly labeled image. It then calculates the average pixel value within the same region of the input target image to constrain the reverse learning process. The G network parameters are continuously optimized during the process of learning to generate the input target image from the correctly labeled image. This is used to identify the correctly labeled image and the generated original input image, where G is the inverse generator; This is the original image domain discriminator.

[0096] Different synthetic target maps can be generated by arbitrarily changing one or more of the correctly labeled map, intra-class mean, and noise in the trained G network.

[0097] Step S2 further includes: constraining by optimizing a cyclic loss, where the cyclic loss is:

[0098]

[0099] Where, p data Let G(F(s)) represent the probability distribution of the given data; G(F(s)) is the original image s generated by the forward generator F to generate the labeled image F(s), and then reconstructed by the inverse generator G to obtain the image G(F(s))∈S; F(G(t)) is the labeled image t generated by the inverse generator G to generate the original image G(t), and then reconstructed by the forward generator F to obtain the labeled image F(G(t))∈T.

[0100] The multimodal fused image obtained in step S1 is input into the pre-trained generator network, and denoising is achieved through the following processing flow:

[0101] Feature extraction: The encoder extracts multi-scale features through cascaded convolutional layers, where shallow convolutions capture edge and texture details, and deep convolutions extract semantic-level features.

[0102] Residual learning: Introducing skip connections between the encoder and decoder to fuse low-level features with high-level features channel by channel, suppressing the loss of spatial information caused by downsampling.

[0103] Image reconstruction: The decoder restores the image resolution by progressively upsampling, and the number of output channels is consistent with the number of input data modalities, thus preserving the inherent characteristics of multimodal data.

[0104] For the time-domain signal acquired by the ultrasonic sensor, a combined wavelet transform and threshold shrinkage denoising method is employed. Specifically, this includes:

[0105] Wavelet decomposition: Selecting symlets as the mother wavelet to perform multi-scale decomposition on the original signal, obtaining high-frequency detail coefficients and low-frequency approximation coefficients.

[0106] Thresholding: Apply a soft thresholding function to high-frequency coefficients to suppress noise components.

[0107]

[0108] in, c represents the coefficients after wavelet denoising at the j-th layer; j λ represents the original coefficients of the j-th level wavelet decomposition; λ is the adaptive threshold parameter; sign(c j ) is a symbolic function; max(|c j |-λ,0) is the soft thresholding function.

[0109] Signal reconstruction: The processed coefficients are subjected to inverse wavelet transform to reconstruct the denoised ultrasonic signal.

[0110] This step effectively separates noise components from true edge features in multimodal data through the adversarial training mechanism and cycle consistency constraints of generative adversarial networks. The generator network achieves multimodal joint denoising of optical images, depth data, and infrared images through an encoder-decoder structure and residual learning; the parallel wavelet denoising of ultrasonic signals is specifically optimized for their temporal characteristics.

[0111] S3. Perform edge detection and gap analysis on the image data after image denoising to obtain a gap distribution map;

[0112] In this embodiment, edge detection and gap analysis of the image data after image denoising are achieved by combining deep learning segmentation and statistical methods, specifically including the following technical contents:

[0113] Edge detection is performed on the denoised multimodal data using the UNet model, including:

[0114] Obtain the UNet network model and two Sobel operators;

[0115] Based on multimodal data, at least thousands of two-dimensional image data are obtained through projection from different angles;

[0116] The two-dimensional image data generated by projection is used as the training and testing set for the model;

[0117] The UNet network model is trained based on the training set, and the prediction results are obtained from the query data to obtain a preliminary segmented image. Based on the preliminary segmented image obtained by UNet, the edge information of the image is enhanced by the background thresholding method. Based on the preliminary segmented image with enhanced edge information, the gradient information of each pixel is obtained by calculating the horizontal and vertical gradients of each pixel in the preliminary segmented image using the Sobel operator.

[0118] Based on the gradient information of the image, the binarized image is filtered using a dual thresholding method, and after edge smoothing, the accurate edge contour is obtained.

[0119] Furthermore, a UNet network is trained based on the training set, and the prediction results are used on the query data to obtain a preliminary segmented image, specifically:

[0120] Use the following loss function for training:

[0121]

[0122] Where K is the set of indices of all samples in the current training batch; k∈K is the index of a single sample in the sample set, representing the k-th input sample; I is the set of indices of all pixels in a single sample image; i∈I is the index of a single pixel in the pixel set, representing the i-th pixel; This represents the predicted output value of the UNet model for the i-th pixel in the k-th sample. It is the true binary label value of the i-th pixel in the k-th sample.

[0123] The initial learning rate is lrbase, and an adaptive learning rate degradation method is used: if the training set loss decreases less than a pre-set parameter c within 30 epochs, the learning rate is reduced by a factor of l until the learning rate is less than cmin, at which point training stops. During prediction on the test set, the largest connected region predicted by the model is used as the final prediction result.

[0124] Furthermore, based on the initial segmentation image obtained from UNet, after enhancing the edge information of the image using a background thresholding method, the gradient information of each pixel is obtained by calculating the horizontal and vertical gradients of each pixel in the initial segmentation image using the Sobel operator. Specifically:

[0125] The gray value of the Hth percentile is used as the threshold T. For pixels in the image whose threshold is less than T, the threshold is raised to T to obtain the image data Q.

[0126] Two kernels of size k×k are used to calculate the horizontal and vertical gradients for each pixel in the image data Q. The absolute values ​​of the horizontal and vertical gradients are added together to calculate the gradient value of each pixel.

[0127] Furthermore, based on the gradient information of the image, the binarized image is filtered using a dual threshold method. After smoothing the edges, the accurate edge contour is obtained. Specifically, two thresholds are set: those higher than the high value are marked as strong edge pixels, those lower than the low value are suppressed, and those in the middle are marked as weak edge pixels. The breakpoints in the strong edge pixels are connected with the weak edge pixels to obtain the connected binary edge image.

[0128] The UNet model is particularly useful in image segmentation and edge detection, especially when processing oral images, allowing for customized adjustments to address the minute gaps between restorations and abutment teeth. UNet's advantage lies in its unique symmetrical structure, enabling the encoder to extract image features, the decoder to restore spatial resolution, and ultimately, fine-grained edge detection. To adapt to the edge features of restorations in oral images, the UNet model needs to be trained on oral data, with specific optimizations for the minute gaps between restorations and abutment teeth. Increasing convolutional layer depth, adjusting kernel size, and improving skip connections can enhance the model's sensitivity and accuracy in detecting these minute gaps.

[0129] The statistical results are visualized as a gap distribution map in the form of a heatmap. The specific process includes:

[0130] Spatial mesh generation: The contact area is divided into uniform mesh cells (preferably, the mesh size is 100μm×100μm), and the average gap value within each cell is calculated.

[0131] Color mapping: Define color encoding rules based on the range of gap values, for example:

[0132] Red indicates a gap value greater than 100μm;

[0133] Yellow indicates a gap value between 50μm and 100μm;

[0134] Green indicates a gap value of less than 50 μm.

[0135] Overlay display: The heat map is registered and overlaid with the original 3D model, and multi-dimensional data fusion display is achieved through transparency control.

[0136] This step utilizes the UNet model to achieve high-precision edge segmentation and combines geometric analysis and statistical methods to quantify the gap distribution characteristics. Analysis of variance (ANOVA) is used to characterize the fit from multiple dimensions, including mean, standard deviation, and contact area percentage, providing data support for the subsequent scoring model.

[0137] S4. Evaluate the gap distribution map using the scoring criteria to obtain the fit distribution map; three levels of evaluation intervals can be established: perfect fit <50μm, minor gap 50–100μm is clinically acceptable, and >100μm requires re-making;

[0138] In this embodiment, the evaluation of the scoring criteria for the gap distribution map is achieved by constructing a dynamic quantitative scoring model, which specifically includes the following technical aspects:

[0139] The scoring model, through a comprehensive analysis integrating gap distribution characteristics and contact area ratio, defines the comprehensive fit score as a weighted sum of an exponential decay term and a linear compensation term, with the mathematical expression as follows:

[0140]

[0141] Where, d i w represents the gap value at the i-th measurement point. i is the weighting coefficient for the corresponding region; CAS% is the contact area percentage; β is the material attenuation coefficient; α is the gap weighting factor; γ is the contact area compensation coefficient; n is the total number of valid measurement points.

[0142] The material attenuation coefficient α is dynamically adapted based on the biomechanical properties of the restoration material. Preferably, the coefficient is positively correlated with the material's elastic modulus and is determined by the following formula:

[0143]

[0144] Among them, E material E represents the elastic modulus of the prosthesis material. reference The elastic modulus is used as a reference material (such as tooth enamel). This design makes the gap sensitivity of high-stiffness materials (such as zirconia) more significant, while the scoring of low-stiffness materials (such as resin) focuses more on the contact area.

[0145] The gap attenuation factor β is based on the standard deviation σ of the gap distribution. d Dynamic adjustment is defined as:

[0146] β = 15 + 0.5·σ d ;

[0147] Where β is the gap attenuation factor; σ d The standard deviation is denoted as .

[0148] This formula results in a high dispersion (σ) in the gap distribution. d When σ is relatively large, the penalty of the exponential decay term is adaptively reduced to avoid scoring distortion due to local outliers. Preferably, σ... d At 30μm, the anti-interference mode is enabled, and an upper limit constraint is applied to β.

[0149] Weighting coefficient w i The settings are differentiated based on the biomechanical importance of the measurement point's location:

[0150] Occlusal surface area: The weighting coefficient is set to 1.2 to reflect the stringent fit requirements of high stress concentration areas under chewing load;

[0151] Adjacent surface area: The weight coefficient is set to 1.0 to balance functional and aesthetic requirements;

[0152] Gingival margin region: The weighting coefficient is set to 0.8 to reduce the scoring bias caused by adaptive compensation of gingival tissue.

[0153] Preferably, the region division is automatically identified through three-dimensional model curvature analysis, and local areas with a curvature radius of less than 0.5 mm are identified as occlusal functional areas.

[0154] The contact area compensation coefficient γ is dynamically adjusted based on the type of restoration material and surface roughness.

[0155] All-ceramic material: The compensation coefficient is set to 0.5 to reflect the dependence of its low-friction characteristics on the contact area;

[0156] Metallic materials: The compensation coefficient is set to 0.4 to reduce the interference of oxide film deformation on the contact area;

[0157] Composite resin: The compensation coefficient is calculated as 0.6-0.02·Ra, where Ra is the surface roughness parameter (unit: μm). This design allows for targeted compensation in the scoring of contact area loss caused by rough surfaces.

[0158] The degree of fit is graded based on the scoring results, and the threshold setting rules are as follows:

[0159] Perfect fit: The score is higher than the preset upper limit threshold;

[0160] Slight gap: The score is within the middle threshold range;

[0161] Obvious gap: The score is below the lower limit threshold.

[0162] Preferably, the threshold is dynamically adjusted according to the type of restoration. For example, the perfect fit threshold for all-ceramic restorations is 5% higher than that for resin restorations.

[0163] This step utilizes a dynamic quantitative scoring model to organically combine the nonlinear influence of gap distribution with the linear contribution of contact area, achieving a multi-dimensional comprehensive assessment of fit. The dynamic adjustment mechanism of each parameter in the model addresses the insufficient adaptability of traditional fixed threshold methods in scenarios involving material diversity and clinical complexity.

[0164] S5. Overlay the conformity distribution map onto the real environment using augmented reality technology;

[0165] In this embodiment, an augmented reality (AR) technology is used to overlay a density distribution map onto the real-world environment. This is achieved through spatial registration with multi-device adaptation, dynamic rendering, and real-time data synchronization. Specifically, the following technical aspects are included:

[0166] A unified spatial coordinate system is established through the cross-platform OpenXR framework to achieve pose alignment of different AR devices. Preferably, for AR glasses devices, a point cloud map of the oral cavity environment is constructed using visual inertial odometry (VIO)-based SLAM technology. Through feature point matching and bundle adjustment optimization, the fit distribution map is registered with the real oral cavity scene at the sub-millimeter level.

[0167] For smart tablets or mobile devices, an improved perspective n-point (PnP) algorithm is used to solve the camera pose. Specifically, the RANSAC algorithm is used to remove mismatched point pairs, and reprojection error constraints are introduced to optimize the rotation matrix R and translation vector T. device =R·T scan +t;

[0168] Where R is the rotation matrix; t is the translation vector; T scan For the scanning system coordinate system; T device For the AR device coordinate system, its mathematical expression is:

[0169] T device =R·T scan +t;

[0170] Where R is the rotation matrix; t is the translation vector; T scan For the scanning system coordinate system; T device This is the coordinate system for the AR device.

[0171] The Vulkan graphics engine is used to render the density distribution map in real time. Preferably, the rendering effect is dynamically adjusted according to the gap value.

[0172] Color coding rules:

[0173] The red channel (RGB255,0,0) indicates that the gap value is greater than the preset upper limit threshold (e.g., 100μm);

[0174] The yellow channel (RGB255,255,0) indicates that the gap value is in the middle threshold range (e.g., 50-100μm);

[0175] The green channel (RGB0,255,0) indicates that the gap value is lower than the preset lower limit threshold (e.g., 50μm).

[0176] Transparency control algorithm:

[0177]

[0178] Where, θ lower and θ upper These are the lower and upper limits of the grading threshold, respectively; α blend The transparency blending coefficient; max(d) i) represents the maximum gap value within the current rendering area.

[0179] Design differentiated interaction logic for different device types:

[0180] A MediaPipe gesture recognition model is deployed to parse operation commands by detecting the coordinates of key points on the hand. Preferably, a fist-clenching gesture lasting 2 seconds triggers a switch in the 2D profile view, and the index finger's sliding trajectory is mapped to the scaling adjustment of the heatmap, with the scaling step dynamically adapted according to the sliding speed.

[0181] A multi-touch priority strategy is adopted, defining the three-finger pinch action as the activation condition for the multi-time point data comparison mode, and the two-finger rotation operation is transformed into the observation angle adjustment of the three-dimensional model around the Z-axis, with the angle resolution set to 1°.

[0182] Low-latency transmission of the close-fitting distribution map and rendering commands is achieved through the WebRTC protocol. Preferably, the transport layer employs the UDP protocol and forward error correction (FEC) mechanism to ensure that the end-to-end latency remains below 50ms even under network jitter conditions.

[0183] Data compression employs lightweight feature encoding based on EfficientNet-Lite, mapping the original point cloud data and rendering parameters into low-dimensional feature vectors, achieving a compression ratio of up to 8:1.

[0184] For security verification, each frame of data is appended with a 32-byte SHA-256 hash digest. The receiving end verifies the data integrity by comparing the digests, preventing the AR overlay content from being tampered with during transmission.

[0185] This step utilizes the OpenXR framework to achieve cross-device spatial registration, combined with dynamic rendering algorithms and interactive control strategies, enabling the fit assessment results to be accurately superimposed onto the real oral cavity environment.

[0186] The apparatus for assessing the marginal fit of a fixed denture described below can be used in conjunction with the method for assessing the marginal fit of a fixed denture described above.

[0187] Please see the appendix Figure 5 The present invention also provides an evaluation device for the marginal fit of fixed dentures, comprising:

[0188] The first processing module is used to acquire and enhance two-dimensional high-resolution images, three-dimensional depth data, and data of micro-cracks or irregular surfaces in saliva and blood environments through high-definition optical probes, depth cameras, ultrasonic sensors, and infrared scanning sensors, and to fuse the acquired data into multimodal image data through a cross-modal feature pyramid network.

[0189] The second processing module is used to denoise the multimodal image data;

[0190] The third processing module is used to perform edge detection and gap analysis on the image data after image denoising to obtain a gap distribution map.

[0191] The fourth processing module is used to evaluate the gap distribution map according to the scoring criteria to obtain the fabric tightness distribution map;

[0192] The fifth processing module is used to overlay the conformity distribution map onto the real environment using augmented reality technology.

[0193] The device in this embodiment can be used to execute the above method embodiments, and its principle and technical effects are similar, so they will not be described again here.

[0194] The assessment system for the marginal fit of fixed dentures described below can be used in conjunction with the assessment method for the marginal fit of fixed dentures described above.

[0195] Please see the appendix Figure 6 The present invention also provides a system for evaluating the marginal fit of fixed dentures, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the above-described method for evaluating the marginal fit of fixed dentures is performed.

[0196] The system in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so they will not be described again here.

[0197] The storage medium described below and the method for evaluating the marginal fit of a fixed denture described above can be referred to in correspondence.

[0198] Please see the appendix Figure 7 The present invention also provides a storage medium storing a computer program thereon, which executes the above-described method for evaluating the marginal fit of a fixed denture when running.

[0199] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0200] The storage medium in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so it will not be described again here.

[0201] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the marginal fit of a fixed denture, characterized by, The method comprises the following steps: S1, acquiring multi-modal image data of the contact area between the denture and the abutment; S2, performing image denoising on the multi-modal image data; S3, performing edge detection and gap analysis on the image data after image denoising processing to obtain a gap distribution map; S4, evaluating the gap distribution map according to a scoring standard to obtain a fit distribution map; three evaluation intervals can be set: perfect fit < 50 μm, slight gap 50-100 μm clinically acceptable, and > 100 μm requiring remaking; S5, superimposing the fit distribution map into the real environment through augmented reality technology; The S3 comprises: S31, using a UNet model to perform edge detection on the denoised multi-modal data; S32, analyzing whether the edge of the prosthesis is overhanging, missing or irregular according to the results of edge detection; S33, using variance analysis algorithm to statistically analyze the distribution of the edge gap in the contact area to obtain a gap distribution map; The S4 comprises: A quantitative scoring model is constructed, and the fit comprehensive score is calculated by the following formula: ； in, For the first The gap value at each measurement point; These are the weighting coefficients for the corresponding regions; This represents the percentage of the contact area. The material attenuation coefficient; This is the gap weighting factor; This is the contact area compensation coefficient; This represents the total number of valid measurement points.

2. The method for evaluating the marginal fit of a fixed denture according to claim 1, wherein The S1 comprises: S11, acquiring two-dimensional high-resolution images of the contact area between the denture and the abutment through a high-definition optical probe; S12, acquiring three-dimensional depth data of the contact area between the denture and the abutment through a depth camera; S13, acquiring micro-cracks or irregular surfaces of the contact area between the denture and the abutment through an ultrasonic sensor; S14, enhancing the two-dimensional high-resolution images, three-dimensional depth data, micro-cracks or irregular surfaces under the saliva and blood environment through an infrared scanning sensor; S15, fusing the enhanced two-dimensional high-resolution images, three-dimensional depth data, micro-cracks or irregular surfaces through a cross-modal feature pyramid network to obtain multi-modal image data.

3. The method of claim 1, wherein the method is used for evaluating the marginal fit of a fixed denture. The S2 comprises: S21, training a generative adversarial network using a cycle consistency algorithm, wherein the generative adversarial network comprises a generator network for generating high-quality images consistent with the actual situation of the oral prosthesis and a discriminator network for judging whether the images are real; S22, performing denoising processing on the multi-modal data through the trained generative adversarial network.

4. The method of claim 1, wherein the method is used for evaluating the marginal fit of a fixed denture. The S5 comprises: Three-dimensional space superposition is realized through AR glasses or smart tablets and mobile phones, and the specific steps are as follows: A unified coordinate system is established using a cross-platform OpenXR framework Device difference processing: AR glasses: construct a point cloud map of the oral environment through SLAM technology; Smart tablets and mobile phones: solve the camera pose using an improved PnP algorithm; Coordinate system conversion matrix: ； wherein, is a rotation matrix; is a translation vector; is a scan system coordinate system; is an AR device coordinate system; Dynamic rendering: render the gap heat map through the Vulkan engine; Real-time data synchronization: Transmission protocol: WebRTC low-latency transmission; Data compression: lightweight feature encoding based on EfficientNet-Lite; Security check: SHA-256 hash check.

5. An evaluation device for evaluating marginal fit of a fixed denture, which implements the evaluation method for evaluating marginal fit of a fixed denture according to any one of claims 1 to 4, characterized by Comprise: The first processing module is used for collecting and enhancing two-dimensional high-resolution images, three-dimensional depth data, micro-crack or irregular surface data in a saliva or blood environment through a high-definition optical probe, a depth camera, an ultrasonic sensor and an infrared scanning sensor, and fusing the collected data into multi-modal image data through a cross-modal feature pyramid network. The second processing module is used for image denoising on the multi-modal image data. The third processing module is used for edge detection and gap analysis on the image data after image denoising processing, to obtain a gap distribution map. The fourth processing module is used for evaluating the gap distribution map according to a scoring standard, to obtain a distribution map of the tightness. The fifth processing module is used for superimposing the tightness distribution map into a real environment through augmented reality technology.

6. A system for evaluating the marginal fit of a fixed denture, characterized by The computer program is executed to perform the method for evaluating the edge tightness of the fixed denture according to any one of claims 1-4. The computer program is executed to perform the method for evaluating the edge tightness of the fixed denture according to any one of claims 1-4.

7. A storage medium having stored thereon a computer program, characterized in that ​

Images