A method for fixed restoration suitability assessment based on image data
By acquiring and processing image data of dental implant fixed prostheses, and combining semantic segmentation and optical flow algorithms, the micro-displacement field of the prosthesis and abutment is calculated, which solves the problem that existing technologies cannot accurately assess the suitability of dental implant fixed prostheses, and realizes accurate assessment and improvement under dynamic occlusion.
Patent Information
- Application Number
- CN202610371293.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-25
- Publication Date
- 2026-07-03
AI Technical Summary
Existing technologies, when assessing the suitability of dental implant fixed prostheses, neglect dynamic occlusal scenarios, resulting in significant discrepancies between assessment results and actual clinical use. This makes it impossible to accurately reflect the true suitability of the prosthesis and also presents issues of radiation or strong subjectivity.
An image-based evaluation method was adopted, which collected continuous temporal images of the patient's simulated occlusion process and static baseline images of the non-occlusal state. Combined with semantic segmentation and optical flow algorithms, the micro-displacement field of the prosthesis and abutment was calculated, and a suitability score was output using a temporal anomaly detection model.
It achieves accurate reflection of the true fit of fixed prostheses in dynamic occlusal scenarios, eliminates interference from scanning equipment movement and patient head shaking, provides accurate fit assessment, and avoids the risks of prosthesis loosening and bacterial growth.
Smart Images
Figure CN122335693A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to a method for assessing the suitability of fixed prostheses based on image data. Background Technology
[0002] Dental implantation is the mainstream method for restoring missing or damaged teeth. It mainly consists of three parts: the implant, the abutment, and the fixed prosthesis. The implant, as an artificial tooth root, is inserted into the jawbone and fixedly supported through osseointegration. The abutment connects the implant and the fixed prosthesis, transmitting occlusal forces and supporting the prosthesis. The fixed prosthesis is mostly an overlay structure, fitted over the abutment, used to restore the chewing function and aesthetics of the teeth. Its fit with the abutment directly determines the restorative effect and long-term stability. After the implant surgery, the suitability of the fixed prosthesis needs to be monitored long-term, with a focus on assessing the tightness of its inner wall against the outer wall of the abutment and the bottom edge against the abutment. This is to avoid complications such as microbial invasion and peri-implantitis caused by poor fit, and to ensure the lifespan of the prosthesis.
[0003] Currently, the suitability assessment of fixed prostheses mostly adopts static assessment methods, mainly using techniques such as Micro-CT, two-dimensional imaging, or probe palpation to measure the gap size between the prosthesis and abutment in a static state to determine suitability. Traditional assessment methods ignore the dynamic occlusal scenario during oral chewing, and cannot capture the relative displacement between the prosthesis and abutment during occlusal micro-movements. It is difficult to discover hidden fit problems in a static state. In addition, some static assessment methods have problems such as radiation, need to destroy test pieces, or strong subjectivity. The assessment results deviate significantly from the actual clinical use state and cannot accurately reflect the true suitability of the prosthesis, making it difficult to meet the needs of accurate clinical assessment. Summary of the Invention
[0004] In view of this, the present invention proposes a method for assessing the fit of fixed prostheses based on image data, which combines dynamic occlusal scenarios for assessment and can accurately reflect the true fit of fixed prostheses.
[0005] The technical solution of this invention is implemented as follows: A method for assessing the fit of a fixed prosthesis based on image data includes the following steps: Step S1: Acquire continuous time-series images of the patient's simulated occlusion process after implantation of the fixed prosthesis, as well as static baseline images of the non-occlusal state. Step S2: Denoise the continuous temporal images and static reference images, and separate the fixed restoration area and the abutment area through semantic segmentation; Step S3: Based on the fixation prosthesis region and the abutment region, lock the internal mating surface between the inner wall of the fixation prosthesis and the outer wall of the abutment, as well as the edge mating surface between the bottom edge of the fixation prosthesis and the edge of the abutment. Step S4: Use the optical flow algorithm to calculate the pixel displacement vectors of the internal and edge bonding surfaces of the corresponding consecutive frames of the continuous temporal image to obtain the preliminary micro-displacement field; Step S5: Using the internal bonding surface and edge bonding surface of the static reference image as a reference, the preliminary micro-displacement field is corrected to obtain the relative micro-displacement field. Step S6: Input the relative micro-displacement fields corresponding to the internal bonding surface and the edge bonding surface into the temporal anomaly detection model, and output the suitability score by the temporal anomaly detection model.
[0006] Preferably, step S1 includes the following steps: Step S11: Use a high-speed intraoral scanner to acquire static baseline images of the patient in a non-occlusal state after implantation of the fixed prosthesis; Step S12: When the patient performs three simulated biting states—light biting, normal biting, and forceful biting—a high-speed intraoral scanner is used to acquire continuous time-series images of the simulated biting process.
[0007] Preferably, step S2 includes the following specific steps: Step S21: Adaptive median filtering algorithm is used to remove noise, artifacts and interfering pixels from continuous time-series images and static reference images; Step S22: Using the trained CNN semantic segmentation network, segment the denoised continuous temporal images and the static reference images respectively; Step S23: Feature extraction and classification are performed using a CNN semantic segmentation network to separate the fixed prosthesis region and the abutment region.
[0008] Preferably, the specific steps for locking the internal bonding surface in step S3 include: Step S31: Extract the three-dimensional point cloud contours of the inner wall of the fixed restoration and the outer wall of the abutment respectively, and calculate the geometric center, principal direction vector and bounding box size; Step S32: Use the iterative nearest point algorithm to coarsely register the point cloud of the inner wall of the fixed restoration with the point cloud of the outer wall of the abutment; Step S33: For the region after coarse registration, construct a bidirectional topological adjacency graph between the inner wall of the fixed repair body and the outer wall of the abutment, and traverse the triangular facet matching relationships. Step S34: Identify a set of continuous facets where the angle between the facet normals is less than a preset angle threshold and the distance between facets is less than a preset bonding threshold, and lock the projection area of the set of continuous facets as the internal bonding surface.
[0009] Preferably, the specific steps for locking the edge bonding surface in step S3 include: Step S35: Extract the edge loop at the bottom of the fixed prosthesis and the edge loop at the corresponding position of the abutment, perform smooth fitting, and calculate the radial distance distribution of the fitted curve; Step S36: When the radial distance distribution falls within the preset threshold of the width of the annular sealing strip and the edge ring closure meets the requirements, lock the annular contact area between the bottom edge of the fixed repair body and the abutment as the edge contact surface.
[0010] Preferably, step S4 includes the following specific steps: Step S41: Perform region cropping on adjacent consecutive frame images in the continuous temporal image to obtain pixel regions that only contain the internal bonding surface and the edge bonding surface; Step S42: Construct a sparse optical flow estimation algorithm. For pixel regions containing only the internal bonding surface, use the weighted least squares method to solve the optical flow equation and calculate the first displacement vector of each pixel point in the internal bonding surface between consecutive frame images. Step S43: For pixel regions containing only edge-fitting surfaces, use a circular window to perform optical flow calculations, track the displacement changes of pixels on the edge loop, and obtain the second displacement vector; Step S44: Integrate the first displacement vector and the second displacement vector according to the pixel coordinate order, and generate the preliminary micro-displacement field of the inner bonding surface and the preliminary micro-displacement field of the edge bonding surface respectively.
[0011] Preferably, after calculating the first displacement vector and the second displacement vector, an effectiveness filter is performed to remove abnormal vectors whose displacement amplitude exceeds a preset threshold and invalid vectors whose optical flow convergence fails.
[0012] Preferably, step S5 includes the following specific steps: Step S51: Establish a reference coordinate system based on the coordinates of all pixels of the inner bonding surface and the edge bonding surface in the static reference image; Step S52: Map the pixels of the internal and edge bonding surfaces corresponding to the continuous temporal images to the corresponding pixels of the static reference image in the reference coordinate system. Step S53: Based on coordinate mapping, using the fixed feature points at the connection between the abutment and the implant as a reference, identify and separate the rigid body displacement components of the abutment itself through feature point displacement analysis. Step S54: Based on the coordinate mapping relationship of the separated rigid body displacement components, perform coordinate transformation and displacement correction on the preliminary micro-displacement field to obtain the relative micro-displacement field.
[0013] Preferably, step S6 includes the following specific steps: Step S61: Normalize and feature-stitched the relative micro-displacement fields corresponding to the internal bonding surface and the edge bonding surface to generate a fused temporal feature vector. Step S62: Input the fused temporal feature vector into the trained temporal anomaly detection model, and the temporal anomaly detection model outputs the internal fitness score and the marginal fitness score respectively; Step S63: Perform a weighted fusion of the internal fitness score and the marginal fitness score to generate the overall fitness score.
[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention discloses a method for assessing the fit of fixed prostheses based on image data. During image data acquisition, in addition to static images, continuous temporal images of the patient simulating occlusion are also acquired. After preprocessing the continuous temporal images and the static reference images, the internal contact surfaces of the inner wall of the fixed prosthesis and the outer wall of the abutment, as well as the edge contact surfaces of the bottom edge of the fixed prosthesis and the abutment, are identified. Then, an optical flow algorithm is introduced to construct preliminary micro-displacement fields for the internal and edge contact surfaces to reflect internal and edge offsets during occlusion. Simultaneously, the static reference image is used as a benchmark to correct the preliminary micro-displacement fields, eliminating irrelevant displacement interference caused by scanning equipment movement or slight head movements of the patient. This ensures that the displacement data focuses on the relative movement of the fixed prosthesis and the abutment. Finally, a temporal anomaly detection model can detect the relative micro-displacement fields and output a fit score. By introducing a dynamic occlusion scenario, the true fit of the fixed prosthesis can be accurately reflected. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart of a method for assessing the suitability of a fixed prosthesis based on image data according to the present invention; Figure 2 This is a flowchart of step S1 of the image data-based method for assessing the fit of a fixed prosthesis according to the present invention. Figure 3 This is a flowchart of step S2 of the image data-based method for assessing the fit of a fixed prosthesis according to the present invention; Figure 4 This is a flowchart of step S3 of the image data-based method for assessing the fit of a fixed prosthesis according to the present invention. Figure 5 This is a flowchart of step S4 of the image data-based method for assessing the fit of a fixed prosthesis according to the present invention. Figure 6 This is a flowchart of step S5 of the image data-based method for assessing the fit of a fixed prosthesis according to the present invention; Figure 7 This is a flowchart of step S6 of a method for assessing the suitability of a fixed prosthesis based on image data according to the present invention. Detailed Implementation
[0017] To better understand the technical content of this invention, a specific embodiment is provided below, and the invention will be further described in conjunction with the accompanying drawings.
[0018] See Figures 1 to 7 The present invention provides a method for assessing the fit of a fixed prosthesis based on image data, comprising the following steps: Step S1: Acquire continuous time-series images of the patient's simulated occlusion process after implantation of the fixed prosthesis, as well as static baseline images of the non-occlusal state. Step S2: Denoise the continuous temporal images and static reference images, and separate the fixed restoration area and the abutment area through semantic segmentation; Step S3: Based on the fixation prosthesis region and the abutment region, lock the internal mating surface between the inner wall of the fixation prosthesis and the outer wall of the abutment, as well as the edge mating surface between the bottom edge of the fixation prosthesis and the edge of the abutment. Step S4: Use the optical flow algorithm to calculate the pixel displacement vectors of the internal and edge bonding surfaces of the corresponding consecutive frames of the continuous temporal image to obtain the preliminary micro-displacement field; Step S5: Using the internal bonding surface and edge bonding surface of the static reference image as a reference, the preliminary micro-displacement field is corrected to obtain the relative micro-displacement field. Step S6: Input the relative micro-displacement fields corresponding to the internal bonding surface and the edge bonding surface into the temporal anomaly detection model, and output the suitability score by the temporal anomaly detection model.
[0019] After a period of time following the implantation of a fixed prosthesis, the patient needs to have the fit of the prosthesis assessed. Since the fixed prosthesis is fitted onto the abutment, the assessment includes the degree of fit between the inner wall of the prosthesis and the outer wall of the abutment, as well as the degree of fit between the bottom edge of the prosthesis and the edge of the abutment. These two types of fit are defined as internal fit and marginal fit. Traditional assessment methods mostly collect data under static conditions. However, patients open and close their mouths frequently during daily activities such as eating and speaking, resulting in frequent biting. If the fixed prosthesis has a poor fit, it will cause problems during biting. During the fitting process, issues such as loosening of the fixed prosthesis and bacterial growth in the gaps can easily occur. Therefore, when assessing the fit of patients with implanted fixed prostheses, in addition to acquiring traditional static reference images of the non-occlusal state, continuous time-series images of the patient simulating the occlusal process are also acquired. Based on the continuous time-series images, internal and marginal fit are assessed, and corrections are made using the static reference images to eliminate interference, thereby obtaining an accurate fit score to truly reflect the patient's fit for the fixed prosthesis implantation. When the fit score is low, timely improvement measures can be taken to avoid problems such as fixed prosthesis dislodgement and bacterial growth.
[0020] After acquiring continuous temporal images and static reference images, denoising preprocessing was performed on both to improve image clarity. Then, semantic segmentation was used to separate the fixed prosthesis region and abutment region from each frame of the continuous temporal images and static reference images. These regions further identified the internal fitting surface related to internal fit and the edge fitting surface related to edge fit. The internal fitting surface refers to the fitting surface between the inner wall of the fixed prosthesis and the outer wall of the abutment, while the edge fitting surface refers to the fitting surface between the bottom edge of the fixed prosthesis and the edge of the abutment. After determining the internal... After the occlusion surface and edge occlusion surface are determined, an optical flow algorithm is introduced to calculate the pixel displacement vectors of the internal and edge occlusion surfaces in consecutive frames, and a preliminary micro-displacement field is constructed. The preliminary micro-displacement field can reflect the changes of each pixel at the position of the internal and edge occlusion surfaces during the occlusion process. Then, using a static reference image as a reference, the preliminary micro-displacement field is corrected to eliminate irrelevant displacement interference caused by the movement of the scanning device or slight shaking of the patient's head, ensuring that the displacement data is focused on the relative movement of the prosthesis and the abutment. Finally, after obtaining the relative micro-displacement field, it can be processed by a temporal anomaly detection model to output the corresponding suitability score.
[0021] Preferably, step S1 includes the following steps: Step S11: Use a high-speed intraoral scanner to acquire static baseline images of the patient in a non-occlusal state after implantation of the fixed prosthesis; Step S12: When the patient performs three simulated biting states—light biting, normal biting, and forceful biting—a high-speed intraoral scanner is used to acquire continuous time-series images of the simulated biting process.
[0022] A high-speed intraoral scanner was used to acquire images of the patient's oral cavity, ensuring that the images were clear and covered the entire structure of the fixed prosthesis and abutment. First, a static reference image was acquired in the non-occlusal state, which was used as a reference for subsequent correction. Then, the patient was asked to bite lightly, bite normally, and bite forcefully in sequence. Continuous time-series images of the entire dynamic occlusion process were then acquired. During the acquisition process, interference from oral soft tissue, saliva, and light reflection was avoided to ensure the effectiveness of the images.
[0023] Preferably, step S2 includes the following specific steps: Step S21: Adaptive median filtering algorithm is used to remove noise, artifacts and interfering pixels from continuous time-series images and static reference images; Step S22: Using the trained CNN semantic segmentation network, segment the denoised continuous temporal images and the static reference images respectively; Step S23: Feature extraction and classification are performed using a CNN semantic segmentation network to separate the fixed prosthesis region and the abutment region.
[0024] An adaptive median filtering algorithm can be used to denoise the acquired continuous temporal images and static reference images, removing the corresponding noise to improve image clarity. After denoising, a trained CNN semantic segmentation network is used to segment the denoised continuous temporal images and static reference images respectively. The CNN semantic segmentation network is trained on images of dental prostheses and abutments and has accurate segmentation capabilities. Through the feature extraction and classification functions of the CNN semantic segmentation network, the fixed prosthesis region and the abutment region can be accurately separated, and the boundary contours of the two can be clearly defined. At the same time, interference from other oral tissues such as the gingiva can be eliminated, ensuring the integrity of the segmented fixed prosthesis and abutment contours.
[0025] Preferably, the specific steps for locking the internal bonding surface in step S3 include: Step S31: Extract the three-dimensional point cloud contours of the inner wall of the fixed restoration and the outer wall of the abutment respectively, and calculate the geometric center, principal direction vector and bounding box size; Step S32: Use the iterative nearest point algorithm to coarsely register the point cloud of the inner wall of the fixed restoration with the point cloud of the outer wall of the abutment; Step S33: For the region after coarse registration, construct a bidirectional topological adjacency graph between the inner wall of the fixed repair body and the outer wall of the abutment, and traverse the triangular facet matching relationships. Step S34: Identify a set of continuous facets where the angle between the facet normals is less than a preset angle threshold and the distance between facets is less than a preset bonding threshold, and lock the projection area of the set of continuous facets as the internal bonding surface.
[0026] After obtaining the fixed prosthesis area and abutment area, it is necessary to further clarify the specific areas of the internal camouflage surface and the marginal camouflage surface to reduce irrelevant areas and improve the accuracy of the suitability assessment. Specifically, when locking the internal camouflage surface, three-dimensional point cloud reconstruction technology is first used to generate three-dimensional point cloud models of the inner wall of the fixed prosthesis and the outer wall of the abutment. The geometric center, principal direction vector and bounding box size of the two are calculated. Then, the iterative nearest point algorithm is used to perform coarse registration of the point cloud of the inner wall of the fixed prosthesis and the point cloud of the outer wall of the abutment to initially lock the overall spatial area of the prosthesis covering the abutment, eliminate gingival and background interference, and determine the approximate range of the camouflage surface. Finally, a bidirectional topological adjacency graph of the inner wall of the fixed prosthesis and the outer wall of the abutment is constructed. The triangular facet matching relationship of the two is traversed to identify a set of continuous facets with a facet normal vector angle ≤5° and a facet spacing ≤1μm (preset subpixel level camouflage threshold). The projection area of the continuous facet set is defined as the internal camouflage surface, and its facet number and three-dimensional coordinate range are marked.
[0027] Preferably, the specific steps for locking the edge bonding surface in step S3 include: Step S35: Extract the edge loop at the bottom of the fixed prosthesis and the edge loop at the corresponding position of the abutment, perform smooth fitting, and calculate the radial distance distribution of the fitted curve; Step S36: When the radial distance distribution falls within the preset threshold of the width of the annular sealing strip and the edge ring closure meets the requirements, lock the annular contact area between the bottom edge of the fixed repair body and the abutment as the edge contact surface.
[0028] When locking the edge bonding surface, two sets of edge loops are first extracted: the edge loop at the bottom of the fixed restoration and the edge loop at the corresponding position of the abutment. Then, B-spline curves are used to smooth and fit the two loops to eliminate contour noise. The radial distance distribution between the two loops after fitting is calculated, and the radial distance value of each point is recorded. The bonding of the two edge loops is verified based on the radial distance distribution. First, a threshold for the width of the annular sealing strip is set to determine whether the radial distance distribution falls entirely within the threshold range. At the same time, a contour closure detection algorithm is used to detect the closure degree of the edge loops. If the closure degree is ≥99.5% and there are no breaks or distortions, and the radial distance distribution meets the sealing strip width requirements, then the annular area is locked as the bonding area between the bottom edge of the fixed restoration and the abutment, and defined as the edge bonding surface. If any condition is not met, the edge loop fitting parameters are re-optimized and the detection is repeated until the conditions are met and the edge bonding surface is locked.
[0029] Preferably, step S4 includes the following specific steps: Step S41: Perform region cropping on adjacent consecutive frame images in the continuous temporal image to obtain pixel regions that only contain the internal bonding surface and the edge bonding surface; Step S42: Construct a sparse optical flow estimation algorithm. For pixel regions containing only the internal bonding surface, use the weighted least squares method to solve the optical flow equation and calculate the first displacement vector of each pixel point in the internal bonding surface between consecutive frame images. Step S43: For pixel regions containing only edge-fitting surfaces, use a circular window to perform optical flow calculations, track the displacement changes of pixels on the edge loop, and obtain the second displacement vector; Step S44: After calculating the first displacement vector and the second displacement vector, perform validity filtering to remove abnormal vectors with displacement amplitudes exceeding a preset threshold and invalid vectors that fail to converge optical flow. Then, integrate the first displacement vector and the second displacement vector according to the pixel coordinate order, and generate the preliminary micro-displacement field of the inner bonding surface and the preliminary micro-displacement field of the edge bonding surface, respectively.
[0030] The sparse optical flow estimation algorithm of this invention is the pyramid Lucas-Kanade optical flow algorithm with 3-5 layers, which can achieve sub-pixel level displacement calculation accuracy. Before constructing the initial micro-displacement field, consecutive frames in a continuous time series image are selected, and the time interval between frames should not be too large to ensure that the displacement changes during the interlocking micro-motion process can be effectively captured. Region clipping is performed to retain only the pixel regions of the internal mating surface and the edge mating surface, and interference from irrelevant regions is shielded. Then, for the pixel regions containing only the internal mating surface, the window size of the optical flow algorithm is set, and the weighted least squares method is used to solve the optical flow equation. The horizontal and vertical displacement vectors of each pixel point in the internal mating surface between consecutive frames are calculated, and the horizontal and vertical displacement vectors are output as the first displacement vector.
[0031] For pixel regions containing only edge-fitting surfaces, a circular window is used for optical flow calculation, focusing on tracking the displacement changes of edge loop pixels to avoid calculation errors caused by edge contour breaks. Finally, a second displacement vector can be obtained. After obtaining the first and second displacement vectors, validity filtering is performed to remove abnormal and invalid vectors to ensure the reliability of the displacement data. Then, the first and second displacement vectors are integrated in pixel coordinate order to generate preliminary micro-displacement fields for the corresponding internal and edge-fitting surfaces, and the displacement magnitude and direction information of each pixel are recorded.
[0032] Preferably, step S5 includes the following specific steps: Step S51: Establish a reference coordinate system based on the coordinates of all pixels of the inner bonding surface and the edge bonding surface in the static reference image; Step S52: Map the pixels of the internal and edge bonding surfaces corresponding to the continuous temporal images to the corresponding pixels of the static reference image in the reference coordinate system. Step S53: Based on coordinate mapping, using the fixed feature points at the connection between the abutment and the implant as a reference, identify and separate the rigid body displacement components of the abutment itself through feature point displacement analysis. Step S54: Based on the coordinate mapping relationship of the separated rigid body displacement components, perform coordinate transformation and displacement correction on the preliminary micro-displacement field to obtain the relative micro-displacement field.
[0033] The initial micro-displacement field contains errors caused by equipment movement and head shaking. Therefore, it is necessary to use a static reference image as a benchmark for correction. First, the pixels of the internal and edge bonding surfaces in the static reference image are extracted, and a unified reference coordinate system is established. The initial spatial position information of each reference pixel is recorded to provide a unique reference standard for subsequent micro-displacement correction. Based on the established reference coordinate system, an image feature point matching algorithm is used to accurately map the pixels of the corresponding bonding surfaces in the continuous time-series images obtained in step S4 with the reference pixels of the static reference image, establishing a one-to-one positional relationship between the two to ensure the correlation of displacement calculation. Based on the coordinate mapping, using the fixed feature points at the connection between the abutment and the implant as references, the rigid displacement components of the abutment itself are identified and separated through feature point displacement analysis, eliminating the errors caused by scanning. Irrelevant displacement interference from equipment movement or slight head shaking of the patient ensures that displacement data focuses on the relative motion between the prosthesis and the abutment. Finally, based on the coordinate mapping relationship after separating the rigid body displacement components, coordinate transformation and displacement correction can be performed on the preliminary micro-displacement field to accurately calculate the true relative displacement of the fixed prosthesis relative to the abutment, obtaining the preliminary corrected relative micro-displacement field. Then, the displacement consistency of the preliminary corrected relative micro-displacement field is checked. A continuity detection algorithm is used to determine whether the displacement difference between adjacent pixels exceeds a preset threshold. If there is a region of abrupt displacement, the region is locally recorrected and smoothed to ensure the continuity and rationality of the relative micro-displacement field. Using a static reference image as a fixed reference, the correction can eliminate this error, ensuring that the relative micro-displacement field only reflects the true micro-movements of the prosthesis and the abutment, guaranteeing the accuracy of subsequent scoring.
[0034] Preferably, step S6 includes the following specific steps: Step S61: Normalize and feature-stitched the relative micro-displacement fields corresponding to the internal bonding surface and the edge bonding surface to generate a fused temporal feature vector. Step S62: Input the fused temporal feature vector into the trained temporal anomaly detection model, and the temporal anomaly detection model outputs the internal fitness score and the marginal fitness score respectively; Step S63: Perform a weighted fusion of the internal fitness score and the marginal fitness score to generate the overall fitness score.
[0035] After obtaining the relative micro-displacement field, a fitness assessment can be performed using a temporal anomaly detection model. The relative micro-displacement field first needs to be standardized to normalize the data to the [0,1] interval, eliminating numerical differences caused by different image acquisition conditions. Then, the two standardized sets of relative micro-displacement fields are concatenated to generate a fused temporal feature vector, ensuring that the feature dimension matches the input dimension of the temporal anomaly detection model. The fused temporal feature vector is then input into the trained Transformer temporal anomaly detection model. The model automatically captures temporal and spatial distribution anomaly features in the micro-displacement field through a multi-head attention mechanism and outputs two paths: an internal fitness score and an edge fitness score. These scores accurately reflect the fixation stability of the internal bonding surface and the sealing reliability of the edge bonding surface, respectively, avoiding the misrepresentation of a particular bonding surface by a single score. Different weights are assigned to the internal and edge fitness scores in a 7:3 ratio, and finally, a weighted fusion is performed to obtain the overall fitness score.
[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for assessing the fit of a fixed prosthesis based on image data, characterized in that, Includes the following steps: Step S1: Acquire continuous time-series images of the patient's simulated occlusion process after implantation of the fixed prosthesis, as well as static baseline images of the non-occlusal state. Step S2: Denoise the continuous temporal images and static reference images, and separate the fixed restoration area and the abutment area through semantic segmentation; Step S3: Based on the fixation prosthesis region and the abutment region, lock the internal mating surface between the inner wall of the fixation prosthesis and the outer wall of the abutment, as well as the edge mating surface between the bottom edge of the fixation prosthesis and the edge of the abutment. Step S4: Use the optical flow algorithm to calculate the pixel displacement vectors of the internal and edge bonding surfaces of the corresponding consecutive frames of the continuous temporal image to obtain the preliminary micro-displacement field; Step S5: Using the internal bonding surface and edge bonding surface of the static reference image as a reference, the preliminary micro-displacement field is corrected to obtain the relative micro-displacement field. Step S6: Input the relative micro-displacement fields corresponding to the internal bonding surface and the edge bonding surface into the temporal anomaly detection model, and output the suitability score by the temporal anomaly detection model.
2. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps of step S1 include: Step S11: Use a high-speed intraoral scanner to acquire static baseline images of the patient in a non-occlusal state after implantation of the fixed prosthesis; Step S12: When the patient performs three simulated biting states—light biting, normal biting, and forceful biting—a high-speed intraoral scanner is used to acquire continuous time-series images of the simulated biting process.
3. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps of step S2 include: Step S21: Adaptive median filtering algorithm is used to remove noise, artifacts and interfering pixels from continuous time-series images and static reference images; Step S22: Using the trained CNN semantic segmentation network, segment the denoised continuous temporal images and the static reference images respectively; Step S23: Feature extraction and classification are performed using a CNN semantic segmentation network to separate the fixed prosthesis region and the abutment region.
4. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps for locking the internal bonding surface in step S3 include: Step S31: Extract the three-dimensional point cloud contours of the inner wall of the fixed restoration and the outer wall of the abutment respectively, and calculate the geometric center, principal direction vector and bounding box size; Step S32: Use the iterative nearest point algorithm to coarsely register the point cloud of the inner wall of the fixed restoration with the point cloud of the outer wall of the abutment; Step S33: For the region after coarse registration, construct a bidirectional topological adjacency graph between the inner wall of the fixed repair body and the outer wall of the abutment, and traverse the triangular facet matching relationships. Step S34: Identify a set of continuous facets where the angle between the facet normals is less than a preset angle threshold and the distance between facets is less than a preset bonding threshold, and lock the projection area of the set of continuous facets as the internal bonding surface.
5. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps for locking the edge bonding surface in step S3 include: Step S35: Extract the edge loop at the bottom of the fixed prosthesis and the edge loop at the corresponding position of the abutment, perform smooth fitting, and calculate the radial distance distribution of the fitted curve; Step S36: When the radial distance distribution falls within the preset threshold of the width of the annular sealing strip and the edge ring closure meets the requirements, lock the annular contact area between the bottom edge of the fixed repair body and the abutment as the edge contact surface.
6. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps of step S4 include: Step S41: Perform region cropping on adjacent consecutive frame images in the continuous temporal image to obtain pixel regions that only contain the internal bonding surface and the edge bonding surface; Step S42: Construct a sparse optical flow estimation algorithm. For pixel regions containing only the internal bonding surface, use the weighted least squares method to solve the optical flow equation and calculate the first displacement vector of each pixel point in the internal bonding surface between consecutive frame images. Step S43: For pixel regions containing only edge-fitting surfaces, use a circular window to perform optical flow calculations, track the displacement changes of pixels on the edge loop, and obtain the second displacement vector; Step S44: Integrate the first displacement vector and the second displacement vector according to the pixel coordinate order, and generate the preliminary micro-displacement field of the inner bonding surface and the preliminary micro-displacement field of the edge bonding surface respectively.
7. The method for assessing the fit of a fixed prosthesis based on image data according to claim 6, characterized in that, After calculating the first and second displacement vectors, an effectiveness filter is performed to remove abnormal vectors whose displacement amplitude exceeds a preset threshold and invalid vectors that fail to converge optical flow.
8. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps of step S5 include: Step S51: Establish a reference coordinate system based on the coordinates of all pixels of the inner bonding surface and the edge bonding surface in the static reference image; Step S52: Map the pixels of the internal and edge bonding surfaces corresponding to the continuous temporal images to the corresponding pixels of the static reference image in the reference coordinate system. Step S53: Based on coordinate mapping, using the fixed feature points at the connection between the abutment and the implant as a reference, identify and separate the rigid body displacement components of the abutment itself through feature point displacement analysis. Step S54: Based on the coordinate mapping relationship of the separated rigid body displacement components, perform coordinate transformation and displacement correction on the preliminary micro-displacement field to obtain the relative micro-displacement field.
9. The method for assessing the fit of a fixed prosthesis based on image data according to claim 1, characterized in that, The specific steps of step S6 include: Step S61: Normalize and feature-stitched the relative micro-displacement fields corresponding to the internal bonding surface and the edge bonding surface to generate a fused temporal feature vector. Step S62: Input the fused temporal feature vector into the trained temporal anomaly detection model, and the temporal anomaly detection model outputs the internal fitness score and the marginal fitness score respectively; Step S63: Perform a weighted fusion of the internal fitness score and the marginal fitness score to generate the overall fitness score.