Method for quantitatively evaluating degree of gingival recession by intraoral image and system thereof
By using a virtual cementoenamel boundary baseline and three-dimensional normal vector projection, the problems of starting point identification and perspective distortion in gingival recession assessment are solved, enabling accurate quantitative assessment of the degree of gingival recession, which is suitable for tracking the treatment efficacy and long-term monitoring of periodontal disease.
Patent Information
- Application Number
- CN202610802437.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-04
- Publication Date
- 2026-08-25
AI Technical Summary
Existing technologies for assessing gingival recession suffer from difficulties in identifying the cementoenamel boundary, scale blurring based on two-dimensional images, and perspective distortion under non-standard shooting angles, resulting in insufficient measurement accuracy and reliability.
By introducing a virtual cementoenamel boundary baseline generation mechanism, the starting point of gingival recession is reconstructed using a topological mapping algorithm. Combined with endogenous anatomical anchor points and three-dimensional normal vector projection, perspective distortion is eliminated, enabling accurate three-dimensional distance measurement.
It provides a reliable starting point reference in cases of cementoenamel boundary loss at the time of follow-up visit, eliminates scale ambiguity and perspective distortion, and improves the accuracy and applicability of gingival recession assessment. It is particularly suitable for periodontal disease treatment follow-up and long-term monitoring.
Smart Images

Figure CN122636563A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oral medical image analysis technology, and in particular to a method and system for quantitative assessment of the degree of gingival recession using intraoral images. Background Technology
[0002] Gingival recession refers to the receding of the gingival margin relative to the cementoenamel junction towards the tooth root. Accurate quantitative assessment of the degree of gingival recession is crucial for the early diagnosis of periodontal disease, treatment planning, efficacy monitoring, and long-term prognosis. In clinical practice, repeated measurements of gingival recession are often necessary; therefore, accurate, repeatable, and easy-to-use quantitative assessment methods are irreplaceable for standardized periodontal treatment procedures. Traditional gingival recession assessment methods primarily rely on manual measurements by clinicians using periodontal probes. While this contact-based manual measurement method has been used in periodontal treatment for a long time, it has many inherent limitations in practical clinical application. Due to these limitations, non-contact gingival recession assessment methods based on digital imaging have received widespread attention from both academia and clinicians in recent years. Acquiring patients' oral imaging data using image acquisition devices such as intraoral scanners, cone-beam computed tomography (CBCT) scanners, and digital cameras, and combining this with image analysis algorithms to achieve quantitative measurement of gingival morphology, has become a major research direction in this field. Compared with traditional manual measurement, this type of method has advantages such as being non-contact, repeatable, longitudinally traceable, and digitally archived, providing important technical support for the precision and digital transformation of periodontal diagnosis and treatment.
[0003] For example, Chinese invention patent CN119205680B discloses a method for precise quantitative assessment of periodontal disease based on IOS images and CBCT images, which includes the following steps: performing image segmentation on intraoral scan IOS images to obtain IOS segmentation results for each tooth; performing image segmentation on CBCT images to obtain CBCT segmentation results for each tooth; preprocessing and registering the IOS segmentation results and CBCT segmentation results to obtain a data fusion model; obtaining the gingival contour and tooth long axis based on the data fusion model; obtaining gingival contour points based on the gingival contour; obtaining the alveolar bone contour and GBD distance on the alveolar bone along the direction of the tooth long axis based on the gingival contour points; and assessing periodontal disease based on the GBD distance.
[0004] For example, Chinese invention patent CN111563887B discloses an intelligent analysis method and device for oral images, including an oral image acquisition device and a mobile terminal set with an oral image database, an oral fusion feature extraction module, a feature database, a target classifier module, and a tooth detection result output module. The method includes acquiring user oral images, extracting fusion features of teeth and gums from the oral images, using a single support vector machine classification model to evaluate the level of each tooth aesthetics detection target, outputting the classification results of each support vector machine classification model, and comprehensively calculating the oral aesthetics index from the classification results.
[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:
[0006] Existing technologies, while attempting to address some issues of non-contact measurement by introducing digital models, suffer from three main bottlenecks: First, the determination of the measurement starting point relies on the premise that the physical cementoenamel boundary is visible at the time of follow-up examination. However, in clinical practice, the cementoenamel boundary of the target tooth is often not directly identifiable during follow-up examinations due to factors such as tartar buildup, caries, restoration obstruction, or severe atrophy. This makes existing methods unsuitable for a large number of high-value clinical cases. Second, existing assessment methods based on two-dimensional intraoral images face a fundamental problem of scale ambiguity. Two-dimensional images lack depth information and absolute scale references. Existing methods either rely on external reference objects, increasing the complexity of the imaging operation, or can only output relative assessment results, failing to provide millimeter-level precision in atrophy measurement. Third, while existing assessment methods based on three-dimensional intraoral scanning can obtain high-precision results, intraoral scanning equipment is expensive. In scenarios where intraoral images are captured using ordinary digital cameras or smartphones, the shooting angle is often not strictly perpendicular to the measured tooth surface. The perspective distortion caused by oblique incidence results in a severe nonlinear deviation between the pixel distance measured in the image and the actual physical distance. Summary of the Invention
[0007] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for quantitative assessment of the degree of gingival recession using intraoral images. The technical solution is as follows:
[0008] Intraoral imaging methods for quantitative assessment of gingival recession include:
[0009] Historical intraoral images of the target tooth during the baseline period and current intraoral images during the follow-up period are extracted. Reference gingival morphological feature data of the target tooth are extracted from the historical intraoral images during the baseline period. The reference gingival morphological feature data includes the set of gingival margin apexes of the target tooth during the baseline period and the cementoenamel boundary curvature parameters of the target tooth during the baseline period.
[0010] Based on baseline gingival morphological feature data, the gingival morphological features in the historical intraoral images during the baseline period are mapped to the current intraoral images during the follow-up period. A virtual cementoenamel boundary baseline is generated in the current intraoral image of the target tooth during the follow-up period. The virtual cementoenamel boundary baseline serves as the starting point reference for gingival recession measurement.
[0011] At least one morphological constant is extracted from the current intraoral image of the target tooth during the follow-up period as an endogenous anatomical anchor point. The morphological constant is the anatomical geometry of the target tooth itself.
[0012] The standard physical dimension parameters that match the target tooth type are retrieved from a pre-built database of standard anatomical dimensions of teeth. The target tooth type is determined according to the general tooth position numbering method, and each tooth position type corresponds to a set of standard physical dimension parameters.
[0013] Using the image measurements of the endogenous anatomical anchor points and standard physical size parameters as constraints, the spatial pose matrix of the current shooting angle of the current intraoral image during the follow-up period is calculated, and a nonlinear pixel-to-physical distance mapping field for the local tooth surface of the target tooth is generated based on the spatial pose matrix.
[0014] The actual gingival margin of the target tooth is extracted from the current intraoral image during the follow-up visit using an image segmentation algorithm.
[0015] Based on the spatial pose matrix, the three-dimensional normal vector of the target tooth surface is calculated.
[0016] The virtual cementoenamel boundary baseline and the actual gingival margin are projected onto a projection plane orthogonal to the three-dimensional normal vector. Based on the pixel-to-physical distance mapping field, the vertical vector distance between the two projection lines is calculated. Multiple vertical vector distances constitute the local atrophy set of the target tooth. Based on the local atrophy set, the quantitative assessment value of gingival atrophy of the target tooth is output.
[0017] In addition, an intraoral imaging quantitative assessment system for the degree of gingival recession is provided, specifically including:
[0018] The data extraction module is used to extract historical intraoral images of the target tooth during the baseline period and current intraoral images during the follow-up period. The baseline gingival morphological feature data of the target tooth is extracted from the historical intraoral images during the baseline period. The baseline gingival morphological feature data includes the set of gingival margin apexes of the target tooth during the baseline period and the cementoenamel boundary curvature parameters of the target tooth during the baseline period.
[0019] The baseline generation module is used to map the gingival morphology features in the historical intraoral images during the baseline period to the current intraoral images during the follow-up period based on the baseline gingival morphology feature data. A virtual cementoenamel boundary baseline is generated in the current intraoral image of the target tooth during the follow-up period. The virtual cementoenamel boundary baseline serves as the starting point reference for gingival recession measurement.
[0020] The anchor point extraction module is used to extract at least one morphological constant from the current intraoral image of the target tooth during the follow-up period as an endogenous anatomical anchor point. The morphological constant is the anatomical geometry of the target tooth itself.
[0021] The parameter query module is used to query the standard physical size parameters that match the target tooth type from a pre-built database of standard anatomical dimensions of teeth. The target tooth type is determined according to the general tooth position numbering method, and each tooth position type corresponds to a set of standard physical size parameters.
[0022] The data processing module is used to calculate the spatial pose matrix of the current shooting angle of the current intraoral image during the follow-up period, using the image measurement values of the endogenous anatomical anchor points and standard physical size parameters as constraints, and to generate a nonlinear pixel-to-physical distance mapping field for the local tooth surface of the target tooth based on the spatial pose matrix.
[0023] The edge line extraction module is used to extract the actual gingival edge line of the target tooth from the current intraoral image during the follow-up visit using an image segmentation algorithm.
[0024] The normal vector calculation module is used to calculate the three-dimensional normal vector of the target tooth surface based on the spatial pose matrix.
[0025] The results output module projects the virtual cementoenamel boundary baseline and the actual gingival margin line onto a projection plane orthogonal to the three-dimensional normal vector, respectively. Based on the pixel-to-physical distance mapping field, it calculates the vertical vector distance between the two projection lines. Multiple vertical vector distances constitute a set of local atrophy amounts of the target tooth. Based on the set of local atrophy amounts, it outputs a quantitative assessment value of gingival atrophy of the target tooth.
[0026] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:
[0027] 1. The quantitative assessment method for gingival recession using intraoral images provided by this invention solves the problem of lost starting point for gingival recession measurement when the physical cementoenamel boundary is missing or unidentifiable at the time of follow-up visit. This is achieved by introducing a virtual cementoenamel boundary baseline generation mechanism based on cross-period topological mapping between historical intraoral images of the target tooth during the baseline period and the current intraoral image during the follow-up visit. In clinical practice, the cementoenamel boundary of the target tooth is often not directly identifiable during follow-up visits due to tartar buildup, caries, restoration obstruction, or severe recession. This invention fully utilizes the consistency and temporal continuity of the tooth position and anatomical structure of the same target tooth between the baseline and follow-up periods. Through a topological mapping algorithm, the gingival margin morphology and cementoenamel boundary curvature information of the target tooth in the historical intraoral images of the baseline period are transferred to the image region of the target tooth in the current intraoral image during the follow-up visit. A virtual cementoenamel boundary baseline corresponding to the anatomical structure of the same target tooth in the baseline period is reconstructed in the two-dimensional image space, thus providing a reliable starting point reference for the measurement of recession during the follow-up visit. Meanwhile, by introducing multiple safeguard mechanisms such as bending energy constraint, anatomical rationality verification, pull-back adjustment, and dynamic re-screening of historical intraoral images from the baseline period, the risk of the mapping output deviating from the anatomical pattern of the baseline period in terms of morphology is effectively avoided. This significantly expands the clinical applicability of quantitative assessment methods for gingival recession, and is particularly suitable for clinical application scenarios that require before-and-after comparisons, such as periodontal disease efficacy tracking, longitudinal efficacy assessment, and long-term follow-up monitoring.
[0028] 2. This invention solves the scale ambiguity problem caused by the lack of physical scale reference in two-dimensional intraoral images by constructing a pose calculation algorithm based on dual constraints of endogenous anatomical anchor points and a priori database. Two-dimensional images themselves do not contain depth information or absolute scale references. This invention utilizes the morphological constant of the target tooth itself, which is relatively stable across individuals and unaffected by gingival recession, as an endogenous anatomical anchor point, ensuring that the scale reference originates from the target tooth itself without any external reference. Simultaneously, a priori database of standard anatomical dimensions, categorized by tooth position type, is introduced to provide reasonable constraints on the true physical dimensions of the anchor points. These two types of constraints complement each other—the endogenous anchor point provides precise single-point scale alignment, and the priori database provides a statistically reasonable boundary for the overall scale—effectively overcoming the scale ambiguity problem in monocular depth estimation. This ensures that the calculated spatial pose matrix and pixel-to-physical distance mapping field possess absolute physical scale, significantly reducing the equipment and operational barriers to accurate quantitative evaluation.
[0029] 3. This invention effectively eliminates the nonlinear measurement deviation caused by perspective distortion under non-standard shooting angles by designing an orthogonal projection plane atrophy measurement mechanism based on three-dimensional normal vector guidance. In scenarios where intraoral images are captured using ordinary digital cameras or smartphones, the limited space of the oral cavity and the ease of operation often prevent the shooting angle from being strictly perpendicular to the tooth surface being measured. Perspective distortion caused by oblique incidence results in a severe nonlinear deviation between the measured pixel distance in the image and the actual physical distance, and the degree of deviation exhibits a complex, non-uniform distribution depending on the shooting angle. This invention, by inverting the three-dimensional normal vector of the target tooth surface based on the spatial pose matrix, constructs a standardized projection plane orthogonal to the normal vector. The virtual cementoenamel boundary baseline and the actual gingival margin are projected onto the same standardized plane for vertical vector distance measurement. This ensures that the atrophy measurement always occurs in a standard coordinate system parallel to the tooth surface, eliminating the perspective compression effect caused by the tilted shooting angle. Meanwhile, by performing fine-grained scale correction on the distance calculation within the projection plane through a nonlinear pixel-to-physical distance mapping field, the measurement results are ensured to reflect the real physical distance rather than the image pixel distance. This allows the invention to maintain stable measurement accuracy even in clinical practice scenarios with a high degree of freedom in shooting angles, significantly improving its adaptability to natural shooting conditions. Attached Figure Description
[0030] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0031] Figure 1 This is a schematic diagram of the method for quantitative assessment of the degree of gingival recession using intraoral images, as provided in the embodiments of this application.
[0032] Figure 2 A schematic diagram of the intraoral image quantitative assessment system module for the degree of gingival recession provided in this application embodiment.
[0033] Figure 3 This is a schematic diagram of the logic flow for quantitative assessment of the degree of gingival recession using intraoral images, as provided in the embodiments of this application. Detailed Implementation
[0034] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present disclosure are shown in the drawings, it should be understood that embodiments of the present disclosure may be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure.
[0035] It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure. In the description of the embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "this embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects.
[0036] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0037] Please see Figure 1 The image shows a quantitative assessment method for the degree of gingival recession using intraoral images, which specifically includes:
[0038] like Figure 3 The diagram illustrates the logical flow of quantitative assessment of gingival recession using intraoral images provided in this embodiment. It includes an intraoral image input stage, a baseline gingival morphological feature data extraction stage, a virtual cementoenamel boundary baseline generation stage, a target tooth morphological constant extraction stage and a standard anatomical dimension prior database query stage, a target tooth actual gingival margin extraction stage, a monocular depth estimation and spatial pose matrix calculation stage, a target tooth surface three-dimensional normal vector calculation stage, an orthogonal projection stage of the virtual cementoenamel boundary baseline and the actual gingival margin, a vertical vector distance calculation and local recession set construction stage, and a quantitative assessment value output stage for gingival recession. Solid arrows indicate the main flow data transfer relationship between each stage, while dashed arrows indicate the feedback calibration and data import relationship. The overall architecture of this embodiment is presented intuitively, clearly expressing the logical sequence between each stage.
[0039] The baseline historical intraoral image and the current intraoral image during the follow-up visit of the target tooth are extracted from the intraoral images. The baseline gingival morphological feature data includes the set of vertices and curvature parameters of the gingival margin during the baseline period.
[0040] When multiple baseline intraoral images of the target tooth are available, at least one image is selected as the valid baseline intraoral image based on the principles of closest acquisition time, intact cementoenamel junction visible in the baseline image, and uninterrupted gingival margin of the target tooth in the baseline image. It should be noted that in actual clinical practice, multiple candidate baseline intraoral images of the same target tooth may exist in the patient's past medical records. These candidate baseline intraoral images may differ in terms of acquisition time, shooting angle, lighting conditions, and the anatomical integrity of the target tooth.
[0041] The sources of selectable baseline intraoral images include, but are not limited to: intraoral photographs from each follow-up visit archived in the hospital's electronic medical record system, self-taken intraoral photographs uploaded by the target patient in home oral health monitoring applications, and intraoral imaging data retained by the target patient at other dental clinics and accessed with authorization. Before the screening begins, the system first collects all historical intraoral images from the above sources whose tooth position numbers match the target tooth, forming a candidate baseline historical intraoral image set.
[0042] The principle of "closest acquisition time for baseline period" means that, among the candidate set of historical intraoral images for baseline period, the historical intraoral image whose acquisition time is closest to the current follow-up visit time is preferentially selected as the valid historical intraoral image for baseline period. It should be noted that the principle of "closest acquisition time" must be used in conjunction with a lower time limit threshold to avoid selecting baseline images and follow-up visit images that are too close in time (e.g., less than a week apart), resulting in the difference in atrophy between the two periods being lower than the measurement accuracy and losing the comparative significance. In this embodiment, the minimum time interval between the baseline period and the follow-up visit time is preset to one month, meaning that only historical intraoral images acquired at least one month before the follow-up visit time will be included in the candidate set. Simultaneously, to avoid excessively large differences in anatomical structures due to excessively long acquisition times for baseline images, this embodiment presets the maximum time interval between the baseline period and the follow-up visit time to three years, meaning that historical intraoral images acquired more than three years before the follow-up visit time are no longer considered as priority candidates.
[0043] The principle of complete visibility of the cementoenamel boundary in baseline images specifically means that, in the candidate set of historical intraoral images from the baseline period, the cementoenamel boundary curve of the target tooth in the selected historical intraoral images must be completely identifiable on the buccal or labial side and not covered by calculus, restorations, caries, or other obstructions. The specific determination method for complete visibility of the cementoenamel boundary is as follows: For each candidate historical intraoral image from the baseline period, a pre-trained image segmentation algorithm is used to identify the enamel and root regions of the target tooth, and the continuous visible length of the boundary line between the two regions is detected on the image. When the continuous visible length is not less than half of the maximum mesiodistal width of the crown of the target tooth, the candidate historical intraoral image from the baseline period is deemed to satisfy the principle of complete visibility of the cementoenamel boundary; otherwise, it is deemed not to satisfy the principle.
[0044] The principle of uninterrupted gingival margin in the baseline image of the target tooth specifically means that, in the candidate historical intraoral image set of the baseline period, the gingival margin curve of the target tooth in the selected historical intraoral images must be a continuous and uninterrupted contour line on the buccal or labial side, without any gingival margin breakage, loss, or significant distortion caused by inflammation, edema, trauma, or artificial occlusion. The specific method for determining uninterrupted gingival margin is as follows: For each candidate historical intraoral image of the baseline period, the boundary contour between the tooth region and the gingival region of the target tooth is identified using an image segmentation algorithm. Pixel-level interruptions or significant gaps are detected along the horizontal direction on this boundary contour. When the cumulative length of the interruption or gap does not exceed one-tenth of the maximum mesiodistal width of the target tooth crown, the candidate historical intraoral image of the baseline period is deemed to meet the principle of uninterrupted gingival margin; otherwise, it is deemed not to meet the principle. In this embodiment, the cumulative interruption length ratio threshold for uninterrupted gingival margin is preset to 0.1, which can be adjusted within the range of 0.05 to 0.2 according to actual clinical conditions.
[0045] It should be noted that the target tooth refers to a tooth with or suspected of having gingival recession that requires quantitative assessment. The cementoenamel boundary is the anatomical boundary between the crown and root, and clinically, this boundary is usually used as the starting point for assessing the degree of gingival recession. The mesial side refers to the side of the target tooth facing the midline of the dental arch, and the distal side refers to the side of the target tooth away from the midline of the dental arch. The baseline intraoral image refers to the intraoral image of the target tooth taken at the time of the previous visit, serving as the morphological reference baseline for subsequent recession assessment; the current intraoral image at the follow-up visit refers to the intraoral image of the target tooth taken at the time of the current follow-up visit. The images taken at the baseline and follow-up visits are of the same target tooth, and its tooth position, crown morphology, and other anatomical features remain consistent, providing a natural basis for cross-period topological mapping.
[0046] The extraction method for the gingival margin vertex set of the target tooth in the baseline period is as follows: Within the image region of the target tooth in the historical intraoral image of the baseline period, the gingival-teeth boundary contour of the target tooth in the baseline period is sampled at equal intervals along the horizontal direction. The image coordinates of each sampling point are used as a gingival margin vertex, and all sampling points constitute the gingival margin vertex set of the target tooth in the baseline period. The number of sampling points is preset to be no less than twenty. The extraction method for the cementoenamel boundary curvature parameter of the target tooth in the baseline period is as follows: The cementoenamel boundary curve of the target tooth in the historical intraoral image of the baseline period is identified. After smoothing and filtering the curve, its local curvature is calculated point by point. All local curvature values constitute the cementoenamel boundary curvature parameter of the target tooth in the baseline period.
[0047] Based on baseline gingival morphological data, a topological mapping algorithm is used to map historical intraoral images from the baseline period and current intraoral images from the follow-up period to the current intraoral image from the follow-up period of the target tooth. A virtual cementoenamel boundary baseline is generated in the current intraoral image from the follow-up period of the target tooth. The virtual cementoenamel boundary baseline serves as the starting point reference for gingival recession measurement.
[0048] For historical intraoral images from the baseline period, the set of gingival margin vertices is extracted as the first control point set, and the equidistant sampling points on the cementoenamel boundary curve are extracted as the second control point set. The first and second control point sets together constitute the source domain control point set.
[0049] Topological mapping refers to a nonlinear transformation process that transfers the morphological features of one image region to another image region while preserving neighborhood relationships and spatial continuity. The source domain control point set is the input to the mapping transformation, provided by the image region of the target tooth in the historical intraoral images of the baseline period; the target domain control point set is the output of the mapping transformation, located within the image region of the target tooth in the current intraoral image of the follow-up visit. The first control point set describes the gingival margin morphology of the target tooth in the baseline period, and the second control point set describes the cementoenamel boundary morphology of the target tooth in the baseline period. There is a fixed vertical positional relationship between the two control point sets in the baseline image. This vertical positional relationship implies the anatomical positional pattern of the gingival margin relative to the cementoenamel boundary in the baseline period (healthy reference state). This pattern, after being mapped across periods to the follow-up visit, serves as a positional reference for the virtual cementoenamel boundary baseline in the follow-up visit relative to the actual gingival margin in the follow-up visit.
[0050] The first set of control points describes the gingival margin morphology in the baseline historical intraoral images, while the second set of control points describes the cementoenamel boundary morphology in the baseline historical intraoral images. There is a fixed longitudinal positional relationship between the two sets of control points in adjacent tooth images. This longitudinal positional relationship implies the anatomical positional pattern of the lower gingival margin relative to the cementoenamel boundary in the baseline historical intraoral images.
[0051] During the follow-up period of the target tooth, in the current intraoral image, the candidate positions of target domain control points corresponding to the source domain control point set are determined based on the crown contour of the target tooth, thus obtaining the target domain control point set. The specific method is as follows:
[0052] In the current intraoral image during the follow-up period of the target tooth, the crown contour of the target tooth is identified and used as the geometric reference of the target domain. The crown contour of the historical intraoral image during the baseline period is registered and aligned with the crown contour of the target tooth. Based on the positional relationship between the source domain control points and the source domain crown contour, the corresponding positions of the target domain control points relative to the target domain crown contour are deduced. The corresponding positions are the candidate positions of the target domain control points.
[0053] A continuous deformation function from the source domain control point set to the target domain control point set is constructed using a thin plate spline interpolation algorithm. The continuous deformation function includes affine transformation components and nonlinear bending components.
[0054] Thin-plate spline interpolation refers to the mathematical function corresponding to the minimum bending energy surface naturally formed by an infinitely thin elastic metal plate after forced displacement at multiple fixed points. The affine transformation component describes the overall translation, rotation, and scaling transformation from the source domain to the target domain, reflecting the overall geometric positional differences of the same target tooth between the baseline and follow-up periods caused by differences in shooting angle and distance. The nonlinear bending component describes the local surface bending transformation outside of the affine transformation, reflecting the local morphological differences of the target tooth between the two periods due to pathological changes such as gingival recession and tooth wear.
[0055] Calculate the bending energy value of the continuous deformation function and compare it with the upper limit of bending energy. When the bending energy value is higher than the upper limit of bending energy, it is determined that the mapping result does not meet the anatomy rationality. Expand the sampling density of the source domain control point set and reconstruct the continuous deformation function until the bending energy value is lower than the upper limit of bending energy.
[0056] The bending energy value is the integral of the square of the second derivative of the continuous deformation function over the entire mapping region, reflecting the degree of nonlinear bending of the deformation function. The higher the bending energy value, the more severe the local distortion introduced by the deformation function, which may produce a virtual baseline that does not conform to the anatomical laws of teeth.
[0057] The method for obtaining the upper limit of bending energy includes: collecting no less than one hundred clinical healthy dental arch samples, performing thin plate spline mapping between adjacent teeth for each sample and calculating the bending energy value. In this embodiment, the 95th percentile of the statistical distribution of all bending energy values is taken as the upper limit of bending energy.
[0058] When the bending energy value is higher than the upper limit of bending energy, a new sampling point is inserted between adjacent sampling points in the first control point set and the second control point set to halve the sampling interval and reconstruct the continuous deformation function. If the bending energy value is still higher than the upper limit, the sampling interval halving operation is repeated until the bending energy value drops below the upper limit.
[0059] Applying a continuous deformation function to the cementoenamel boundary curve of the historical intraoral image during the baseline period yields the virtual cementoenamel boundary baseline in the current intraoral image of the target tooth during the follow-up period.
[0060] The converged continuous deformation function is applied to all sampling points on the cementoenamel boundary curve of the historical intraoral image during the baseline period. The corresponding coordinates of these sampling points in the current intraoral image area during the follow-up period of the target tooth are obtained. All corresponding coordinates are connected in the original order to obtain a continuous curve, which is the virtual cementoenamel boundary baseline.
[0061] Obtaining the virtual cementoenamel boundary baseline within the current intraoral image during the follow-up period of the target tooth also includes verifying the anatomical rationality of the virtual cementoenamel boundary baseline, specifically including:
[0062] Several calibration points were uniformly sampled along the virtual cementoenamel boundary baseline. The local curvature value of each calibration point was calculated and compared with the upper and lower limits of the standard curvature, respectively. The upper and lower limits of the standard curvature were obtained from the prior database of standard anatomical dimensions of teeth according to the target tooth type.
[0063] Anatomical rationality verification refers to applying secondary constraints to the output of thin plate spline mapping to ensure that the morphology of the virtual cementoenamel boundary baseline conforms to the anatomical rules of the target tooth type.
[0064] The uniform sampling method for the verification points is as follows: sampling is carried out along the virtual cementoenamel boundary baseline at fixed arc length intervals, and the number of verification points on each baseline is preset to be no less than thirty.
[0065] The local curvature value is calculated as follows: for each verification point, take two adjacent verification points before and after it to form a five-point window, fit a quadratic curve to the five-point window, and the curvature of the quadratic curve at the verification point is the local curvature value of the verification point.
[0066] A priori database of standard anatomical dimensions of teeth refers to a pre-constructed dataset that stores the statistical distribution of anatomical parameters for different tooth types. Tooth types are classified according to a common tooth position numbering method, and each tooth position type corresponds to a set of anatomical parameters.
[0067] The process for setting the upper and lower limits of standard curvature is as follows: For each tooth position type, collect no less than 300 clinical healthy samples, extract the local curvature on the cementoenamel boundary curve for each sample, and statistically analyze the local curvature distribution of all sites in all samples. In this embodiment, the 97.5th percentile of the local curvature distribution is used as the upper limit of standard curvature, and the 2.5th percentile of the local curvature distribution is used as the lower limit of standard curvature.
[0068] When the local curvature value falls between the upper and lower limits of the standard curvature, the verification point is determined to be a reasonable point for dissection, and the position coordinates of the verification point are retained.
[0069] When the local curvature value is higher than the upper limit of the standard curvature, the verification point is determined to be an excessive curvature point. The local curvature value of the verification point is then pulled back to the upper limit of the standard curvature. The pullback range is proportional to the difference between the local curvature value and the upper limit of the standard curvature.
[0070] When the local curvature value is lower than the standard curvature lower limit, the verification point is determined to be an excessively flat point. The local curvature value of the verification point is then pulled back to the standard curvature lower limit. The pullback range is proportional to the difference between the local curvature value and the standard curvature lower limit.
[0071] The specific execution method of pullback adjustment is as follows: For excessively curved points, subtract a pullback amount from their local curvature value. The pullback amount is equal to the difference between the local curvature value of the verification point and the upper limit of the standard curvature multiplied by the pullback coefficient.
[0072] For excessively flat points, add a pullback amount to their local curvature value. The pullback amount is equal to the difference between the standard curvature lower limit and the local curvature value of the verification point multiplied by the pullback coefficient.
[0073] The process for setting the pull-back coefficient is as follows: Through offline experiments, a group of clinical samples with known actual cementoenamel junction positions were pulled back and adjusted within the range of 0.5 to 0.9 to find the pull-back coefficient value that minimizes the error between the pulled-back position and the actual position. In this embodiment, the pull-back coefficient is preset to 0.8. Directly modifying the local curvature value would disrupt the continuity of the curve.
[0074] The actual adjustment method for the position of the verification point is as follows: the local curvature value after the pull-back is used to back-calculate the image coordinates where the verification point should be located through curve integration, and the curve segments between adjacent verification points are smoothed to ensure that the entire virtual baseline remains continuous and smooth after adjustment.
[0075] The proportion of excessively curved and excessively straight points among all verification points is counted. When the proportion exceeds the threshold for anatomical deviation, it is determined that the selection of historical intraoral images in the baseline period is inappropriate, triggering a re-screening of historical intraoral images in the baseline period.
[0076] The virtual cementoenamel boundary baseline, adjusted after pulling back from all verification points, is used as the final output virtual cementoenamel boundary baseline.
[0077] When the proportion of excessively curved and excessively flat points is low, it indicates that the historical intraoral images from the baseline period can provide a reasonable morphological reference, with only a few check points showing deviations, which can be resolved by adjusting the images back. When the proportion is high, it indicates that the historical intraoral images from the baseline period as a whole cannot provide a reasonable morphological reference for the current intraoral images during the follow-up visit. For example, the time between the baseline period and the follow-up visit may be too long, resulting in significant changes in anatomical structures during that period. Alternatively, the historical intraoral images from the baseline period may have problems such as insufficient image quality, blurry images, or severe occlusion. In this case, adjusting the images back is insufficient, and the historical intraoral images from the baseline period must be re-screened.
[0078] The process for setting the anatomical deviation percentage threshold is as follows: a mapping experiment is conducted on no less than fifty clinical dental arch samples. The maximum percentage of excessively curved points and excessively straight points in the successfully mapped cases is counted. This maximum percentage is multiplied by 1.2 to obtain the anatomical deviation percentage threshold. In this embodiment, the anatomical deviation percentage threshold is set to 30%.
[0079] Expanding the scope of health assessment criteria includes: broadening the search range for historical intraoral images during the baseline period. Specifically, this involves extending the search from the most recent historical acquisition to images acquired at even earlier times. If the patient has multiple historical acquisition images stored in the system, priority is given to historical images acquired more recently than the follow-up visit and meeting the image quality requirements. If a valid baseline image still cannot be obtained, the user is prompted to acquire additional baseline images or to use other assessment methods.
[0080] At least one morphological constant is extracted from the current intraoral image of the target tooth during the follow-up period as an endogenous anatomical anchor point. The morphological constant is the anatomical geometry of the target tooth itself.
[0081] Multiple candidate morphological constants were extracted simultaneously from the current intraoral images of the target tooth during the follow-up period. The candidate morphological constants included the maximum mesiodistal width of the crown of the target tooth, the interdental papilla base distance between the target tooth and the mesial adjacent tooth, the interdental papilla base distance between the target tooth and the distal adjacent tooth, and the vertical height of the crown of the target tooth.
[0082] It should be noted that morphological constants refer to the anatomical and geometrical quantities of the target tooth that are relatively stable across individuals and are not affected by gingival recession.
[0083] The maximum mesiodistal width of a tooth crown refers to the maximum horizontal distance of the crown in the mesiodistal direction, measured from the outermost apexes of the mesial and distal contact areas. The interdental papilla base distance refers to the horizontal distance between the bases of the gingival papillae of two adjacent teeth. The vertical height of a tooth crown refers to the vertical distance from the highest point of the cusp to the lowest point of the gingival margin on the occlusal surface.
[0084] For each candidate morphological constant, its identifiability confidence in the image is calculated. The identifiability confidence is determined by the geometric mean of the image segmentation confidence of the boundary points at both ends of the morphological constant.
[0085] The identifiability confidence of each candidate morphological constant is compared with the confidence threshold. Candidate morphological constants with an identifiability confidence higher than the confidence threshold are identified as valid morphological constants, while candidate morphological constants with an identifiability confidence lower than the confidence threshold are eliminated.
[0086] The specific calculation method for the identifiability confidence is as follows: Each candidate morphological constant is defined by two endpoints. For example, the maximum width of a tooth crown in the mesial-distal direction is defined by the mesial endpoint and the distal endpoint. The image segmentation algorithm outputs the probability value of each pixel belonging to the category while outputting the pixel-level classification result. The probability value of the category to which the endpoint belongs at each endpoint is used as the image segmentation confidence of that endpoint. The geometric mean of the image segmentation confidence of the two endpoints of a morphological constant is calculated. The geometric mean is the identifiability confidence of the morphological constant.
[0087] The process of setting the confidence threshold is as follows: collect no less than 200 clinical intraoral image samples with different lighting, angles and blur levels. For each sample, human experts mark which morphological constants can be accurately identified. Statistically calculate the identifiability confidence distribution of the accurately identifiable morphological constants. In this embodiment, the 10th percentile of this distribution is used as the confidence threshold. In this embodiment, the confidence threshold is preset to 0.7.
[0088] When the number of valid morphological constants falls below the lower limit of the number of anchor points, the confidence threshold is lowered and the data is re-filtered until the number of valid morphological constants reaches the lower limit of the number of anchor points.
[0089] When the number of valid morphological constants is higher than or equal to the lower limit of the number of anchor points, all valid morphological constants are output as endogenous anatomical anchor points, and the identifiability confidence of each endogenous anatomical anchor point is used as the weighting weight of each constraint in the subsequent pose calculation algorithm.
[0090] The process of setting the lower limit of the number of anchor points is as follows: Through offline experiments, the accuracy of monocular depth estimation under different numbers of anchor points was tested. It was found that when the number of anchor points is less than 2, there is serious scale ambiguity in the pose matrix. In this embodiment, the lower limit of the number of anchor points is set to 2.
[0091] The confidence threshold is lowered by decreasing it in a preset step size of 10% of the initial confidence threshold each time. After each reduction, the threshold is re-filtered. If the minimum anchor point count is still not met, the threshold is lowered further until it reaches a preset minimum protection value. The purpose of setting this minimum protection value is to prevent excessively lowering the threshold from including obviously erroneous morphological constants in the anchor point set. In this embodiment, the minimum protection value is preset to 0.4. If the minimum anchor point count is still not met even after lowering the threshold, the current image is determined to lack depth estimation capabilities, and the user is prompted to retake the image.
[0092] The method of passing the identifiable confidence as a weight is as follows: the identifiable confidence of each endogenous anatomical anchor point is normalized to the interval [0, 1] and then used as the weight coefficient of each constraint in the subsequent pose calculation algorithm. The anchor point with higher identifiable confidence contributes more to the joint loss function, so that the pose calculation result is more likely to be dominated by reliable anchor points.
[0093] Retrieve standard physical dimension parameters that match the target tooth type from a pre-built database of standard anatomical dimensions of teeth.
[0094] In this embodiment of the invention, the pose calculation algorithm is specifically a monocular depth estimation algorithm. It uses the image measurement values of the endogenous anatomical anchor points extracted from the current intraoral image during the follow-up visit and the standard physical size parameters queried from the prior database of standard anatomical dimensions of teeth as constraints to calculate the spatial pose matrix of the shooting perspective of the current intraoral image during the follow-up visit, and generates a nonlinear pixel-to-physical distance mapping field for the local tooth surface of the target tooth based on the spatial pose matrix.
[0095] It should be noted that monocular depth estimation refers to the algorithmic process of retrieving three-dimensional spatial depth information based solely on a single two-dimensional image. The spatial pose matrix is a mathematical transformation matrix describing the position and orientation of the camera relative to the target tooth, including three translational components and three rotational components. Once the spatial pose matrix is determined, each pixel in the image can be back-projected to its corresponding position in three-dimensional physical space.
[0096] The standard physical dimensions of the target tooth type are retrieved from the prior database of standard anatomical dimensions of teeth. The standard physical dimensions are expressed as a statistical distribution consisting of the mean and standard deviation.
[0097] For each endogenous anatomical anchor point, a scale consistency constraint term is constructed. The scale consistency constraint term is the difference between the physical measurement value obtained by back-projecting the image measurement value of the endogenous anatomical anchor point through the spatial pose matrix and the mean value of the standard physical size parameters.
[0098] For each endogenous anatomical anchor point, a distribution rationality constraint term is constructed. The distribution rationality constraint term is the ratio between the distance of the physical measurement value from the mean of the standard physical dimensional parameter and the standard deviation.
[0099] Standard physical dimensional parameters are represented by a statistical distribution of mean and standard deviation because there are natural dimensional fluctuations among individuals for the same tooth position type, and a single fixed value cannot characterize these fluctuations. The process for setting standard physical dimensional parameters is as follows: at least three thousand clinical healthy samples are collected for each tooth position type; the actual physical dimensions of each morphological constant are measured using a 3D scanning device; and the mean and standard deviation of all measurements are calculated to constitute the standard physical dimensional parameters of that morphological constant for that tooth position type.
[0100] The joint constraint loss function is obtained by weighting the sum of squares of the scale consistency constraint and the sum of squares of the distribution rationality constraint. The weights are determined by the identifiability confidence of the endogenous anatomical anchors.
[0101] The initial solution of the spatial pose matrix is obtained by minimizing the joint constraint loss function through an iterative optimization method. The spatial pose matrix includes rotation and translation components.
[0102] Calculate the deviation of the physical measurement values of all endogenous anatomical anchor points corresponding to the initial solution from the standard physical size parameters. Compare the maximum deviation with the upper limit of deviation. When the maximum deviation is higher than the upper limit of deviation, expand the search range of the pose calculation algorithm and iterate again until the maximum deviation is lower than the upper limit of deviation.
[0103] The final converged spatial pose matrix is output as the spatial pose matrix of the current shooting perspective.
[0104] The specific implementation of the iterative optimization method is as follows: a pre-trained monocular depth estimation neural network is used to perform preliminary depth inference on the current intraoral image of the target tooth during the follow-up period to obtain a coarse solution of the pose matrix. The coarse solution is used as the starting point of iterative optimization. The six components of the pose matrix are gradually adjusted using the Levenberg-Marquardt algorithm. After each iteration, the joint constraint loss function is recalculated. When the decrease in the loss function of two consecutive iterations is lower than the preset convergence threshold, it is determined to be converged, and the current pose matrix is output as the initial solution.
[0105] The specific calculation method for deviation is as follows: for each endogenous anatomical anchor point, its deviation is defined as the absolute value of the difference between the physical measurement value obtained by back projection of the anchor point and the mean value of the standard physical size parameter, divided by the standard deviation of the standard physical size parameter.
[0106] In this embodiment of the invention, the process of setting the upper limit of deviation is as follows: the distribution of deviation under normal calculation conditions is statistically analyzed through clinical trials, and the 95th percentile of the distribution is multiplied by 1.2 times to obtain the upper limit of deviation. In this embodiment, the upper limit of deviation is set to 3.
[0107] The search range is expanded by doubling the radius of the reliable region of the Levenburg-Marquardt algorithm and re-executing the iterative optimization. If it still fails to converge, a new starting point is randomly generated in a larger pose space for re-optimization to avoid getting trapped in local optima.
[0108] Based on the aforementioned spatial pose matrix calculation, a feedback calibration mechanism between the spatial pose matrix and the virtual cementoenamel boundary baseline can be established to improve overall accuracy. The purpose of this calibration mechanism is that the generation of the virtual cementoenamel boundary baseline depends on cross-period topological mapping, while the calculation of the spatial pose matrix depends on endogenous anatomical anchor points and a priori database in the current intraoral images during follow-up visits. The input information sources for these two processes differ, and independent deviations may occur due to differences in imaging conditions between the baseline and follow-up periods, anatomical changes in the target tooth, etc. By jointly verifying the outputs of the two processes in three-dimensional physical space, inconsistencies can be identified and corrected. The specific process of this mechanism includes:
[0109] Based on the spatial pose matrix, the virtual cementoenamel boundary baseline is back-projected onto the three-dimensional physical space to obtain the three-dimensional virtual cementoenamel boundary curve.
[0110] The standard longitudinal position range of the cementoenamel boundary is retrieved from the prior database of standard anatomical dimensions of teeth for the target tooth type. The standard longitudinal position range of the cementoenamel boundary is expressed as the longitudinal distance between the maximum crown outline and the cementoenamel boundary.
[0111] The actual longitudinal distance is obtained by calculating the longitudinal distance between the three-dimensional virtual cementoenamel boundary curve and the maximum outline of the crown of the target tooth in three-dimensional physical space.
[0112] It should be noted that the purpose of the feedback calibration mechanism is as follows: the generation of the virtual cementoenamel boundary baseline in two-dimensional image space depends on thin-plate spline mapping, while the calculation of the spatial pose matrix depends on endogenous anatomical anchor points and a priori database. These two processes are relatively independent. When the outputs of the two processes are jointly checked in three-dimensional space, if contradictions are found, it indicates that at least one process has a deviation, requiring feedback calibration.
[0113] The pre-determined process for the standard longitudinal position range of the cementoenamel boundary is as follows: collect no less than 300 clinical healthy samples for each tooth position type, measure the longitudinal distance of the maximum crown contour line relative to the cementoenamel boundary for each sample, obtain the statistical distribution of this distance, and use the 2.5% quantile and 97.5% quantile of this distribution as the lower and upper bounds of the standard longitudinal position range of the cementoenamel boundary.
[0114] Identify the maximum crown contour of the target tooth in three-dimensional physical space. The maximum crown contour is the curve corresponding to the widest cross section of the crown in three-dimensional space. Calculate the distance from each point on the three-dimensional virtual cementoenamel boundary curve to the maximum crown contour. The average of the distances is the actual longitudinal distance.
[0115] When the actual longitudinal distance falls within the standard longitudinal position range of the cementoenamel boundary, the virtual cementoenamel boundary baseline is determined to be consistent with the spatial pose matrix, and no calibration is required.
[0116] When the actual longitudinal distance falls outside the standard longitudinal position range of the cementoenamel boundary, it is determined that there is a deviation in the virtual cementoenamel boundary baseline or the spatial pose matrix, triggering feedback calibration: calculate the deviation between the actual longitudinal distance and the standard longitudinal position range of the cementoenamel boundary, and allocate the deviation to the position adjustment of the virtual cementoenamel boundary baseline and the attitude adjustment of the spatial pose matrix. The allocation ratio is determined by the reciprocal of the solution confidence of each.
[0117] The calibrated virtual cementoenamel boundary baseline and the calibrated spatial pose matrix are applied to subsequent measurement steps.
[0118] The solution confidence level refers to the quantitative evaluation value of the reliability of the output of each module. The solution confidence level of the virtual cementoenamel boundary baseline is jointly determined by the reciprocal of the bending energy value of the thin plate spline mapping and the pass rate of the anatomical rationality verification; the solution confidence level of the spatial pose matrix is jointly determined by the mean confidence level of the identifiable endogenous anatomical anchor points and the convergence speed of the iterative optimization.
[0119] A module with a higher confidence level indicates that its output is more reliable and should be subject to less calibration; a module with a lower confidence level indicates that its output is more likely to be biased and should be subject to more calibration.
[0120] The specific execution method for adjusting the virtual cementoenamel boundary baseline position is as follows: the virtual cementoenamel boundary baseline is translated as a whole along the vertical height direction of the crown, and the translation amount is equal to its allocated calibration amount. The specific execution method for adjusting the spatial pose matrix posture is as follows: the rotation component of the spatial pose matrix is slightly perturbed according to its allocated calibration amount, and then an iterative optimization is performed again. After calibration, the actual longitudinal distance is recalculated. If it still falls outside the standard range, the calibration process is repeated, up to three times. If it still does not fall within the standard range after three repetitions, the user is prompted to re-acquire images.
[0121] The actual gingival margin of the target tooth is extracted from the intraoral image using an image segmentation algorithm.
[0122] It should be noted that the actual gingival margin is extracted as follows: a pre-trained semantic segmentation neural network is used to perform binary classification of the tooth and gingival regions on the intraoral image, and the boundary contour between the two regions is extracted as the actual gingival margin. The training samples of the semantic segmentation neural network include no less than 5,000 intraoral images manually annotated by dental specialists, covering various lighting, angles, and pathological conditions to ensure the network's generalization ability in real-world clinical scenarios.
[0123] Based on the spatial pose matrix, the three-dimensional normal vector of the target tooth surface is calculated.
[0124] Multiple measurement reference points were uniformly sampled within the tooth surface area between the virtual cementoenamel boundary baseline and the actual gingival margin.
[0125] For each measurement reference point, its three-dimensional spatial coordinates are obtained by back-projection based on the spatial pose matrix, and the neighboring pixels within a preset radius around the measurement reference point are extracted to form a neighborhood three-dimensional point set.
[0126] It should be noted that the three-dimensional normal vector refers to the overall orientation of the target tooth surface within the current measurement area, physically meaning a direction perpendicular to the tooth surface. The three-dimensional normal vector serves as the basis for subsequently constructing an orthogonal projection plane and determines the directional standard for measuring tooth atrophy.
[0127] The tooth surface region between the virtual cementoenamel boundary baseline and the actual gingival margin is precisely the clinical area where gingival recession occurs. Limiting the measurement reference points to this region ensures that the three-dimensional normal vector accurately reflects the local tooth surface orientation of the recession area, rather than the average orientation of the entire tooth. In this embodiment, the preset radius is set to the physical distance corresponding to 5 pixels, which is obtained by inverse calculation using a nonlinear pixel-to-physical distance mapping field. The uniform sampling number of the measurement reference points is preset to no less than twenty to ensure the statistical stability of subsequent principal component analysis.
[0128] Principal component analysis is performed on the neighborhood 3D point set to obtain the three principal component directions and their corresponding eigenvalues. The eigenvector corresponding to the smallest eigenvalue is used as the local normal vector of the measurement reference point.
[0129] The physical meaning of principal component analysis is: a neighborhood set of three-dimensional points is approximately distributed on a local surface in three-dimensional space. Among the three principal component directions of principal component analysis, the first two principal component directions describe the extension direction of the local surface, and the third principal component direction (the eigenvector corresponding to the minimum eigenvalue) is orthogonal to the extension direction of the local surface, which is the normal vector direction of the surface.
[0130] The initial value of the global three-dimensional normal vector of the target tooth surface is obtained by averaging the local normal vectors of all measurement reference points.
[0131] Calculate the angle between the local normal vector of each measurement reference point and the initial value of the global three-dimensional normal vector, and compare the maximum value of the angle with the normal consistency threshold.
[0132] When the maximum value of the included angle is higher than the normal consistency threshold, the measurement reference point is determined to be a normal anomaly point. The normal anomaly point is removed from the set of measurement reference points, and the updated global three-dimensional normal vector is obtained by re-averaging the local normal vectors of the remaining measurement reference points.
[0133] When the maximum value of the included angle is lower than the normal consistency threshold, the initial value of the overall three-dimensional normal vector is used as the output of the three-dimensional normal vector of the target tooth surface.
[0134] The process of setting the normal consistency threshold is as follows: For no less than one hundred clinical healthy tooth surface samples, extract the angle distribution between all local normal vectors and the global normal vector on the surface of each sample, and use the 95th percentile of the distribution as the normal consistency threshold. In this embodiment, the normal consistency threshold is set to 15 degrees.
[0135] The mechanism for removing abnormal normal points is designed to address situations where calculus, restorations, or abnormal protrusions are present on the tooth surface. The local normal vectors of these abnormal areas deviate from the overall tooth surface direction, and failure to remove them would contaminate the estimation of the overall three-dimensional normal vectors. The iterative process after removal is executed a maximum of three times, removing the measurement reference point with the largest current angle each time, ensuring that the removal process does not excessively shrink the set of measurement reference points.
[0136] The virtual cementoenamel boundary baseline and the actual gingival margin are projected onto a projection plane orthogonal to the three-dimensional normal vector. Based on the pixel-to-physical distance mapping field, the vertical vector distance between the two projection lines is calculated. Multiple vertical vector distances constitute the local atrophy set of the target tooth. Based on the local atrophy set, the quantitative assessment value of gingival atrophy of the target tooth is output.
[0137] Based on the spatial pose matrix, each pixel in the current intraoral image of the target tooth during the follow-up visit is back-projected to the three-dimensional physical coordinate system to obtain the three-dimensional spatial coordinates corresponding to each pixel.
[0138] For any two adjacent pixels in an image, calculate the ratio of the Euclidean distance between their corresponding three-dimensional spatial coordinates to the pixel distance between the two pixels in the image, and use this ratio as the local scale factor for that pixel location.
[0139] Spatial interpolation of the local scale factor at all pixel locations yields a continuous local scale factor distribution field covering the entire target tooth in the current intraoral image during the follow-up period.
[0140] It should be noted that the physical meaning of the local scale factor is: the actual physical distance (in millimeters) corresponding to a unit pixel distance at that pixel location. Due to the perspective projection effect in intraoral images, the unit pixel of the tooth surface area closer to the lens corresponds to a smaller physical distance, while the unit pixel of the tooth surface area farther from the lens corresponds to a larger physical distance. Therefore, the local scale factor takes different values in different regions of the image, forming a spatial distribution field.
[0141] The specific execution method of spatial interpolation is as follows: the local scale factor calculated for all adjacent pixel pairs in the image is distributed as discrete points in the image space. The bicubic spline interpolation algorithm is used to expand the discrete points into a continuous field covering the current intraoral image of the entire target tooth during the follow-up period, ensuring that the corresponding local scale factor can be queried at any pixel position.
[0142] The spatial gradient magnitude is calculated for a continuous local scale factor distribution field, and then compared with the gradient uniformity threshold.
[0143] When the spatial gradient magnitude is lower than the gradient uniformity threshold, it is determined to be a uniformly mapped region, and the arithmetic mean of the scale factors within the region is used as the uniform mapping coefficient in subsequent physical distance calculations.
[0144] When the spatial gradient magnitude is higher than the gradient uniformity threshold, it is determined to be a non-uniform mapping region. In the subsequent physical distance calculation, the local scale factor is integrated along the connection path to obtain the physical distance between any two points.
[0145] The continuous local scale factor distribution field is output as a nonlinear pixel-to-physical distance mapping field.
[0146] The distinction between uniformly mapped and non-uniformly mapped regions aims to balance computational efficiency and measurement accuracy. When the intraoral image of the target tooth during a follow-up visit is approximately a frontal view (the shooting angle is close to perpendicular to the tooth surface), the local scale factor of the entire image region is approximately constant. In this case, using the arithmetic mean as a uniform mapping coefficient can obtain sufficient accuracy with minimal computation. However, when the shooting angle deviates significantly from the vertical direction, the local scale factor varies significantly in different regions. It is necessary to integrate the local scale factor along the connection path to accurately reflect the nonlinear mapping relationship.
[0147] The process of setting the gradient uniformity threshold is as follows: by testing clinical samples with different shooting angles, the upper limit of the distribution of the gradient amplitude of the local scale factor when the shooting angle deviation is less than five degrees is found, and this upper limit is used as the gradient uniformity threshold. In this embodiment, the gradient uniformity threshold is set to five percent of the mean of the local scale factor.
[0148] The specific execution method of integration along a connected path is as follows: a connected path (usually a straight line) is determined between two pixels in the image. Sampling is performed along this connected path with a step size of one pixel. The local scale factor of each sampled point is queried, and the sum of the local scale factors of all sampled points is taken as the physical distance between the two points. This integration process accurately reflects the nonlinear change of the true physical distance between any two points under the perspective effect.
[0149] The process of outputting quantitative assessment values of gingival recession in the target tooth includes:
[0150] Using the three-dimensional normal vector as the normal of the projection plane and the geometric center point of the virtual cementoenamel boundary baseline as the origin of the projection plane, the specific position and orientation of the projection plane orthogonal to the three-dimensional normal vector in three-dimensional space are determined.
[0151] It should be noted that the definition of an orthogonal projection plane has two core elements: the normal direction and the origin position. The purpose of selecting a three-dimensional normal vector for the normal direction is to ensure that the projection plane is parallel to the surface of the target tooth, thereby eliminating perspective compression caused by the tilt of the shooting angle. The purpose of selecting the geometric center point of the virtual cementoenamel boundary baseline for the origin position is to ensure that the projection plane is located at the center of the measurement area, minimizing the accuracy loss caused by differences in projection distance.
[0152] The geometric center point of the virtual cementoenamel boundary baseline is calculated by taking the arithmetic mean of the three-dimensional spatial coordinates of all sampling points on the virtual cementoenamel boundary baseline, and then obtaining the three-dimensional spatial coordinates of the geometric center point.
[0153] The virtual cementoenamel boundary baseline is projected onto the projection plane along the three-dimensional normal vector direction to obtain the first projection curve.
[0154] The actual gingival margin line is projected onto the projection plane along the three-dimensional normal vector direction to obtain the second projection curve.
[0155] Multiple measurement starting points are uniformly sampled along the first projection curve. For each measurement starting point, a measurement ray is constructed in the projection plane along a local tangent direction perpendicular to the measurement starting point. The measurement ray intersects the second projection curve at a measurement endpoint.
[0156] The specific execution method of projection along the direction of the three-dimensional normal vector is as follows: for each three-dimensional point on the original curve, draw a straight line along the direction of the three-dimensional normal vector; the intersection of this straight line and the projection plane is the projection point; perform the same operation on all points on the original curve and connect them in the original order to obtain the projected curve. The purpose of projecting two curves onto the same plane is to establish a unified two-dimensional measurement coordinate system, so that subsequent distance calculations have clear geometric meaning.
[0157] The design of the measurement ray perpendicular to the local tangent direction is based on the clinical definition that the direction of gingival recession is along the long axis of the tooth (i.e., perpendicular to the local direction of the cementoenamel junction). Therefore, the direction of the measurement ray must be perpendicular to the local tangent of the first projection curve (the projection of the virtual cementoenamel junction baseline). The local tangent is calculated as follows: at each measurement starting point, two sampling points before and after it are taken to form a five-point window. A straight line is fitted to this five-point window, and the direction of the straight line is the local tangent direction of that measurement starting point.
[0158] The number of uniform samples at the measurement starting point is preset to be no less than twenty, to ensure that the shrinkage amount is fully sampled along the spatial distribution of the shrinkage area.
[0159] The physical distance between each measurement starting point and the corresponding measurement ending point is calculated based on the nonlinear pixel-to-physical distance mapping field. This physical distance is then used as the vertical vector distance of that point. The vertical vector distances corresponding to all measurement starting points together constitute the set of local shrinkage amounts.
[0160] Based on the local atrophy set, four comprehensive evaluation indicators were calculated: maximum atrophy depth, average atrophy depth, atrophy width, and atrophy area.
[0161] Four comprehensive assessment indicators are used as the output of quantitative assessment values for gingival recession in the target tooth.
[0162] The specific calculation methods for the four comprehensive evaluation indicators are as follows: the maximum atrophy depth is the maximum value among all local atrophy amounts at all measurement sites; the average atrophy depth is the arithmetic mean of the local atrophy amounts at all measurement sites; the atrophy width is the mesiodistal span corresponding to consecutive measurement sites where the local atrophy amount exceeds the preset atrophy threshold; and the atrophy area is the integral value of the local atrophy amount along the atrophy width direction. The preset process for the atrophy threshold is as follows: the upper limit of the fluctuation range of the normal physiological gingival margin relative to the cementoenamel boundary is used as the atrophy threshold through clinical statistics. In this embodiment, the atrophy threshold is preset to 0.5 mm.
[0163] The four comprehensive assessment indicators describe the state of gingival recession from different dimensions: the maximum recession depth reflects the most severe degree of recession, corresponding to the clinical recession grading standard; the average recession depth reflects the overall level of recession and is suitable for longitudinal efficacy assessment; the recession width reflects the lateral range of recession and determines the coverage of the treatment plan; the recession area is a composite indicator that comprehensively considers depth and width and can be used as a core parameter for quantitative comparison before and after treatment.
[0164] Please see Figure 2 The image shown is a quantitative intraoral image assessment system for the degree of gingival recession according to this embodiment, which specifically includes:
[0165] The data extraction module is used to extract historical intraoral images of the target tooth during the baseline period and current intraoral images during the follow-up period. The baseline gingival morphological feature data of the target tooth is extracted from the historical intraoral images during the baseline period. The baseline gingival morphological feature data includes the set of gingival margin apexes of the target tooth during the baseline period and the cementoenamel boundary curvature parameters of the target tooth during the baseline period.
[0166] The baseline generation module is used to map the gingival morphology features in the historical intraoral images during the baseline period to the current intraoral images during the follow-up period based on the baseline gingival morphology feature data. A virtual cementoenamel boundary baseline is generated in the current intraoral image of the target tooth during the follow-up period. The virtual cementoenamel boundary baseline serves as the starting point reference for gingival recession measurement.
[0167] The anchor point extraction module is used to extract at least one morphological constant from the current intraoral image of the target tooth during the follow-up period as an endogenous anatomical anchor point. The morphological constant is the anatomical geometry of the target tooth itself.
[0168] The parameter query module is used to query the standard physical size parameters that match the target tooth type from a pre-built database of standard anatomical dimensions of teeth. The target tooth type is determined according to the general tooth position numbering method, and each tooth position type corresponds to a set of standard physical size parameters.
[0169] The data processing module is used to calculate the spatial pose matrix of the current shooting angle of the current intraoral image during the follow-up period, using the image measurement values of the endogenous anatomical anchor points and standard physical size parameters as constraints, and to generate a nonlinear pixel-to-physical distance mapping field for the local tooth surface of the target tooth based on the spatial pose matrix.
[0170] The edge line extraction module is used to extract the actual gingival edge line of the target tooth from the current intraoral image during the follow-up visit using an image segmentation algorithm.
[0171] The normal vector calculation module is used to calculate the three-dimensional normal vector of the target tooth surface based on the spatial pose matrix.
[0172] The results output module projects the virtual cementoenamel boundary baseline and the actual gingival margin line onto a projection plane orthogonal to the three-dimensional normal vector, respectively. Based on the pixel-to-physical distance mapping field, it calculates the vertical vector distance between the two projection lines. Multiple vertical vector distances constitute a set of local atrophy amounts of the target tooth. Based on the set of local atrophy amounts, it outputs a quantitative assessment value of gingival atrophy of the target tooth.
[0173] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the above functions can be divided into different functional modules to complete all or part of the functions described above.
[0174] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.
[0175] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for quantitative assessment of the degree of gingival recession using intraoral images, characterized in that, include: Extract the baseline period historical intraoral images and the current intraoral images during the follow-up period of the target tooth. Extract the baseline gingival morphological feature data of the target tooth from the baseline period historical intraoral images. The baseline gingival morphological feature data includes the set of gingival margin apex of the target tooth during the baseline period and the cementoenamel boundary curvature parameter of the target tooth during the baseline period. Based on the baseline gingival morphology data, the gingival morphology features in the historical intraoral images during the baseline period are mapped to the current intraoral images during the follow-up period. A virtual cementoenamel boundary baseline is generated in the current intraoral image during the follow-up period of the target tooth. The virtual cementoenamel boundary baseline serves as the starting point reference for gingival recession measurement. At least one morphological constant is extracted from the current intraoral image during the follow-up period of the target tooth as an endogenous anatomical anchor point, wherein the morphological constant is the anatomical geometry of the target tooth itself; The standard physical dimension parameters that match the target tooth type are retrieved from a pre-built database of standard anatomical dimensions of teeth. The target tooth type is determined according to a common tooth position numbering method, and each tooth position type corresponds to a set of standard physical dimension parameters. Using the image measurement values of the endoscopic anchor points and the standard physical size parameters as constraints, the spatial pose matrix of the current shooting angle of the current intraoral image during the follow-up period is calculated, and a nonlinear pixel-to-physical distance mapping field for the local tooth surface of the target tooth is generated based on the spatial pose matrix. The actual gingival margin of the target tooth is extracted from the current intraoral image during the follow-up visit using an image segmentation algorithm. Based on the spatial pose matrix, calculate the three-dimensional normal vector of the target tooth surface; The virtual cementoenamel boundary baseline and the actual gingival margin are projected onto a projection plane orthogonal to the three-dimensional normal vector, respectively. Based on the pixel-to-physical distance mapping field, the vertical vector distance between the two projection lines is calculated. The multiple vertical vector distances constitute a set of local atrophy amounts of the target tooth. Based on the set of local atrophy amounts, a quantitative assessment value of gingival atrophy of the target tooth is output.
2. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 1, characterized in that: The mapping of gingival morphological features from historical intraoral images during the baseline period to current intraoral images during the follow-up visit, based on the baseline gingival morphological feature data, specifically includes: When multiple intraoral images of the target tooth are available for reference period, at least one image is selected as a valid intraoral image of the reference period based on the principles of the closest acquisition time of the reference period, the complete visibility of the cementoenamel junction in the reference period image, and the continuous and uninterrupted gingival margin of the target tooth in the reference period image. The set of gingival margin vertices is extracted from the effective reference period historical intraoral images as the first control point set, and the equidistant sampling points on the cementoenamel boundary curve are extracted as the second control point set. The first control point set and the second control point set together constitute the source domain control point set. During the follow-up period of the target tooth, in the current intraoral image, the candidate positions of the target domain control points corresponding to the source domain control point set are determined based on the crown contour of the target tooth, thus obtaining the target domain control point set; Construct a continuous deformation function from the source domain control point set to the target domain control point set, wherein the continuous deformation function includes affine transformation components and nonlinear bending components; Calculate the bending energy value of the continuous deformation function, compare the bending energy value with the upper limit of bending energy, and when the bending energy value is higher than the upper limit of bending energy, it is determined that the mapping result does not meet the anatomy rationality. Increase the sampling density of the source domain control point set and reconstruct the continuous deformation function until the bending energy value is lower than the upper limit of bending energy. Applying the continuous deformation function to the cementoenamel boundary curve of the historical intraoral image during the baseline period yields the virtual cementoenamel boundary baseline in the current intraoral image of the target tooth during the follow-up period.
3. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 2, characterized in that: The process of obtaining the virtual cementoenamel boundary baseline in the current intraoral image during the follow-up period of the target tooth also includes verifying the anatomical rationality of the virtual cementoenamel boundary baseline, specifically including: Several verification points are uniformly sampled along the virtual cementoenamel boundary baseline. The local curvature value of each verification point is calculated and compared with the upper limit and lower limit of the standard curvature, respectively. When the local curvature value falls between the upper limit and the lower limit of the standard curvature, the verification point is determined to be a reasonable point for dissection, and the position coordinates of the verification point are retained. When the local curvature value is higher than the upper limit of the standard curvature, the verification point is determined to be an excessively curved point. The local curvature value of the verification point is then pulled back to the upper limit of the standard curvature. The pullback range is proportional to the difference between the local curvature value and the upper limit of the standard curvature. When the local curvature value is lower than the standard curvature lower limit, the verification point is determined to be an excessively flat point. The local curvature value of the verification point is then pulled back to the standard curvature lower limit. The pullback range is proportional to the difference between the local curvature value and the standard curvature lower limit. The proportion of excessively curved points and excessively straight points among all verification points is counted. When the proportion exceeds the anatomical deviation proportion threshold, it is determined that the selection of historical intraoral images in the baseline period is inappropriate, triggering the re-screening of historical intraoral images in the baseline period and expanding the screening range of historical intraoral images in the baseline period. The virtual cementoenamel boundary baseline, adjusted after pulling back from all verification points, is used as the final output virtual cementoenamel boundary baseline.
4. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 1, characterized in that: The extraction of at least one morphological constant from the current intraoral image during the follow-up period of the target tooth as an endoanatomical anchor point specifically includes: Multiple candidate morphological constants are extracted simultaneously from the current intraoral image of the target tooth during the follow-up period. The candidate morphological constants include the maximum mesiodistal width of the crown of the target tooth, the interdental papilla base distance between the target tooth and the mesial adjacent tooth, the interdental papilla base distance between the target tooth and the distal adjacent tooth, and the vertical height of the crown of the target tooth. For each candidate morphological constant, its identifiability confidence in the image is calculated. The identifiability confidence is determined by the geometric mean of the image segmentation confidence of the two boundary points of the candidate morphological constant. The identifiability confidence of each candidate morphological constant is compared with the confidence threshold. Candidate morphological constants with identifiability confidence higher than the confidence threshold are identified as valid morphological constants, while candidate morphological constants with identifiability confidence lower than the confidence threshold are eliminated. When the number of effective morphological constants is lower than the lower limit of the number of anchor points, the confidence threshold is reduced and the filter is re-selected until the number of effective morphological constants reaches the lower limit of the number of anchor points. When the number of effective morphological constants is higher than or equal to the lower limit of the number of anchor points, all effective morphological constants are output as endogenous anatomical anchor points, and the identifiability confidence of each endogenous anatomical anchor point is used as the weighting weight of each constraint condition in the subsequent pose calculation algorithm.
5. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 4, characterized in that: The process of calculating the spatial pose matrix of the current shooting viewpoint includes: The standard physical dimension parameters of the target tooth type are retrieved from the prior database of standard anatomical dimensions of teeth. The standard physical dimension parameters are expressed as a statistical distribution consisting of the mean and standard deviation. For each endogenous anatomical anchor point, a scale consistency constraint term is constructed. The scale consistency constraint term is the difference between the physical measurement value obtained by back-projecting the image measurement value of the endogenous anatomical anchor point through the spatial pose matrix and the mean value of the standard physical size parameter. For each endogenous anatomical anchor point, a distribution rationality constraint is constructed, wherein the distribution rationality constraint is the ratio between the distance of the physical measurement value from the mean of the standard physical size parameter and the standard deviation; The joint constraint loss function is obtained by weighted summation of the sum of squares of the scale consistency constraint and the sum of squares of the distribution rationality constraint, with the weights determined by the identifiability confidence of the endogenous anatomical anchors. The initial solution of the spatial pose matrix is obtained by minimizing the joint constraint loss function through an iterative optimization method. The spatial pose matrix includes rotation and translation components. Calculate the deviation of the physical measurement values of all endogenous anatomical anchor points corresponding to the initial solution from the standard physical size parameters. Compare the maximum value of the deviation with the upper limit of the deviation. When the maximum value of the deviation is higher than the upper limit of the deviation, expand the search range of the pose calculation algorithm and iterate again until the maximum value of the deviation is lower than the upper limit of the deviation. The final converged spatial pose matrix is output as the spatial pose matrix of the current shooting perspective.
6. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 5, characterized in that: After outputting the final converged spatial pose matrix as the spatial pose matrix of the current shooting viewpoint, the process also includes establishing a feedback calibration mechanism between the spatial pose matrix and the virtual cementoenamel boundary baseline. The specific process includes: Based on the spatial pose matrix, the virtual cementoenamel boundary baseline is back-projected onto the three-dimensional physical space to obtain the three-dimensional virtual cementoenamel boundary curve. The standard longitudinal position range of the cementoenamel boundary for the target tooth type is retrieved from the prior database of standard anatomical dimensions of teeth. The standard longitudinal position range of the cementoenamel boundary is expressed as the longitudinal distance between the maximum outline of the crown and the cementoenamel boundary. The actual longitudinal distance is obtained by calculating the longitudinal distance between the three-dimensional virtual cementoenamel boundary curve and the maximum outline of the crown of the target tooth in three-dimensional physical space. When the actual longitudinal distance falls within the standard longitudinal position range of the cementoenamel boundary, it is determined that the virtual cementoenamel boundary baseline and the spatial pose matrix are consistent and no calibration is required. When the actual longitudinal distance falls outside the standard longitudinal position range of the cementoenamel boundary, it is determined that there is a deviation in the virtual cementoenamel boundary baseline or the spatial pose matrix, triggering feedback calibration. Specifically, this includes calculating the deviation of the actual longitudinal distance from the standard longitudinal position range of the cementoenamel boundary, allocating the deviation to the position adjustment of the virtual cementoenamel boundary baseline and the attitude adjustment of the spatial pose matrix. The allocation ratio is determined by the reciprocal of the solution confidence of each of them, where the solution confidence is a quantitative evaluation value of the output reliability. The calibrated virtual cementoenamel boundary baseline and the calibrated spatial pose matrix are applied to subsequent measurement steps.
7. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 1 or 5, characterized in that: Based on the spatial pose matrix, the three-dimensional normal vector of the target tooth surface is calculated. The specific process includes: Multiple measurement reference points were uniformly sampled in the tooth surface area between the virtual cementoenamel boundary baseline and the actual gingival margin line; For each measurement reference point, its three-dimensional spatial coordinates are obtained by back projection based on the spatial pose matrix, and the neighboring pixels within a preset radius around the measurement reference point are extracted to form a neighborhood three-dimensional point set. Principal component analysis is performed on the neighborhood three-dimensional point set to obtain three principal component directions and their corresponding eigenvalues. The eigenvector corresponding to the minimum eigenvalue is used as the local normal vector of the measurement reference point. The initial value of the global three-dimensional normal vector of the target tooth surface is obtained by averaging the local normal vectors of all measurement reference points. Calculate the angle between the local normal vector of each measurement reference point and the initial value of the overall three-dimensional normal vector, and compare the maximum value of the angle with the normal consistency threshold. When the maximum value of the included angle is higher than the normal consistency threshold, the measurement reference point is determined to be a normal anomaly point. The normal anomaly point is removed from the set of measurement reference points, and the updated global three-dimensional normal vector is obtained by re-averaging the local normal vectors of the remaining measurement reference points. When the maximum value of the included angle is lower than the normal consistency threshold, the initial value of the overall three-dimensional normal vector is output as the three-dimensional normal vector of the target tooth surface.
8. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 5, characterized in that: Based on this spatial pose matrix, a nonlinear pixel-to-physical distance mapping field is generated for a local tooth surface of the target tooth, specifically including: Based on the spatial pose matrix, each pixel in the current intraoral image of the target tooth during the follow-up period is back-projected to the three-dimensional physical coordinate system to obtain the three-dimensional spatial coordinates corresponding to each pixel. For any two adjacent pixels in an image, calculate the ratio of the Euclidean distance between their corresponding three-dimensional spatial coordinates to the pixel distance between the two pixels in the image, and use the ratio as the local scale factor of the pixel location. Spatial interpolation of the local scale factor at all pixel locations yields a continuous local scale factor distribution field covering the entire target tooth during the current intraoral image at the follow-up period. Calculate the spatial gradient magnitude for the continuous local scale factor distribution field, and compare the spatial gradient magnitude with a gradient uniformity threshold; When the spatial gradient magnitude is lower than the gradient uniformity threshold, it is determined to be a uniform mapping region, and the arithmetic mean of the scale factors within the region is used as the uniform mapping coefficient in the subsequent physical distance calculation. When the spatial gradient magnitude is higher than the gradient uniformity threshold, it is determined to be a non-uniform mapping region. In the subsequent physical distance calculation, the local scale factor is integrated along the connection path to obtain the physical distance between any two points. The continuous local scale factor distribution field is output as the nonlinear pixel-to-physical distance mapping field.
9. The method for quantitative assessment of the degree of gingival recession using intraoral images as described in claim 1, characterized in that: The specific process for outputting the quantitative assessment value of gingival recession in the target tooth includes: Using the three-dimensional normal vector as the normal of the projection plane and the geometric center point of the virtual cementoenamel boundary baseline as the origin of the projection plane, the specific position and orientation of the projection plane orthogonal to the three-dimensional normal vector in three-dimensional space are determined. The virtual cementoenamel boundary baseline is projected onto the projection plane along the direction of the three-dimensional normal vector to obtain the first projection curve. The actual gingival margin line is projected onto the projection plane along the direction of the three-dimensional normal vector to obtain a second projection curve; Multiple measurement starting points are uniformly sampled along the first projection curve. For each measurement starting point, a measurement ray is constructed in the projection plane along a local tangent direction perpendicular to the measurement starting point. The measurement ray intersects the second projection curve at a measurement endpoint. Based on the nonlinear pixel-to-physical distance mapping field, the physical distance between each measurement starting point and the corresponding measurement ending point is calculated as the vertical vector distance of that point. The vertical vector distances corresponding to all measurement starting points together constitute the set of local shrinkage amounts. Based on the aforementioned set of local atrophy amounts, four comprehensive evaluation indicators were calculated: maximum atrophy depth, average atrophy depth, atrophy width, and atrophy area. The four comprehensive evaluation indicators are used as the quantitative evaluation values of gingival recession in the target tooth.
10. A quantitative assessment system for the degree of gingival recession using intraoral images, characterized in that: Specifically, it includes: The data extraction module is used to extract the baseline historical intraoral images and the current intraoral images during the follow-up period of the target tooth. The baseline gingival morphological feature data of the target tooth is extracted from the baseline historical intraoral images. The baseline gingival morphological feature data includes the set of gingival margin apexes of the target tooth during the baseline period and the cementoenamel boundary curvature parameter of the target tooth during the baseline period. The baseline generation module is used to map the gingival morphology features in the historical intraoral images during the baseline period to the current intraoral images during the follow-up period based on the baseline gingival morphology feature data, and generate a virtual cementoenamel boundary baseline in the current intraoral image during the follow-up period of the target tooth. The virtual cementoenamel boundary baseline serves as the starting point reference for gingival recession measurement. Anchor point extraction module is used to extract at least one morphological constant as an endogenous anatomical anchor point from the current intraoral image of the target tooth during the follow-up period. The morphological constant is the anatomical geometry of the target tooth itself. The parameter query module is used to query the standard physical size parameters that match the target tooth type from a pre-built prior database of standard anatomical dimensions of teeth. The target tooth type is determined according to the general tooth position numbering method, and each tooth position type corresponds to a set of standard physical size parameters. The data calculation module is used to calculate the spatial pose matrix of the current shooting angle of the current intraoral image during the follow-up period, using the image measurement value of the endogenous anatomical anchor point and the standard physical size parameter as constraints, and to generate a nonlinear pixel-to-physical distance mapping field for the local tooth surface of the target tooth based on the spatial pose matrix. The edge line extraction module is used to extract the actual gingival edge line of the target tooth from the current intraoral image during the follow-up visit using an image segmentation algorithm; The normal vector calculation module is used to calculate the three-dimensional normal vector of the target tooth surface based on the spatial pose matrix. The result output module is used to project the virtual cementoenamel boundary baseline and the actual gingival margin line onto a projection plane orthogonal to the three-dimensional normal vector, and calculate the vertical vector distance between the two projection lines based on the pixel-to-physical distance mapping field. The multiple vertical vector distances constitute a set of local atrophy amounts of the target tooth, and the quantitative assessment value of gingival atrophy of the target tooth is output based on the set of local atrophy amounts.
Citation Information
Patent Citations
A method and device for intelligent analysis of oral cavity images
CN111563887B
Precise quantitative evaluation method of periodontal disease based on ios image and cbct image
CN119205680B
