A method for oral image segmentation for oral diagnosis and treatment

Abstract

This invention specifically relates to a method for oral image segmentation in oral diagnosis and treatment, belonging to the field of oral diagnosis and treatment technology. It includes acquiring evaluation indicators such as axial convergence deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation. This invention constructs a fully automated solution from image input to quality rating, significantly improving the standardization and efficiency of diagnosis and treatment. Based on a precise segmentation mask, it automatically extracts multi-dimensional quantitative indicators such as axial convergence deviation and shoulder curvature standard deviation through algorithms such as plane fitting and curvature calculation, comprehensively covering core diagnostic and treatment requirements such as crown retention, sealing, and restorative space.

Description

Technical Field

[0001] This invention relates to the field of oral diagnosis and treatment technology, and in particular to an oral image segmentation method for oral diagnosis and treatment. Background Technology

[0002] In dental restoration, tooth preparation is a crucial step, and its quality directly affects the retention, sealing, and lifespan of dental crowns.

[0003] Traditional tooth preparation quality assessment mainly relies on the doctor's subjective experience, through visual observation or simple instrument measurement. This has problems such as inconsistent assessment standards, low accuracy, and strong subjectivity, which can easily lead to unstable restoration results and even complications such as microleakage, gingivitis, and restoration breakage.

[0004] With the development of digital oral technology, equipment such as 3D intraoral scanners and dental microscopes have been widely used, enabling the acquisition of high-precision oral image data.

[0005] However, existing oral cavity image segmentation techniques still have many shortcomings:

[0006] The segmentation targets are mostly large structures such as teeth and gums, making it difficult to accurately segment and prepare small critical structures such as the shoulder (only 0.5-1mm wide).

[0007] The segmentation network lacks targeted optimization and is easily affected by image noise, reflection, shadows, etc., resulting in insufficient segmentation accuracy. Thirdly, there is a lack of an integrated solution from image segmentation to quality assessment. The segmentation results cannot directly provide quantitative support for diagnosis and treatment assessment, requiring additional manual intervention, which is inefficient.

[0008] Therefore, there is an urgent need for a method that can achieve precise segmentation of key oral structures and quantitatively assess the quality of tooth preparation based on the segmentation results, in order to overcome the shortcomings of existing technologies and improve the accuracy and standardization of dental restoration treatment. Summary of the Invention

[0009] The purpose of this invention is to provide a method for oral image segmentation for oral diagnosis and treatment in order to solve the above-mentioned problems.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A method for oral image segmentation for oral diagnosis and treatment, comprising:

[0012] Acquire 3D / 2D oral cavity images, and after integrity verification, output standardized, high signal-to-noise ratio image data through preprocessing;

[0013] The prepared tooth body, adjacent teeth, gingival margin, and prepared shoulder are segmented, and a segmentation mask is generated;

[0014] The evaluation indicators include axial surface convergence deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation.

[0015] After comprehensive analysis and processing of various evaluation indicators, a comprehensive evaluation coefficient is obtained, and the corresponding tooth preparation quality grade is matched.

[0016] Preferably, the acquisition of 3D / 2D oral images, after integrity verification, and subsequent preprocessing to output standardized, high signal-to-noise ratio image data, specifically includes:

[0017] Obtain a three-dimensional digital model or a two-dimensional color image of the oral cavity;

[0018] Preprocessing the 3D model:

[0019] Coordinate alignment: The STL model is automatically aligned to the Frankfort plane as the standard jaw plane, and the bilateral infraorbital points and external auditory canal points in the model are extracted to construct a reference coordinate system;

[0020] Region of Interest (ROI) trimming: Based on dental anatomy features, the crown-root boundary of the target tooth is automatically identified, and excess alveolar bone, other teeth and soft tissue models are trimmed off using this boundary as a reference.

[0021] Smooth the trimmed model;

[0022] 2D image preprocessing:

[0023] Illumination normalization: Decompose the reflection component and illumination component of the image, and adjust the illumination component;

[0024] Color space conversion: Convert the RGB image to the HSV color space, enhance color differences, and then convert it back to the RGB space for output;

[0025] Image scaling and alignment: Scaling the image to a preset pixel size and automatically rotating the image based on the tooth's central axis.

[0026] Preferably, the specific components include:

[0027] Core objective of segmentation: tooth preparation;

[0028] Reference targets: adjacent teeth, gingival margin, prepared shoulder, occlusal surface;

[0029] The segmentation network uses an improved U-Net++ network as its architecture;

[0030] A convolutional block attention module is introduced, adding channel attention and spatial attention mechanisms after each convolutional block of the encoder;

[0031] And design an edge-aware loss function;

[0032] In the last layer of the decoder, 1×1 convolutions are used instead of 3×3 convolutions, and batch normalization layers are added.

[0033] Preferably, the method further includes:

[0034] Training the data includes dataset construction, data augmentation, and determining training parameters;

[0035] Post-segmentation processing and accuracy verification:

[0036] Morphological optimization of the segmentation mask output by the network includes:

[0037] Connectivity analysis: Remove isolated noise regions with an area less than 50 pixels;

[0038] Edge smoothing: Morphological closing operations are used to fill holes in the segmented area, and then median filtering is used to smooth the edges to ensure that the edge lines such as shoulders and ridges are continuous and without breaks;

[0039] Regional consistency check: Check the positional relationship between the prepared tooth and the adjacent tooth. If there is a positional conflict, the dividing boundary is automatically adjusted to a preset reasonable range.

[0040] Accuracy verification: For the segmentation results of each input image, the Dice coefficient, cross-union ratio and pixel accuracy of each target region are automatically calculated;

[0041] Output: Generates a multi-channel segmentation mask, with each channel corresponding to a segmentation target. The mask corresponds one-to-one with the preprocessed original image.

[0042] Preferably, the process of obtaining the axial surface aggregation deviation includes:

[0043] For 3D models:

[0044] Point cloud data of the prepared tooth is extracted from the segmentation mask, and the point cloud is divided into four axial regions: buccal, lingual, mesial, and distal, based on the tooth's central axis.

[0045] For each axial region, the point cloud is fitted to a plane using the least squares method;

[0046] Calculate the angle between two planes based on their normal vectors.

[0047] The axial convergence degree is obtained by adding the buccal-lingual tilt angle and the mesial-distal tilt angle.

[0048] The normal range of the axial surface cohesion degree is preset. If the obtained axial surface cohesion degree is not within the normal range, the difference between the axial surface cohesion degree and the normal range is calculated and recorded as the axial surface cohesion deviation.

[0049] For 2D images:

[0050] If only a single 2D image is entered, two orthogonal images will be taken in the cheek-tongue direction and the near-to-mid direction; if multiple images have been entered, orthogonal view images will be automatically selected.

[0051] For each image, extract the axial edge contour using the prepared tooth segmentation mask;

[0052] Straight lines were fitted to the buccal, lingual, mesial, and distal margin contours to obtain the slope of each margin.

[0053] The included angle is calculated based on the slope of the relative edges, the degree of aggregation is obtained by summing the angles, and the axial aggregation deviation is obtained.

[0054] Preferably, the process of obtaining the standard deviation of the shoulder curvature and the coefficient of variation of the shoulder width includes:

[0055] Extract the edge contour of the pre-designed shoulder from the segmentation mask, and discretize the curve into a sequence of coordinate points:

[0056] For a sequence of discrete coordinate points, the curvature of each point is calculated using the five-point numerical differentiation method;

[0057] The standard deviation of the shoulder curvature was calculated.

[0058] The maximum allowable threshold for the standard deviation of shoulder curvature is preset. If the standard deviation of shoulder curvature is greater than the maximum allowable threshold, the difference between the two is calculated and recorded as the standard deviation of shoulder curvature.

[0059] Extract the gingival margin contour from the segmentation mask, and use this contour as a reference line to offset a preset distance toward the crown direction;

[0060] A predetermined number of measurement points are evenly selected on the outline of the prepared shoulder platform and distributed at equal intervals along the circumference of the shoulder platform. The vertical distance from the surface of the prepared body to the reference line is measured at each point.

[0061] After obtaining each vertical distance, calculate the mean and standard deviation, and divide the standard deviation by the mean to obtain the shoulder width variation coefficient. Preset the maximum allowable threshold for the shoulder width variation coefficient. If the shoulder width variation coefficient is greater than the maximum allowable threshold, calculate the difference between the two and record it as the shoulder width variation deviation coefficient.

[0062] Preferably, the process of obtaining the minimum spatial deviation of the occlusal surface includes:

[0063] Extract the occlusal point cloud or contour of adjacent teeth from the segmentation mask and fit it to the reference plane;

[0064] A predetermined number of measurement points are evenly selected within the preparatory occlusal surface segmentation area, covering the buccal side, lingual side, and central fossa;

[0065] Calculate the vertical distance from each measurement point to the reference plane;

[0066] The obtained vertical distances are sorted in ascending order of their numerical values, and the minimum vertical distance is extracted and recorded as the minimum space of the occlusal surface. A minimum space threshold for the occlusal surface is preset. If the minimum space of the occlusal surface is less than the minimum space threshold, the difference between the two is calculated and recorded as the minimum space deviation of the occlusal surface.

[0067] Preferably, the process of obtaining the minimum adjacent surface gap includes:

[0068] Extract the proximal regions of the preparatory body and adjacent teeth from the segmentation mask;

[0069] The KD tree algorithm is used to quickly calculate the minimum distance or minimum pixel distance between two adjacent surfaces;

[0070] A preset number of cross sections are uniformly selected along the perpendicular direction of the adjacent surfaces. The adjacent surface gap of each cross section is calculated, and all the obtained adjacent surface gaps are sorted in ascending order according to their numerical values. The minimum adjacent surface gap is extracted. The normal range of the minimum adjacent surface gap is preset. If the minimum adjacent surface gap is not within the normal range, the difference between the minimum adjacent surface gap and the normal range is calculated and recorded as the minimum adjacent surface gap deviation.

[0071] Preferably, the step of obtaining a comprehensive evaluation coefficient after comprehensive analysis and processing of various evaluation indicators, and matching it with the corresponding tooth preparation quality grade, specifically includes:

[0072] The weighting factors for the pre-defined axial surface convergence deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation are calculated by multiplying the axial surface convergence deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation with their corresponding weighting factors, and then summing them to obtain the comprehensive evaluation coefficient.

[0073] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0074] 1. This invention significantly improves the standardization and efficiency of diagnosis and treatment by constructing a fully automated solution from image input to quality rating. Based on a precise segmentation mask, it automatically extracts multi-dimensional quantitative indicators such as axial plane aggregation deviation and shoulder curvature standard deviation through algorithms such as plane fitting and curvature calculation, comprehensively covering core diagnosis and treatment requirements such as crown retention, sealing, and restoration space.

[0075] 2. This invention achieves pixel-level precise segmentation of key oral structures through targeted technical optimization, solving the pain point of insufficient segmentation of small targets and complex edges in existing technologies. After segmentation, morphological optimization and accuracy verification ensure that there are no omissions or edge breaks in the segmentation of prepared teeth, adjacent teeth, gingival margins, and other targets, providing reliable basic data support for subsequent diagnosis and treatment quality assessment. Attached Figure Description

[0076] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0077] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0078] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0079] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0080] Example 1

[0081] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0082] Appendix Figure 1 The flowchart of an oral image segmentation method for oral diagnosis and treatment provided in this embodiment of the invention shows the complete steps from acquiring 3D / 2D oral images to obtaining comprehensive evaluation coefficients after comprehensive analysis and processing of various evaluation indicators.

[0083] In this embodiment, it includes:

[0084] Acquire 3D / 2D oral cavity images in compatible formats. After integrity verification, perform targeted preprocessing such as coordinate alignment, cropping, denoising, and illumination normalization to output standardized, high signal-to-noise ratio image data, providing reliable input for subsequent processes.

[0085] Specifically, it includes:

[0086] Obtain a three-dimensional digital model or a two-dimensional color image of the oral cavity;

[0087] 3D digital model: Receive standard STL format files output from an intraoral scanner (such as 3ShapeTRIOS or iTeroElement), requiring a model resolution ≥ 0.05mm and a point cloud density ≥ 1000 points / m². Ensure that the details of the tooth surface (such as the edge of the shoulder) are completely preserved;

[0088] Two-dimensional color images: JPG / PNG format images taken by a dental microscope (such as Carl Zeiss OP-MI100) or a high-definition dental camera, with a resolution of ≥2048×1536 pixels, color mode of RGB, and the image must contain the complete target prepared tooth, adjacent teeth and surrounding gingival tissue, without obvious obstructions (such as instruments, saliva residue).

[0089] The system automatically detects image integrity. If the 3D model has missing surfaces, the 2D image has severe occlusion, or the resolution is insufficient, it will generate an input data failure prompt and clearly indicate the problem type (such as missing mesial surfaces in the 3D model or severe reflection in the gingival area of ​​the 2D image) to guide the user to re-acquire data.

[0090] Preprocessing the 3D model:

[0091] Coordinate alignment: The STL model is automatically aligned to the Frankfort plane (orbitoauricular plane) as the standard jaw plane. The bilateral infraorbital points and external auditory canal points in the model are extracted to construct a reference coordinate system, eliminating model tilt caused by different patient head postures.

[0092] Region of Interest (ROI) trimming: Based on the anatomical features of the tooth, the crown-root boundary (cementoenamel junction) of the target tooth is automatically identified. Using this boundary as a reference, the trimming extends upward to 2mm above the occlusal surface, outward to the midline of the crown of the adjacent tooth, and downward to the root 1 / 3. Excess alveolar bone, other teeth and soft tissue models are trimmed to reduce the amount of computation.

[0093] Denoising and smoothing: The moving least squares (MLS) method was used to smooth the cropped model, removing point cloud noise generated during the scanning process (such as isolated points and sharp protrusions). The smoothing coefficient was set to 0.03 to ensure that the surface morphology of the tooth is realistically restored, while retaining the edge information of key features such as the shoulder.

[0094] 2D image preprocessing:

[0095] Illumination normalization: The Retinex algorithm is used to decompose the reflection component and illumination component of the image. By adjusting the illumination component, reflections (such as strong light reflection on the enamel surface) and shadows (such as shadows at the junction of the gum and tooth) caused by differences in shooting angle and light source intensity are eliminated, so that the image brightness is uniform and the contrast is consistent.

[0096] Color space conversion: Convert the RGB image to the HSV color space, enhance the color difference between the tooth (whiter) and the gum (redder) by adjusting the saturation channel, and then convert it back to the RGB space for output, providing clearer feature differentiation for subsequent segmentation;

[0097] Image scaling and alignment: The image is uniformly scaled to a preset pixel size (2048×2048 pixels) (keeping the aspect ratio unchanged and filling the edges with black), and the image is automatically rotated based on the tooth's central axis to make the buccal and lingual surfaces of the target prepared tooth parallel to the horizontal direction of the image, ensuring the consistency of the direction of subsequent feature calculations;

[0098] The preprocessed 3D model and 2D image are grayscale normalized (pixel values ​​are mapped to the [0,255] range), and the image comparison files before and after preprocessing are saved for subsequent result tracing.

[0099] An improved U-Net++ network (including CBAM attention mechanism and edge-aware loss function) was used to segment key targets such as prepared teeth, adjacent teeth, gingival margins, and prepared shoulders, and pixel-level segmentation masks were generated and verified for accuracy.

[0100] Specifically, it includes:

[0101] Core segmentation target: tooth preparation (only includes the exposed tooth structure after preparation, excluding the original crown portion that has not been removed);

[0102] Reference target:

[0103] Adjacent teeth (teeth on the left / right side that are in direct contact with the target prepared tooth, including the entire crown portion);

[0104] Gingival margin (the junction between the gingiva and the tooth structure, distinguishing between free gingiva and attached gingiva);

[0105] Prepare the shoulder (the edge line of the neck of the tooth body is a key fitting area for crown restoration and needs to be accurate to the pixel level).

[0106] Occlusal surface (the occlusal contact surface of the prepared tooth and adjacent teeth, used for spatial distance measurement);

[0107] The segmentation network uses an improved U-Net++ network as its basic architecture. Compared with the traditional U-Net, it adds a multi-scale feature fusion path, which improves the segmentation ability of small targets (such as pre-existing shoulder platforms with a width of only 0.5-1mm) and complex edges.

[0108] A convolutional block attention module (CBAM) is introduced, adding channel attention and spatial attention mechanisms after each convolutional block of the encoder. By learning weight allocation, the network automatically focuses on key areas such as the crown and shoulder, suppressing interference from irrelevant information such as gingival background and image noise.

[0109] And design an edge-aware loss function: ,in , , These are the preset weighting factors; For Dice's loss; For edge loss; Cross-entropy loss;

[0110] Decoder optimization: Replace 3×3 convolutions with 1×1 convolutions in the last layer of the decoder to reduce parameter redundancy, and add batch normalization (BN) layers to accelerate network convergence and avoid overfitting.

[0111] Training the data includes dataset construction, data augmentation, and determining training parameters;

[0112] Dataset construction: Collect a predetermined number (1000 cases) of clinical dental preparation images (including 500 3D models and 500 2D images), covering different tooth positions such as anterior teeth, premolars, and molars, as well as different preparation quality samples such as excellent, qualified, and unqualified.

[0113] The images were manually annotated by three chief physicians with more than 10 years of experience in dental restoration. The annotation results were used as true labels after passing the consistency test (Kappa coefficient ≥ 0.85).

[0114] Data augmentation: Perform augmentation operations on the training set, such as random flipping (horizontal / vertical), rotation (±15°), scaling (0.8-1.2 times), and adding Gaussian noise (variance ≤0.01), to expand the dataset size and improve the network's generalization ability;

[0115] Training parameters: The Adam optimizer is used, with an initial learning rate of [value missing]. The learning rate decays with the number of iterations (decaying to 0.9 of the original value every 10 epochs), the number of training epochs is 50, the batch size is 8, and an early stopping strategy is adopted (training stops if the validation set loss does not decrease for 5 consecutive epochs).

[0116] Post-segmentation processing and accuracy verification:

[0117] Post-processing steps: Morphological optimization of the segmentation mask output by the network, including:

[0118] Connected component analysis: Remove isolated noise regions with an area less than 50 pixels (such as mis-segmented gingival dots);

[0119] Edge smoothing: Morphological closing operation (3×3 kernel size) is used to fill the small holes in the segmentation area, and then median filtering (3×3 window size) is used to smooth the edges to ensure that the edge lines such as shoulders and ridges are continuous and without breaks;

[0120] Regional consistency check: Check the positional relationship between the prepared tooth and the adjacent tooth (must meet the clinical anatomical distance, such as interproximal space ≥0.1mm). If there is a positional conflict (such as overlapping segmented areas), the segmentation boundary is automatically adjusted to a preset reasonable range.

[0121] Accuracy verification: For each input image segmentation result, the Dice coefficient, intersection-over-union (IoU) ratio, and pixel accuracy (PA) of each target region are automatically calculated, and an accuracy report is generated. If the Dice coefficient of a target region is less than a preset threshold (e.g., <0.92), it will be marked as insufficient segmentation accuracy, and the network will be triggered to re-segment (adjust attention weights) or prompt the user to check the quality of the input image.

[0122] Output: Generates a multi-channel segmentation mask (3D model as point cloud label, 2D image as pixel label), each channel corresponds to a segmentation target (e.g., channel 1 = prepared tooth, channel 2 = adjacent tooth, channel 3 = gingival margin, channel 4 = prepared shoulder, channel 5 = occlusal surface). The mask corresponds one-to-one with the preprocessed original image, which facilitates region localization during subsequent feature calculation.

[0123] Based on the segmentation mask, evaluation indicators including axial plane aggregation deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation are obtained through algorithms such as plane fitting, curvature calculation, and distance measurement.

[0124] The process of obtaining the axial plane convergence deviation includes:

[0125] The sum of the inclination angles of the prepared body relative to the two side walls (buccal-lingual, mesial-distal) reflects the taper characteristics of the prepared body, which directly affects the retention and stability of the crown.

[0126] Too little convergence (<4°) will make it difficult for the crown to be placed, while too much convergence (>14°) will reduce retention. The ideal range is 6-10°.

[0127] For 3D models:

[0128] Point cloud data of the prepared tooth was extracted from the segmentation mask. Based on the tooth's central axis (extracted by principal component analysis PCA), the point cloud was divided into four axial regions: buccal, lingual, mesial, and distal (each region occupies 90° of the circumference of the prepared tooth).

[0129] For each axial region, the point cloud is fitted to a plane using the least squares method to obtain the plane equations for each axial plane (e.g., the cheek plane equation). Equation of the lateral plane: , , , To prepare the three-dimensional coordinates of each point in the point cloud of the axial plane region of the tooth; , , It is the normal vector component of the cheek-side axial plane fitting plane; It is a constant term of the cheek-side axial plane fitting plane; , , These are the normal vector components of the plane fitted to the tongue-side axial surface; It is a constant term for the plane fitting the axial surface of the tongue;

[0130] The angle (i.e., tilt angle) between two opposing planes (buccal-lingual, mesial-distal) is calculated using the normal vectors of the two opposing planes, as shown in the following formula: , , , where are the normal vectors relative to the plane. The angle between the two planes is expressed in degrees.

[0131] buccal-lingual tilt angle ( ) and near-to-far tilt angle ( Add the values ​​together to obtain the axial convergence; if the prepared body is an anterior tooth, only the mesiodistal inclination angle is calculated as the axial convergence.

[0132] The normal range of the axial surface cohesion degree is preset. If the obtained axial surface cohesion degree is not within the normal range, the difference between the axial surface cohesion degree and the normal range is calculated, and the absolute value is recorded as the axial surface cohesion deviation.

[0133] For 2D images:

[0134] If only a single 2D image is entered, the system prompts the user to take two orthogonal images in the cheek-tongue direction and the near-to-mid direction; if multiple images have been entered, the system automatically filters out orthogonal view images.

[0135] For each image, the prepared tooth segmentation mask is used to extract the axial edge contour using the Canny edge detection algorithm;

[0136] Straight lines were fitted to the buccal, lingual, mesial, and distal margin contours to obtain the slope of each margin.

[0137] Calculate the included angle based on the slope of the relative edges, sum them to obtain the degree of convergence (the calculation logic is the same as that of the 3D model, and it needs to be multiplied by the perspective correction coefficient, such as 1.05, to compensate for the perspective error of the 2D image), and obtain the axial plane convergence deviation.

[0138] The process of obtaining the standard deviation of shoulder curvature and the coefficient of variation of shoulder width includes:

[0139] The prepared shoulder is the connection point between the crown and the tooth body. Its continuity (no interruption, no sharp edges) and width uniformity directly affect the seal and fit of the crown, avoiding complications such as microleakage and gingivitis after restoration.

[0140] Extract the edge contour of the pre-designed shoulder from the segmentation mask (3D model is a spatial curve, 2D image is a planar curve), and discretize the curve into a sequence of coordinate points:

[0141] 3D: ;

[0142] 2D: ;

[0143] For a sequence of discrete coordinate points, the curvature of each point is calculated using the five-point numerical differentiation method. The formula is as follows (taking a 2D curve as an example): ,in , The first derivative, , It is the second derivative;

[0144] The standard deviation of shoulder curvature is calculated; the smaller the standard deviation of shoulder curvature, the smoother the shoulder. The maximum allowable threshold of the standard deviation of shoulder curvature is preset. If the standard deviation of shoulder curvature is greater than the maximum allowable threshold, the difference between the two is calculated, and the absolute value is recorded as the standard deviation of shoulder curvature.

[0145] Extract the gingival margin contour from the segmentation mask, and use this contour as a reference to offset it towards the crown by a preset distance (0.5 mm, the lower limit of the clinically standard shoulder width) as a reference line;

[0146] A predetermined number (50) of measurement points are evenly selected on the prepared shoulder profile and distributed equidistantly along the circumference of the shoulder. The vertical distance from the surface of the prepared body to the reference line (i.e., the width of the shoulder) is measured at each point.

[0147] After obtaining each vertical distance, calculate the mean and standard deviation, and divide the standard deviation by the mean to obtain the shoulder width variation coefficient. The maximum allowable threshold for the shoulder width variation coefficient is preset. If the shoulder width variation coefficient is greater than the maximum allowable threshold, calculate the difference between the two, and record the absolute value as the shoulder width variation deviation coefficient.

[0148] The process of obtaining the minimum spatial deviation of the occlusal surface includes:

[0149] The occlusal surface of the prepared dental prosthesis must provide sufficient vertical space for the crown restoration (such as porcelain crown or all-ceramic crown) to ensure the thickness (usually ≥1.5mm) and strength of the restoration and to prevent the restoration from breaking during occlusion.

[0150] Extract the occlusal surface point cloud (3D) or contour (2D) of adjacent teeth from the segmentation mask, and use the RANSAC algorithm to fit the reference plane (excluding abnormal points such as occlusal surface wear and defects).

[0151] A predetermined number (30) of measurement points are evenly selected within the preparatory body occlusal surface segmentation area, covering key locations such as the buccal side, lingual side, and central fossa;

[0152] Calculate the vertical distance from each measurement point to the reference plane (the direction of the distance is the direction of tooth eruption, i.e., perpendicular to the occlusal plane and upwards).

[0153] The obtained vertical distances are sorted in ascending order of numerical value, and the minimum vertical distance is extracted and recorded as the minimum space of the occlusal surface. A minimum space threshold of the occlusal surface is preset. If the minimum space of the occlusal surface is less than the minimum space threshold, the difference between the two is calculated, and the absolute value is recorded as the minimum space deviation of the occlusal surface.

[0154] If the target tooth to be prepared is an anterior tooth (without a clear occlusal surface), then the measurement will be changed to the distance from the incisal edge to the reference plane (≥1.0mm); if the adjacent tooth is missing, the system will prompt the user to enter the occlusal surface data of the opposing tooth or use the standard occlusal plane as a reference.

[0155] The process of obtaining the minimum adjacent face gap includes:

[0156] A clear and continuous interproximal space must be maintained between the prepared tooth and the adjacent tooth to facilitate the placement and cleaning of the crown during restoration, and to avoid the interproximal contact being too tight (causing the crown to fail to be placed) or too loose (causing food impaction).

[0157] Extract the adjacent surface regions of the preparatory body and adjacent teeth from the segmentation mask (3D is point cloud, 2D is contour).

[0158] The KD tree algorithm is used to quickly calculate the minimum distance (3D model) or minimum pixel distance (2D image, which needs to be multiplied by the image scaling factor when converted to actual distance). The ideal clinical range is 0.1-0.3 mm.

[0159] A predetermined number (20) of cross sections are uniformly selected along the vertical direction of the proximal surface (the direction of crown height). The interproximal gap of each cross section is calculated, and all the obtained interproximal gaps are arranged in ascending order according to their numerical values. The minimum interproximal gap is extracted. The normal range of the minimum interproximal gap is preset. If the minimum interproximal gap is not within the normal range, the difference between the minimum interproximal gap and the normal range is calculated, and the absolute value is recorded as the minimum interproximal gap deviation.

[0160] After comprehensive analysis and processing of various evaluation indicators, a comprehensive evaluation coefficient is obtained, and the corresponding tooth preparation quality grade is matched.

[0161] Specifically, it includes:

[0162] The weighting factors for the pre-defined axial surface convergence deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation are calculated by multiplying the axial surface convergence deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, occlusal surface minimum space deviation, and occlusal surface minimum space deviation with their corresponding weighting factors, and then summing them to obtain the comprehensive evaluation coefficient.

[0163] By pre-setting weighting factors for core evaluation indicators such as axial plane aggregation deviation and shoulder curvature standard deviation, a comprehensive evaluation coefficient is generated through weighted summation, successfully solving the problem of the one-sidedness of evaluating the quality of tooth preparation with a single indicator.

[0164] Its weighting settings can accurately match clinical treatment needs, highlighting the key impacts of axial convergence and shoulder precision on crown retention and sealing, while also taking into account auxiliary dimensions such as occlusal space and interproximal space. This achieves systematic integration of multi-dimensional quality information, allowing the assessment results to comprehensively reflect the overall level of tooth preparation and avoiding the neglect of other potential treatment risks due to the achievement of a single indicator.

[0165] Transforming scattered quantitative indicators into intuitive comprehensive evaluation coefficients and corresponding quality levels significantly improves the practicality and standardization of clinical applications.

[0166] Traditional assessments rely on doctors' subjective experience and judgment, which are easily affected by individual differences. This method, however, establishes a standardized quality assessment system through a unified weighted calculation logic, reducing human error and making the assessment results more objective and comparable.

[0167] Meanwhile, a clear quality grading system can provide doctors with clear diagnostic and treatment references, help quickly locate deficiencies in the preparation process, adjust and optimize the plan in a timely manner, thereby improving the success rate of crown restoration, reducing the risk of postoperative complications, and forming a complete closed loop from image segmentation and index calculation to quality assessment, providing efficient and reliable decision support for oral diagnosis and treatment.

[0168] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0169] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0170] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0171] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0172] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0173] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0174] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0175] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0176] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology 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.

[0177] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for oral image segmentation for oral diagnosis and treatment, characterized in that, include: Acquire 3D / 2D oral cavity images, and after integrity verification, output standardized, high signal-to-noise ratio image data through preprocessing; The prepared tooth body, adjacent teeth, gingival margin, and prepared shoulder are segmented, and a segmentation mask is generated; The evaluation indicators include axial surface aggregation deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, minimum space deviation of the occlusal surface, and minimum interproximal clearance deviation. After comprehensive analysis and processing of various evaluation indicators, a comprehensive evaluation coefficient is obtained and matched with the corresponding tooth preparation quality grade. The process for obtaining the standard deviation of shoulder curvature and the coefficient of variation of shoulder width includes: extracting the edge contour of the prepared shoulder from the segmentation mask and discretizing the curve into a sequence of coordinate points; calculating the curvature of each point using the five-point numerical differentiation method for the discrete coordinate point sequence; calculating the standard deviation of shoulder curvature; and setting a maximum allowable threshold for the standard deviation of shoulder curvature. If the standard deviation of shoulder curvature is greater than the maximum allowable threshold, the difference between the two is calculated and denoted as the standard deviation of shoulder curvature. Quasi-deviation; extract the gingival margin contour from the segmentation mask, and use this contour as a reference, offsetting it by a preset distance towards the crown direction as a reference line; uniformly select a preset number of measurement points on the prepared shoulder contour, equidistantly distributed along the circumference of the shoulder, and measure the vertical distance from the prepared body surface to the reference line at each point; after obtaining each vertical distance, calculate the mean and standard deviation, and divide the standard deviation by the mean to obtain the shoulder width variation coefficient. The maximum allowable threshold for the shoulder width variation coefficient is preset. If the shoulder width variation coefficient is greater than the maximum allowable threshold, calculate the difference between the two, and record it as the shoulder width variation deviation coefficient.

2. The oral image segmentation method for oral diagnosis and treatment according to claim 1, characterized in that, Acquire 3D / 2D oral cavity images, perform integrity verification, and preprocess to output standardized, high signal-to-noise ratio image data, specifically including: Obtain a three-dimensional digital model or a two-dimensional color image of the oral cavity; Preprocessing of the 3D model includes coordinate alignment and region of interest clipping; The cropped model is then smoothed. Two-dimensional image preprocessing includes illumination normalization, color space conversion, image scaling and alignment.

3. The oral image segmentation method for oral diagnosis and treatment according to claim 1, characterized in that, The prepared tooth, adjacent teeth, gingival margin, and prepared shoulder are segmented, and a segmentation mask is generated. Specifically, this includes: Core segmentation target: tooth preparation; reference targets: adjacent teeth, gingival margin, prepared shoulder, occlusal surface; the segmentation network uses an improved U-Net++ network as its basic architecture; Channel attention and spatial attention mechanisms are added after each convolutional block of the encoder, and an edge-aware loss function is designed. A 1×1 convolution is used in the last layer of the decoder, and a batch normalization layer is added.

4. The oral image segmentation method for oral diagnosis and treatment according to claim 3, characterized in that, Also includes: Training the data includes dataset construction, data augmentation, and determining training parameters; Morphological optimization of the segmentation mask output by the network includes: Connectivity analysis: Remove isolated noise regions with an area less than 50 pixels; Edge smoothing: Morphological closing operations are used to fill the holes in the segmented area, and then median filtering is used to smooth the edges to ensure that the shoulder edge line is continuous without any breaks. Regional consistency check: Check the positional relationship between the prepared tooth and the adjacent tooth; Accuracy verification: For the segmentation results of each input image, the Dice coefficient, cross-union ratio and pixel accuracy of each target region are automatically calculated; Output: Generates a multi-channel segmentation mask, with each channel corresponding to a segmentation target. The mask corresponds one-to-one with the preprocessed original image.

5. The oral image segmentation method for oral diagnosis and treatment according to claim 1, characterized in that, The process of obtaining the axial plane convergence deviation includes: For 3D models: Point cloud data of the prepared tooth body is extracted from the segmentation mask, and the point cloud is divided into four axial regions: buccal, lingual, mesial, and distal, based on the tooth body's central axis. For each axial region, the point cloud is fitted to a plane using the least squares method; Calculate the angle between two planes based on their normal vectors. The axial convergence degree is obtained by adding the buccal-lingual tilt angle and the mesial-distal tilt angle. The normal range of the axial surface cohesion degree is preset, and the difference between the axial surface cohesion degree and the normal range is calculated and recorded as the axial surface cohesion deviation. For 2D images: If only a single 2D image is entered, two orthogonal images will be taken in the cheek-tongue direction and the near-to-mid direction; if multiple images have been entered, orthogonal view images will be automatically selected. For each image, extract the axial edge contour using the prepared tooth segmentation mask; Straight lines were fitted to the buccal, lingual, mesial, and distal margin contours to obtain the slope of each margin. The included angle is calculated based on the slope of the relative edges, the degree of aggregation is obtained by summing the angles, and the axial aggregation deviation is obtained.

6. The oral image segmentation method for oral diagnosis and treatment according to claim 1, characterized in that, The process of obtaining the minimum spatial deviation of the occlusal surface includes: Extract the occlusal point cloud or contour of adjacent teeth from the segmentation mask and fit it to the reference plane; A predetermined number of measurement points are evenly selected within the segmented area of ​​the occlusal surface of the prepared body; Calculate the vertical distance from each measurement point to the reference plane; Extract the minimum vertical distance and record it as the minimum space of the occlusal surface; preset the minimum space threshold of the occlusal surface. If the minimum space of the occlusal surface is less than the minimum space threshold, calculate the difference between the two and record it as the minimum space deviation of the occlusal surface.

7. The oral image segmentation method for oral diagnosis and treatment according to claim 1, characterized in that, The process of obtaining the minimum adjacent face gap includes: Extract the proximal regions of the preparatory body and adjacent teeth from the segmentation mask; The KD-tree algorithm is used to quickly calculate the minimum distance or minimum pixel distance between two adjacent surfaces; A preset number of cross sections are uniformly selected along the vertical direction of the adjacent surfaces. The gap between adjacent surfaces of each cross section is calculated, and the minimum gap between adjacent surfaces is extracted. The normal range of the minimum gap between adjacent surfaces is preset, and the difference between the minimum gap between adjacent surfaces and the normal range is calculated and recorded as the minimum gap between adjacent surfaces deviation.

8. The oral image segmentation method for oral diagnosis and treatment according to claim 7, characterized in that, After comprehensively analyzing and processing the various evaluation indicators, a comprehensive evaluation coefficient is obtained, and a corresponding tooth preparation quality grade is matched, specifically including: After setting weighting factors for axial surface aggregation deviation, shoulder curvature standard deviation, shoulder width variation deviation coefficient, minimum space deviation of occlusal surface, and minimum adjacent surface clearance deviation, a weighted summation is performed to obtain the comprehensive evaluation coefficient.

Images