A method and system for evaluating and warning the stability of maxillary expansion

Abstract

This invention relates to the fields of orthodontics and medical image processing, and particularly to a method and system for assessing and warning of stability after maxillary expansion surgery. The method includes: extracting an anti-artifact bone anchoring zone point cloud from maxillary cone-beam CT data; extracting a palatal mucosal surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring zone point cloud from intraoral scan data, and rigidly registering the anti-artifact bone anchoring zone point cloud with the palatal mucosal surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scan coordinate system to the cone-beam CT coordinate system; determining a local region in the cone-beam CT data based on the cross-modal mapping coordinates, and calibrating the position of the cross-modal mapping coordinates based on the cone-beam CT data to output the final marker coordinates. This invention can obtain high-precision final marker coordinates, improving detection accuracy, and has low computational cost, strong interpretability, and prevents subsequent width calculation distortion and false warnings.

Description

Technical Field

[0001] This invention relates to the fields of orthodontics and medical image processing, and in particular to a method and system for assessing and warning of stability after maxillary expansion surgery. Background Technology

[0002] Transverse hypoplasia of the maxilla is a common dentofacial deformity in clinical practice, and maxillary expansion surgery (including rapid expansion, slow expansion, and surgically assisted expansion) is the main treatment method. Postoperative recurrence rates are high, therefore quantitative assessment of postoperative stability and timely early warning are necessary. Existing technologies include automated assessment and early warning systems based on cone-beam computed tomography (CBCT) 3D reconstruction and deep learning. A typical workflow includes: collecting multi-temporal CBCT data from the patient postoperatively; automatically locating key anatomical landmarks of the maxilla and dental arch (such as the central fossa of the first molar, the cusp of the canine, and the endpoint of the mid-palatal suture) using a 3D convolutional neural network (such as 3D U-Net); calculating stability indicators such as arch width and recurrence rate; and triggering early warnings using statistical process control methods.

[0003] However, patients who have undergone maxillary expansion surgery often have various metal structures in their oral cavity: bone-supported rapid expanders include stainless steel expansion screws, connecting rods, and anchorage screws; tooth-supported expanders have metal bands and palatal hinges; some patients also wear orthodontic brackets, buccal tubes, or implant anchorage screws. These metal objects produce severe beam hardening artifacts and metal streak artifacts during CBCT scans, manifesting as bright and dark bands around the metal, blurred or broken bone boundaries, especially in the mid-palatal suture area near the central screw of the expander, where the CT grayscale value is severely interfered with. Traditional threshold-based bone boundary recognition methods fail. Deep learning models, trained primarily on CBCT samples with no or mild artifacts, experience a decrease in landmark localization accuracy when inputting images with severe metal artifacts, leading to distortion in subsequent width calculations and false alarms. Summary of the Invention

[0004] The main objective of this invention is to provide a method and system for assessing and warning of stability after maxillary expansion surgery, aiming to solve the technical problems mentioned in the background art.

[0005] This invention proposes a method for assessing and providing early warning of stability after maxillary expansion surgery, comprising: Acquire maxillary cone-beam computed tomography (CBCT) data and intraoral scan data at the same follow-up time. Anti-artifact bone anchoring zone point cloud was extracted based on the aforementioned maxillary cone-beam CT data; Based on the intraoral scanning data, the palatal mucosa surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring area point cloud is extracted, and the anti-artifact bone anchoring area point cloud and the palatal mucosa surface morphology point cloud are rigidly registered to obtain a cross-modal space mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system. Based on the intraoral scanning data, key landmarks for the efficacy of arch dilation are obtained, and their coordinates in the intraoral scanning coordinate system are obtained. The coordinates are projected onto the cone-beam CT coordinate system based on the cross-modal spatial mapping matrix to obtain the cross-modal mapping coordinates of the key landmark points of the arch expansion treatment in the cone-beam CT coordinate system; Based on the cross-modal mapping coordinates, a local region is determined in the cone-beam CT data, and the position of the cross-modal mapping coordinates is calibrated based on the cone-beam CT data to output the final marker coordinates; Based on the final marker coordinates, output stability assessment conclusions and early warning information.

[0006] This invention also provides a stability assessment and early warning system after maxillary expansion surgery, comprising: The data acquisition module is used to acquire maxillary cone-beam CT data and intraoral scan data at the same follow-up time. Anchor region extraction module is used to extract anti-artifact bone anchoring zone point cloud based on the maxillary cone-beam CT data; The spatial registration module is used to extract the palatal mucosa surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring area point cloud based on the intraoral scanning data, and to rigidly register the anti-artifact bone anchoring area point cloud with the palatal mucosa surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system. The marker detection module is used to obtain key markers for the efficacy of vasodilation based on the intraoral scanning data, and to obtain their coordinates in the intraoral scanning coordinate system. The coordinate projection module is used to project the coordinates onto the cone-beam CT coordinate system according to the cross-modal space mapping matrix, so as to obtain the cross-modal mapping coordinates of the key landmark points of the arch expansion treatment in the cone-beam CT coordinate system; The local calibration module is used to determine a local region in the cone-beam CT data based on the cross-modal mapping coordinates, and to perform position calibration on the cross-modal mapping coordinates based on the cone-beam CT data, and output the final marker coordinates; The evaluation and early warning module is used to output stability evaluation conclusions and early warning information based on the final marker point coordinates.

[0007] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for assessing and warning of stability after maxillary expansion surgery.

[0008] This application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for assessing and warning of stability after maxillary expansion surgery.

[0009] The beneficial effects of this invention are as follows: This invention utilizes the characteristic that intraoral scanning data is not affected by metal artifacts to detect key landmarks of arch vasodilation efficacy. Then, by rigidly registering intraoral scanning data with cone-beam CT data to establish a spatial mapping relationship, the coordinates of the landmarks are projected into the cone-beam CT space. Finally, calibration is performed in a small neighborhood near the projection position using the gray-scale gradient information of cone-beam CT, thereby obtaining high-precision final landmark coordinates. This improves detection accuracy, reduces computational load, enhances interpretability, and prevents subsequent width calculation distortion and false alarms. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0012] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0014] like Figure 1 As shown, this application provides a method for assessing and providing early warning of stability after maxillary expansion surgery, which includes: S1, acquire maxillary cone-beam CT data and intraoral scan data at the same follow-up time.

[0015] S2, Extract anti-artifact bone anchoring zone point cloud based on the maxillary cone-beam CT data.

[0016] S3. Extract the palatal mucosal surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring area point cloud based on the intraoral scanning data, and rigidly register the anti-artifact bone anchoring area point cloud with the palatal mucosal surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system.

[0017] S4. Obtain key landmarks for arch dilation efficacy based on the intraoral scanning data, and obtain their coordinates in the intraoral scanning coordinate system.

[0018] S5. Based on the cross-modal space mapping matrix, the coordinates are projected onto the cone-beam CT coordinate system to obtain the cross-modal mapping coordinates of the key landmark points for the arch expansion treatment in the cone-beam CT coordinate system.

[0019] After obtaining the cross-modal spatial mapping matrix, which contains rotation and translation parameters from the intraoral scanning coordinate system to the cone-beam CT coordinate system, for the coordinates of each key landmark point for arch vasodilator efficacy detected in the intraoral scanning coordinate system, the coordinates are treated as a three-dimensional spatial point. The rotation and translation operations defined by the matrix are applied sequentially, that is, the orientation of the point is adjusted according to the rotation component in the matrix, and the position of the point is moved according to the translation component, thereby calculating the new coordinates of the point in the cone-beam CT coordinate system. This new coordinate is the cross-modal mapping coordinate. The cross-modal mapping coordinate preserves the relative positional relationship of the original landmark point in the intraoral scanning data, and is transformed into the spatial coordinate system corresponding to the cone-beam CT data, providing an initial position for subsequent local search in the cone-beam CT data.

[0020] S6. Based on the cross-modal mapping coordinates, a local region is determined in the cone-beam CT data, and the position of the cross-modal mapping coordinates is calibrated based on the cone-beam CT data to output the final marker coordinates.

[0021] As described in steps S1-S6 above, this invention aims to solve the technical problem of anatomical landmark detection failure or significant accuracy reduction caused by cone-beam CT image artifacts generated by metal expanders and their accessories. This invention utilizes the characteristic that intraoral scan data is unaffected by metal artifacts to detect key landmarks for expander efficacy. Then, by rigidly registering intraoral scan data with cone-beam CT data to establish a spatial mapping relationship, the landmark coordinates are projected into the cone-beam CT space. Finally, calibration is performed using the grayscale gradient information of cone-beam CT in a small neighborhood near the projected location, thereby obtaining high-precision final landmark coordinates. This method achieves high-precision landmark detection in a metal artifact environment without relying on image artifact correction algorithms (which may introduce bone morphology distortion) or deep learning models (which have limited generalization ability to metal artifacts). Compared with existing methods that directly detect on cone-beam CT images, this approach uses intraoral scan data to bypass artifacts and gradually approximates the location of real bony landmarks through cross-modal registration and local calibration, thereby improving detection accuracy, reducing computational load, and enhancing interpretability.

[0022] S7, based on the final marker coordinates, output stability assessment conclusions and early warning information.

[0023] After obtaining the final landmark coordinates, intraoral scan data was used as an independent reference source to verify the consistency of cone-beam computed tomography (CBCT) results. Based on the verification results, a reliable data source was selected to calculate stability indices, and finally, stability assessment conclusions and early warning information were output. Specifically, firstly, the arch expansion width and recurrence rate were calculated. For the current follow-up time, the arch width was calculated based on the final landmark coordinates of the central fossa of the left and right first molars, and the canine width was calculated similarly. The recurrence rate was obtained by the percentage of lost arch expansion relative to the initial arch expansion. Secondly, the recurrence trend was fitted. Using multiple follow-up time points and their corresponding recurrence rates as input, an exponential decay model or a linear mixed-effects model was used to fit the recurrence trend curve to predict the recurrence rate at future time points. The width value and recurrence rate served as stability indices. Thirdly, early warning thresholds were set and early warning judgments were made. Recurrence rate data of a group of patients with good postoperative stability were collected at each follow-up time point. The mean and standard deviation of each time point were calculated, and the warning limit was set to the mean plus two standard deviations, and the control limit was set to the mean plus three standard deviations. If the current recurrence rate exceeds the control limit, a Level 1 warning is triggered; if the current recurrence rate is between the warning limit and the control limit, a Level 2 warning is triggered; if the predicted recurrence rate at the next follow-up time exceeds the control limit at that time, an early warning is triggered. Finally, a stability assessment report is output, including the stability assessment conclusion and warning information. The calculated width value, recurrence rate, warning level, and recurrence trend graph are integrated into a structured report and pushed to clinicians through the hospital information system. If there are low-confidence markers or data regressions in the aforementioned marker detection steps, these are noted in the report.

[0024] In one embodiment of the present invention, the step of extracting anti-artifact bony anchoring zone point cloud based on the maxillary cone-beam CT data includes: S21, Obtain the palatine horizontal plate region and bilateral nasal floor plate region in the middle and posterior part of the hard palate as the target anchoring area; S22, according to the target anchoring area, set the seed point position in the cone-beam CT data. The seed point position satisfies the following constraints: in the horizontal direction, it is limited to the left and right predetermined widths of the mid-palatal suture; in the sagittal direction, it is limited to the predetermined proportion range after the full length of the mid-palatal suture; and in the vertical direction, it is limited to the preset bone tissue grayscale threshold range of the hard palate cortex. S23. Based on the location of the seed point, a region growing algorithm is used to segment the anti-artifact bone anchoring area point cloud, starting from the seed point and using conditions that the gray level is greater than the preset bone tissue gray level threshold and the gradient is smooth.

[0025] As described in steps S21-S23 above, this invention further defines the step of extracting the point cloud of the anti-artifact bony anchoring region. In the cone-beam CT images of the maxilla after maxillary expansion surgery, the metal expander and its accessories will produce beam hardening artifacts and metal stripe artifacts, resulting in blurred bone tissue boundaries or false bright and dark stripes in the image. Conventional image segmentation methods, such as region growing based on global thresholds or segmentation algorithms based on edge detection, can work normally in artifact-free images, but when metal artifacts are present, due to gray-level anomalies and false boundaries caused by the artifacts, these methods have difficulty distinguishing between real bone boundaries and artifacts, resulting in segmentation results containing artifact noise or missing real bone tissue. To address this problem, this invention combines the spatial distribution pattern of metal artifacts after maxillary expansion surgery with the stability of anatomical structures, and designs a progressive processing path from region selection to coordinate constraints to conditional segmentation. This method can extract the point cloud of the bony region unaffected by artifacts in the cone-beam CT image of the maxilla contaminated with metal artifacts without relying on the image artifact correction algorithm, thus solving the problem that the bony reference region cannot be accurately extracted due to metal artifacts.

[0026] Specifically, in patients after arch expansion surgery, the central screw and anchorage screw of the metal expander are usually located in the anterior middle part of the maxillary arch or the anterior middle part of the palatal suture. The severity of metal artifacts is negatively correlated with the distance from the target area to the metal object: the closer to the metal object, the more severe the artifact; the farther away, the less severe the artifact. The horizontal plate region of the palatine bone in the middle and posterior part of the hard palate (i.e., the posterior third of the mid-palatal suture region) and the bilateral nasal floor plates are typically more than 30 mm away from the metal expander. According to the physical characteristics of X-ray beam hardening artifacts, the artifact intensity decreases exponentially with increasing distance. Therefore, the above-mentioned regions are least affected by artifacts in cone-beam CT images, and the gray value variation coefficient of the cortical bone boundary is significantly lower than that of the dental arch and the anterior region of the mid-palatal suture. In addition, the horizontal plate of the palatine bone, as the posterior plate of the hard palate, is flat and has continuous cortical bone, and has a consistent anatomical position among different individuals. The bilateral nasal floor plates are located on the lateral side of the nasal floor and are also fixed in position. The above two regions do not undergo morphological changes after expander surgery because expander surgery mainly affects the anterior middle part of the mid-palatal suture and the alveolar process, while the posterior hard palate and nasal floor regions do not participate in expander remodeling. Therefore, selecting the above-mentioned region as the target anchoring area can minimize the interference of metal artifacts on bone boundary recognition, so that the image region processed by subsequent segmentation operations has a low artifact intensity, thereby ensuring the authenticity of bone boundary information in the segmentation result.

[0027] After determining the anatomical type of the target anchoring area, the initial position of the seed point needs to be automatically located in the actual cone-beam CT volume data. Due to differences in head posture and scanning range among patients, absolute coordinates cannot be used directly. This step transforms prior anatomical knowledge into relative coordinate constraints. The lateral constraint is set within a predetermined width on both sides of the mid-palatal suture. This predetermined width is obtained by measuring the typical distance of metal artifact diffusion from the center of the metal object to both sides in the lateral direction based on clinical statistics, and using this distance as the predetermined width. For example, by measuring the artifact diffusion range of cone-beam CT images of a group of patients after arch expansion surgery, the 95th percentile of the distance from the visible edge of the artifact to the center of the metal object is taken as the predetermined width. This constraint ensures that the seed point is located outside the artifact-affected area, while excluding structures that may be affected by artifacts, such as the alveolar process and the lateral wall of the maxillary sinus. The sagittal constraint is set within a predetermined proportional range along the entire length of the mid-palatal suture. The predetermined ratio is obtained by measuring the entire length from the anterior to the posterior point of the palatine midline in the cone-beam CT image of the patient after palatal expansion. Then, calculating from posterior to front, a ratio is determined such that the proportion of visible metallic artifacts in the region before the corresponding anterior boundary (i.e., closer to the front) exceeds a preset threshold (e.g., 10%), while the proportion of artifacts in the region after this ratio (i.e., closer to the back) is below this threshold. For example, statistical analysis identifies the posterior 40% as a low-incidence area for artifacts. This constraint restricts the seed point to a safe area far from the metal expander. Vertically, the CT grayscale value of the voxel containing the seed point is required to be within a preset bone tissue grayscale threshold range (e.g., 200 to 1500 Henle units) to ensure that the seed point falls on the cortical bone layer, rather than soft tissue or medullary cavity. This grayscale threshold is obtained by statistically analyzing the grayscale distribution of the cortical bone region in a normal maxillary cone-beam CT image without artifacts, and taking the mean plus or minus a certain standard deviation as the threshold range. For example, the grayscale value of cortical bone typically ranges from 200 to 1500 Hen units. Therefore, the lower limit of the preset bone tissue grayscale threshold range is set to 200 Hen units, and the upper limit is set to 1500 Hen units to distinguish the cortical bone from the soft tissue above and the cancellous bone below. Through these three constraints, the seed point is located within the cortical bone region free from artifact contamination. This step automates the seed point localization process, eliminating the need for manual intervention and preventing the seed point from falling into soft tissue, medullary cavity, or artifact noise due to artifact interference, thus ensuring the correct starting point for subsequent regional growth.

[0028] After obtaining the seed point, it needs to be expanded into a continuous point cloud covering the entire anchoring region. This step uses a region growing algorithm, which starts from the seed point and checks whether adjacent voxels meet the growth conditions. If they do, they are included in the region and the expansion continues outward. The growth conditions consist of two sub-conditions, both of which must be met simultaneously. The first sub-condition is that the voxel's gray value is greater than a preset bone tissue gray value threshold. This threshold is obtained using the same method as the threshold in the vertical constraint. This condition ensures that the included voxels belong to bone tissue and not soft tissue or air. The second sub-condition is gradient smoothing, meaning that the gray-level gradient magnitude within the voxel's neighborhood is less than a preset gradient threshold. This preset gradient threshold is obtained by statistically analyzing the gray-level gradient distribution of artifact-free normal bone tissue regions and taking the mean of the gradient magnitude plus a certain number of standard deviations as the upper limit. A typical characteristic of metal artifacts is that the gray-level values ​​oscillate violently in space, resulting in a large local gradient magnitude, while the gray-level changes of real bone tissue are gradual, resulting in a small local gradient magnitude. Therefore, the gradient smoothing condition can effectively filter out isolated bright or dark voxels caused by artifacts. When both conditions work together, only voxels that simultaneously meet the grayscale threshold and gradient smoothing requirements will be included in the anchoring region point cloud. This segmentation method differs from conventional region growing: conventional region growing typically only uses the grayscale threshold condition, which works normally in artifact-free images, but in images with metal artifacts, it will include artifact noise as well, because the grayscale values ​​of some voxels in the artifact region may also fall within the grayscale range of bone tissue; this step introduces an additional gradient smoothing condition, so that the segmented anchoring region point cloud does not contain artifact noise, while preserving the true bone tissue boundary, thereby obtaining a highly reliable bony reference region.

[0029] In one embodiment of the present invention, the step of rigidly registering the anti-artifact bone anchoring region point cloud with the palatal mucosa surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system includes: S31, the iterative nearest point algorithm is used to iteratively solve for the rotation matrix and translation vector that minimizes the average Euclidean distance between the point cloud of the anti-artifact bone anchoring zone and the point cloud of the palatal mucosa surface morphology. The objective function of the iterative nearest point algorithm is: ; In the formula, p i q represents the i-th point in the point cloud of the palatal mucosa surface morphology (as the source point cloud). i This indicates that the anti-artifact bone anchoring area point cloud is related to the p i The corresponding i-th point (as the target point cloud), n represents the total number of point pairs involved in the calculation, R represents the rotation matrix, and t represents the translation vector. The rotation matrix R and the translation vector t are obtained iteratively by minimizing the above objective function. S32, construct a cross-modal space mapping matrix from the intraoral scan coordinate system to the cone-beam CT coordinate system based on the rotation matrix and translation vector: X CBCT =R·X IOS +t In the formula, X IOS This represents the coordinates of a point in the intraoral scanning coordinate system, X. CBCT This represents the coordinates of the corresponding point in the cone-beam CT coordinate system after the transformation.

[0030] As described in steps S31-S32 above, this invention further specifies the step of obtaining the cross-modal spatial mapping matrix. After extracting the anti-artifact bone anchoring area point cloud, it is necessary to spatially align the bone anchoring area with the corresponding anatomical structure in the intraoral scanning data, thereby establishing a coordinate transformation relationship from the intraoral scanning coordinate system to the cone-beam CT coordinate system. Since the intraoral scanning data records the surface morphology of the palatal mucosa, while the anti-artifact bone anchoring area point cloud in the cone-beam CT data records the surface morphology of the cortical bone, the two do not overlap spatially and there is an offset caused by the mucosal thickness. If conventional global registration methods are used directly (such as registering the point cloud of the entire dentition with the point cloud of the entire maxilla), the point cloud of the dentition and alveolar process region in the cone-beam CT is unreliable due to metal artifacts, resulting in a large registration error. To address this scenario, an iterative nearest-point algorithm is employed to perform rigid registration using the anti-artifact bony anchoring zone point cloud and the corresponding palatal mucosal surface morphology point cloud from intraoral scan data. A specific objective function is used to solve for the rotation matrix and translation vector, ultimately constructing a cross-modal spatial mapping matrix. This matrix accurately transforms the coordinates of key landmarks for arch vasodilation therapy detected in the intraoral scan coordinate system to the cone-beam CT coordinate system, providing accurate initial positions for subsequent local gradient boundary calibration. This solves the technical problem of inaccurate cross-modal data alignment caused by metal artifacts, avoiding registration failures or accuracy degradation due to artifact contamination during full-map registration.

[0031] Specifically, after extracting the anti-artifact bony anchoring zone point cloud, a set of point cloud data in the cone-beam CT coordinate system was obtained, representing the cortical bone surface of the posterior hard palate and nasal floor region. Simultaneously, a palatal mucosal surface morphology point cloud was extracted from the intraoral scan data. This point cloud, located in the intraoral scan coordinate system, represents the palatal mucosal surface covering the aforementioned cortical bone. The two sets of point clouds have the following relationships: anatomically, the palatal mucosal surface and the underlying cortical bone surface have similar geometric contours, with only spatial offset caused by mucosal thickness; in terms of data quality, the anti-artifact bony anchoring zone point cloud is almost unaffected by metal artifacts, and the palatal mucosal surface morphology point cloud is also unaffected by metal artifacts. Therefore, setting the anti-artifact bony anchoring region point cloud as the target point cloud and the palatal mucosa surface morphology point cloud as the source point cloud is to solve the rigid transformation of the source point cloud to the space where the target point cloud is located through the registration algorithm, so that the two sets of point clouds on which the subsequent registration depends both come from regions without artifacts or with low artifacts, thereby ensuring the reliability of the registration process.

[0032] The execution process of the iterative nearest point algorithm is as follows: In each iteration, firstly, based on the current transformation parameters, the closest point in the target point cloud is searched for each point in the source point cloud according to its Euclidean distance, forming point pairs; then, based on these point pairs, the new transformation parameters (rotation matrix and translation vector) are solved using the least squares method, reducing the value of the objective function; the above process is repeated until the convergence condition is met (e.g., the change in transformation parameters is less than a preset threshold or the preset maximum number of iterations is reached). In this scheme, since the two sets of point clouds are roughly corresponding in anatomical position and the initial pose difference is small, the algorithm can converge within a relatively small number of iterations. Through this numerical optimization method, the optimal rotation matrix and translation vector for transforming points in the intraoral scanning coordinate system to the cone-beam CT coordinate system can be quantitatively solved. Furthermore, since the point clouds involved in the calculation are not affected by metal artifacts, the solved transformation parameters have high accuracy.

[0033] After obtaining the rotation matrix and translation vector, a complete rigid body transformation expression is constructed. For point coordinates in the intraoral scanning coordinate system, the corresponding point coordinates in the cone-beam CT coordinate system can be calculated through this transformation. The physical meaning of this cross-modal spatial mapping matrix is: to transform all points in the entire intraoral scanning space (including key landmarks for arch expansion efficacy) into the cone-beam CT space according to the geometric constraints of the periosteum interface. The construction of this mapping matrix does not rely on information about metal artifact regions, but only on the point cloud of the anti-artifact bony anchoring zone with almost no artifacts and its corresponding palatal mucosal surface morphology point cloud. This allows the coordinates of key landmarks for arch expansion efficacy detected in subsequent intraoral scanning data to be accurately projected into the cone-beam CT coordinate system, and the spatial deviation between the projected position and the actual bony landmark is controlled within a small range. Compared with the existing method of directly using the entire dentition or the entire maxilla point cloud for registration, this scheme avoids areas contaminated by artifacts, resulting in lower registration errors.

[0034] It should be noted that the cross-modal spatial mapping matrix in this embodiment is constructed using a direct rigid registration method, aligning the point cloud of the palatal mucosal surface morphology with the point cloud of the anti-artifact bony anchoring zone, without explicit compensation for the mucosal thickness between the two. This is because the core assessment indicators (extension width, recurrence rate, and width change) of this invention are based on the relative changes of the same patient at different follow-up times rather than absolute measurements. For example, the systematic bias generated by registration at the first follow-up time is Δ1 (determined by both mucosal thickness and the registration algorithm), so the coordinates of the marker point measured at that time are the true coordinates plus Δ1; at subsequent follow-up times, since the anatomical structure (such as hard palate mucosal thickness) of the same patient is basically the same in the short term, and the registration algorithm and data acquisition conditions are consistent, the systematic bias Δ1 remains basically unchanged. Therefore, not explicitly compensating for mucosal thickness will not affect the accuracy of recurrence rate calculation and early warning judgment. This method meets the accuracy requirements of clinical assessment of post-extension stability while maintaining the simplicity of the algorithm.

[0035] In one embodiment of the present invention, the step of obtaining key landmarks for arch vasodilator efficacy based on the intraoral scan data and obtaining their coordinates in the intraoral scan coordinate system includes: S41, Based on the intraoral scanning data, a three-dimensional mesh model is constructed, and the maximum principal curvature and minimum principal curvature of each vertex on the mesh model are calculated, wherein the maximum principal curvature and minimum principal curvature are obtained by calculating the eigenvalues ​​of the Hessian matrix after fitting a quadratic surface to the local neighborhood of the vertex of the mesh model. S42, calculate the shape index of each vertex based on the maximum and minimum principal curvatures, using the following formula: ; Where SI represents the shape index, k1 represents the maximum principal curvature, and k2 represents the minimum principal curvature. When SI approaches -1, it represents a bowl-shaped surface (central fossa), and when SI approaches +1, it represents a cap-shaped surface (cusp). S43, based on the shape index, the vertex regions with a shape index less than a first preset shape index threshold are marked as central fossa candidate regions, and the vertex regions with a shape index greater than a second preset shape index threshold are marked as cusp candidate regions; S44, obtain the expected positional sequence information of teeth on the dental arch and the preset geometric template of key landmark points for arch expansion treatment; S45, based on the expected positional order information, select the point that is closest to the geometric template of the key landmark point for arch expansion treatment in each of the central fossa candidate regions and the cusp candidate regions as the key landmark point for arch expansion treatment, and obtain its coordinates in the intraoral scanning coordinate system.

[0036] As described in steps S41-S45 above, this invention further defines the steps for detecting key landmarks of arch expansion efficacy based on surface geometric features. After constructing the cross-modal spatial mapping matrix, it is necessary to detect key landmarks of arch expansion efficacy on intraoral scanning data to obtain the coordinates of these landmarks in the intraoral scanning coordinate system. Since a metal expander and its accessories are present in the patient's mouth after maxillary arch expansion surgery, but intraoral scanning technology is based on structured light or confocal microscopy principles and does not involve X-rays, it is not affected by metal artifacts. However, intraoral scanning data is a three-dimensional mesh model that only records the surface morphology of the crown and palatal mucosa, and does not contain skeletal information. How to automatically and accurately locate anatomical landmarks directly related to arch expansion efficacy (such as the central fossa of the first molar and the cusp of the canine) on intraoral scanning data is the core problem that needs to be solved in this step. Conventional methods may rely on manual point selection by the operator, which is inefficient and subject to observer error; or they may rely on deep learning models for key point detection, but this requires a large amount of labeled data and has limited generalization ability to different dentition morphologies. To address this scenario, this invention utilizes the geometric features of the crown surface (the central fossa is bowl-shaped and the cusp is cap-shaped) to screen candidate regions by calculating the shape index. Then, it combines the expected positional order of the teeth on the dental arch with a preset geometric template for precise matching. This allows for the automatic acquisition of the coordinates of key landmarks for arch expansion efficacy without relying on deep learning models or manual intervention. This solves the technical problem of automatically detecting key landmarks for arch expansion efficacy in intraoral scan data and provides a high-precision initial position for subsequently projecting the landmark coordinates onto cone-beam CT space.

[0037] Specifically, intraoral scan data is stored as a triangular mesh (STL format) or point cloud. First, it needs to be converted into a 3D mesh model with topological connections, where each vertex contains 3D spatial coordinates and each edge records the connections between vertices. After obtaining the mesh model, the maximum and minimum principal curvatures are calculated for each vertex. The principal curvatures are calculated as follows: For a given vertex, all its neighboring vertices in its local neighborhood (e.g., a one-ring or two-ring neighborhood) are taken, and a quadratic surface is fitted to this local region (e.g., using moving least squares or solving the locally parameterized Hessian matrix). Then, the principal curvatures of this quadratic surface at the vertex are calculated, which are the eigenvalues ​​of the Hessian matrix. The eigenvalue with the larger absolute value is the maximum principal curvature, and the one with the smaller absolute value is the minimum principal curvature. The sign of the principal curvature indicates the bending direction of the surface: a positive value indicates a convex surface, and a negative value indicates a concave surface. On the crown surface, the central fossa region is concave (k1 and k2 are both negative or one negative and one positive but the absolute value of the negative is larger), and the cusp region is convex (k1 and k2 are both positive). This step can transform the original three-dimensional mesh model into geometric feature quantities (principal curvatures) at each vertex, providing basic data for subsequent shape index calculations.

[0038] The shape index SI of each vertex is calculated based on the maximum and minimum principal curvatures. The shape index is a dimensionless quantity used to describe the local shape of a surface, with a value range of [-1, 1]. When SI approaches -1, the surface is bowl-shaped (concave, i.e., a central fossa feature); when SI approaches +1, the surface is cap-shaped (convex, i.e., a cusp feature); and when SI approaches 0, the surface is saddle-shaped or planar. In this formula, k1 and k2 are the principal curvatures calculated in the first step, and it is agreed that k2 ≥ k1 to ensure the denominator is non-negative. The range of the arctan function is (-2 / π, 2 / π), which, after multiplying by 2 / π, maps to (-1, 1). This formula characterizes the shape type of the surface through the ratio of the two principal curvatures, unaffected by the absolute magnitude of the curvatures (i.e., unrelated to the steepness of the surface, only related to the shape type). It can transform the principal curvature pair into a single shape index, allowing the geometry of each vertex to be quantified as a scalar value, facilitating threshold setting for classification. Compared to directly using principal curvature values, the shape index is more robust to noise and local scale variations because it depends only on the ratio of curvatures rather than their absolute values. In intraoral scans of patients after arch expansion, local noise may be generated on the crown surface due to obstruction by metal brackets or expander attachments; the ratioistic properties of the shape index can suppress the effects of this noise to some extent.

[0039] After obtaining the shape index of each vertex, it is necessary to filter the vertex regions that may belong to the central fossa or cusp based on the numerical range of the shape index. The first and second preset shape index thresholds are obtained as follows: Intraoral scan data of a normal dentition are collected, and an orthodontic expert manually marks the positions of the central fossa of the first molar and the cusps of the canines. Then, the shape index distribution at these marked points is statistically analyzed. The shape index at the central fossa is usually concentrated between -0.9 and -0.7, so the first preset shape index threshold can be set to -0.8 (i.e., areas with SI < -0.8 are marked as candidate regions for the central fossa); the shape index at the cusp is usually concentrated between 0.7 and 0.9, so the second preset shape index threshold can be set to 0.8 (i.e., areas with SI > 0.8 are marked as candidate regions for the cusp). These thresholds can be fine-tuned according to the actual data distribution. After threshold filtering, several connected regions are obtained, each corresponding to a possible central fossa or cusp location. Because there may be other depressions or protrusions on the crown surface (such as developmental grooves, marginal ridges, etc.), the candidate regions may contain false positives of non-target structures. The technical effect of this step is that it transforms the geometric shape detection problem into a threshold classification problem, enabling the initial localization of the central fossa and cusp, and significantly narrowing the search range for subsequent precise matching.

[0040] After obtaining candidate regions for the central fossa and cusps, it is necessary to identify the true key landmarks for arch expansion efficacy from these candidate regions (i.e., the central fossa of the left and right first molars and the cusps of the left and right canines). This step utilizes two constraints for precise matching. The first constraint is the expected positional order of the teeth on the dental arch. The dental arch morphology has a relatively stable order: from the midline to both sides, the order is central incisor, lateral incisor, canine, first premolar, second premolar, first molar, and second molar. Therefore, the central fossa of the first molar should be located in the posterior region of the dental arch, and the cusps of the canines should be located in the anterior region of the dental arch (behind the incisors and in front of the premolars). By calculating the dental arch curve of the entire intraoral scan data (e.g., by fitting the dental arch morphology using the least squares method), the approximate positional range of each tooth can be determined, thus limiting the candidate regions to the corresponding ranges. The second constraint is a pre-defined geometric template for key landmarks for arch expansion efficacy. The method for constructing the geometric template of key landmarks for arch expansion efficacy is as follows: A set of standard dentition samples (e.g., intraoral scan data collected from healthy volunteers) is selected, and the positions of the central fossa of the first molar and the cusps of the canines are manually marked. The relative position vectors between these landmarks are calculated (e.g., the vector from the central fossa of the left first molar to the cusp of the left canine, the vector from the central fossa of the left first molar to the central fossa of the right first molar, etc.), forming the geometric template of key landmarks for arch expansion efficacy. In the intraoral scan data to be tested, for each candidate region, its geometric similarity to the corresponding type of landmark in the template is calculated. For example, by comparing local curvature distribution or relative distance, the candidate point with the highest similarity is selected as the final landmark. Simultaneously, the nearest distance principle is applied, i.e., the candidate point closest to the center of the expected position interval is selected. These two constraints work together to output the coordinates of the final key landmarks for arch expansion efficacy in the intraoral scan coordinate system. This allows for accurate identification of the target landmark from multiple candidate regions using prior knowledge of the dental arch order and the geometric constraints between landmarks, eliminating false detections of interfering structures such as developmental grooves and marginal ridges.

[0041] It should be noted that the criteria for determining whether the geometric template of the key landmark points for arch expansion efficacy matches the target data are as follows: For the intraoral scan data to be tested, a set of candidate points is selected from the candidate regions selected based on the expected position of the dental arch (for example, the point closest to the dental arch curve is selected from each candidate region, resulting in a total of 4 candidate points: candidate point of the central fossa of the left first molar, candidate point of the central fossa of the right first molar, candidate point of the cusp of the left canine, and candidate point of the cusp of the right canine). The actual distance vector between these 4 candidate points is calculated and compared with the corresponding distance vector in the template. If the absolute value of the difference between the actual value of each vector and the standard value of the template does not exceed twice the corresponding standard deviation (i.e., the 95% confidence interval), then the set of candidate points is determined to match the geometric template and is used as the final key landmark point for arch expansion efficacy. If there is no match, other combinations of candidate points are tried in turn (for example, the next closest point is selected from the candidate regions) until a combination that meets the matching criteria is found. If all combinations do not match, manual review is prompted or other methods (such as the point of maximum curvature) are used as landmark points.

[0042] In one embodiment of the present invention, the step of determining a local region in the cone-beam CT data based on the cross-modal mapping coordinates, performing position calibration on the cross-modal mapping coordinates based on the cone-beam CT data, and outputting the final marker coordinates includes: S61, Based on the cross-modal mapping coordinates, take a cubic neighborhood of a predetermined size in the cone-beam CT volume data with the cross-modal mapping coordinates as the center; S62, Calculate the gray-level variance in the neighborhood of the cube; S63, determine whether the grayscale variance is lower than a first preset variance threshold or higher than a second preset variance threshold: S631, if the grayscale variance is lower than the first preset variance threshold or higher than the second preset variance threshold, it is determined that the area is severely damaged by metal artifacts, and the cross-modal mapping coordinates are directly used as the final marker point coordinates. S632, if the gray-level variance is not lower than the first preset variance threshold and not higher than the second preset variance threshold, then calculate the gray-level gradient modulus of each voxel in the neighborhood of the cube, fit a quadratic surface in the neighborhood of the cube according to the gray-level gradient modulus, search for the local maximum point of the gray-level gradient modulus, and use the coordinates of the local maximum point as the final marker point coordinates.

[0043] As described in steps S61-S63 above, this invention obtains cross-modal mapping coordinates, which represent the initial estimated positions of key bony landmarks in cone-beam CT space for arch vasodilation treatment. However, due to registration errors in the cross-modal mapping matrix (mainly due to individual differences in palatal mucosal thickness and residual errors in the iterative nearest-point algorithm), and slight changes in patient position during intraoral scanning and cone-beam CT acquisition, there is a certain deviation (usually less than 0.5 mm) between these cross-modal mapping coordinates and the actual bony landmark positions. Furthermore, in some cases, these cross-modal mapping coordinates may fall precisely within a local grayscale anomaly region caused by metal artifacts, causing their position to deviate from the actual bone boundary. Therefore, it is necessary to calibrate the cross-modal mapping coordinates to align them with the actual bony boundaries in the cone-beam CT image. To address this scenario, this invention calculates the grayscale gradient modulus within a very small local neighborhood (with a side length not exceeding 5 voxels) centered on the cross-modal mapping coordinates, and searches for the local maximum point of the gradient modulus as the final landmark coordinates. Meanwhile, this method detects whether the region is severely damaged by metal artifacts by calculating the local gray-level variance. If the variance is too low, it is determined that reliable calibration cannot be performed, calibration is abandoned, and a low-confidence label is attached. This technical solution directly solves the technical problem that cross-modal mapping coordinates are inconsistent with the real bone boundary due to registration residual errors or local artifact interference. It achieves sub-voxel-level calibration without introducing a global search, and provides a degradation processing mechanism for extreme artifact cases.

[0044] After obtaining the cross-modal mapping coordinates, the search range for the calibration operation needs to be determined. Since the deviation between the cross-modal mapping coordinates and the actual bony landmarks is already controlled within a small range (typically within 0.5 mm), a global search across the entire cone-beam CT volume data is unnecessary; calibration only needs to be performed within a local neighborhood. The predetermined size of this cubic neighborhood is determined as follows: based on the statistical distribution of the registration error of the cross-modal spatial mapping matrix, the 99th percentile of the error distribution is taken as the half-side length of the neighborhood, ensuring that the probability of the actual bony landmark being located within this neighborhood exceeds 99%. For example, if the standard deviation of the registration error is 0.15 mm, the half-side length can be taken as 3 times the standard deviation (0.45 mm). Considering that the cone-beam CT voxel size is typically 0.3 mm, the half-side length corresponds to approximately 1.5 voxels. For ease of calculation, an integer voxel length can be used, such as a side length of 5 voxels (i.e., extending 2 voxels outwards from the center in each direction, for a total side length of 5 voxels). This neighborhood size is sufficient to cover the registration error range while being small enough to avoid introducing artifact noise from distant locations. This step limits the calibration search range to a very small area, ensuring that true bony landmarks are within the search range while avoiding the erroneous inclusion of distant streaks caused by metal artifacts. Using a larger neighborhood (e.g., with a side length exceeding 10 voxels) may include spurious gradient extrema caused by metal artifacts in the search range, leading to calibration failure; using a smaller neighborhood (e.g., with a side length of 3 voxels) may exclude true bony landmarks from the search range due to larger registration errors.

[0045] In cases of severe metal artifacts, the local neighborhood where the cross-modal mapping coordinates lie may be completely contaminated by artifacts, manifesting as almost constant grayscale values ​​(e.g., uniform dark or bright bands caused by artifacts) or irregular oscillations (e.g., striped grayscale fluctuations caused by artifacts). In this situation, searching for the local maximum of the gradient modulus within this region may yield a false location (e.g., any point in a uniform region or the edge of an artifact stripe), resulting in a calibration result worse than the original cross-modal mapping coordinates. Therefore, this step introduces a quality control mechanism. First, the grayscale variance of all voxels within the cube's neighborhood is calculated. Then, it is determined whether this grayscale variance is lower than a first preset variance threshold or higher than a second preset variance threshold. The first and second preset variance thresholds are obtained as follows: For a set of artifact-free normal maxillary cone-beam CT images, the grayscale variance distribution near the bone boundary is calculated. The 5th percentile of this distribution is taken as the first preset variance threshold (i.e., the minimum variance of the normal bone boundary region), and the 95th percentile is taken as the second preset variance threshold (i.e., the maximum variance of the normal bone boundary region). When the calculated grayscale variance is lower than the first preset variance threshold, it indicates that the grayscale variation in this region is minimal, and there is no clear bone boundary. This is highly likely due to image information corruption caused by uniform dark or bright band artifacts. When the grayscale variance is higher than the second preset variance threshold, it indicates that the grayscale oscillation in this region is severe, and there is no stable bone boundary. This is highly likely due to image information corruption caused by stripe artifacts. In both cases, the system abandons calibration and directly uses cross-modal mapping coordinates as the final marker coordinates. A low-confidence marker is added to this marker for reference by the subsequent stability assessment and early warning system (e.g., reducing the weight of this marker when calculating the recurrence rate, or prompting the doctor to manually verify). If the grayscale variance is not lower than the first preset variance threshold and not higher than the second preset variance threshold, a clear bone tissue boundary is determined to exist in this neighborhood, and subsequent gradient modulus calculation and local maximum search are performed. This step provides a fault-tolerant mechanism for two extreme artifact cases, avoiding the introduction of larger errors due to calibration in unreliable image regions. Compared to existing methods that directly use deep learning to detect landmarks on CBCT, this approach maintains high localization accuracy even in the presence of metal artifacts. This is because the calibration in this approach is performed within a small neighborhood guided by intraoral scan data, which is already close to the true location, rather than searching the entire image, thus avoiding interference from distant artifacts. Compared to existing methods that rely solely on registration without calibration, this approach compensates for residual registration errors and individual differences in soft tissue thickness through local gradient search, further improving accuracy.

[0046] Then, based on the cube's neighborhood, the gray-level gradient magnitude of each voxel within that neighborhood is calculated. After obtaining the gray-level gradient magnitude of each voxel within the cube's neighborhood, the location with the largest gradient magnitude needs to be found. This location corresponds to the steepest gray-level change boundary in the image, i.e., the interface between bone tissue and non-bone tissue. Since voxel coordinates are discrete, directly taking the voxel with the largest gradient magnitude as the calibration result can only achieve voxel-level accuracy (the error is approximately half the voxel size, i.e., 0.15 mm). To achieve sub-voxel-level accuracy, this step uses a quadratic surface fitting method. The specific process is as follows: Taking the voxel with the largest gradient magnitude as the center, several voxels around it (e.g., a 3×3×3 neighborhood) are taken, and a quadratic surface function is fitted to the coordinates and gradient magnitude values ​​of these voxels. The coefficients are solved using the least squares method, and then the coordinates of the extreme points of the quadratic surface (i.e., the points where the first derivative is zero) are calculated. These extreme point coordinates are the final marker point coordinates after sub-voxel calibration. The technical effect of this step is to improve the accuracy of marker positioning from the voxel level to the sub-voxel level, meeting the sub-millimeter accuracy requirement for post-expansion arch stability assessment. If quadratic surface fitting is not performed and the pixel position is directly rounded, the calibrated coordinates may have a systematic deviation of half a voxel from the true boundary. This deviation may affect the accuracy of recurrence rate calculations due to the accumulation of multiple follow-ups.

[0047] In one embodiment of the present invention, the following steps are further included: S64, based on the final marker coordinates, calculate the second width distance between the key markers of the left and right symmetrical arch vasodilation efficacy at the current follow-up time, and obtain the first width distance between the key markers of the left and right symmetrical arch vasodilation efficacy based on the intraoral scan data. S65, obtain the historical width distance corresponding to the previous follow-up time, and calculate the first width change based on the historical width distance and the second width distance; obtain the historical oral scan width distance corresponding to the previous follow-up time, and calculate the second width change based on the historical oral scan width distance and the first width distance; S66, compare the first width change with the first change threshold, and compare the second width change with the second change threshold: S67, if the first width change is greater than the first change threshold and the second width change is less than the second change threshold, it is determined to be abnormal, and the first width distance is used to replace the second width distance, and the data source is marked as intraoral scanning in the output result.

[0048] As described in steps S64-S67 above, this invention adds steps for anomaly detection and data rollback. After completing local gradient boundary calibration and obtaining the final landmark coordinates, the precise location of the key landmarks for arch vasodilation efficacy in the cone-beam CT coordinate system has been obtained. However, in clinical practice, there are two situations that may lead to unreliable final landmark coordinates: First, if the local neighborhood grayscale variance of the cross-modal mapping coordinates is lower than a preset variance threshold, the system will abandon calibration and add a low-confidence marker. In this case, the final landmark coordinates are actually equal to the cross-modal mapping coordinates, and their accuracy is lower than the result after normal calibration. Second, even if the system does not trigger a low-confidence marker, due to extreme metal artifacts or unexpected situations such as patient head movement, the true location of the bony landmarks in the cone-beam CT data may deviate significantly from the location of the landmarks detected in the intraoral scan data, resulting in distortion of the width distance calculated based on the cone-beam CT data. If unreliable final landmark coordinates are directly used for subsequent stability assessment, it may lead to incorrect recurrence rate calculations and false alarms. Therefore, this invention compares the width distance directly measured from intraoral scan data (unaffected by metal artifacts) with the width distance calculated from cone-beam CT data. By analyzing whether the trends of the two measurements over time are consistent, the reliability of the cone-beam CT data is determined, and the system automatically reverts to using intraoral scan data when an anomaly is detected. This solves the technical problem of ensuring the validity of stability assessment data when the coordinates of marker points in cone-beam CT data are unreliable due to artifacts or unexpected situations, thus avoiding misjudgments caused by data quality issues.

[0049] After obtaining the final landmark coordinates at the current follow-up time, the second width distance between the left and right symmetrical key landmarks for arch expansion efficacy is first calculated. Left and right symmetrical key landmarks for arch expansion efficacy refer to paired anatomical landmarks, such as the central fossa of the left and right first molars, or the cusps of the left and right canines. For each pair of landmarks, the second width distance is calculated using Euclidean distance, utilizing the final coordinates of the left and right landmarks (located in the cone-beam CT coordinate system). This distance represents the transverse width after arch expansion measured based on cone-beam CT data. Simultaneously, the first width distance between the same pair of left and right symmetrical key landmarks for arch expansion efficacy is directly measured based on intraoral scan data. The coordinates of the key landmarks for arch expansion efficacy in the intraoral scan coordinate system have already been detected in the intraoral scan data. Since intraoral scan data is not affected by metal artifacts and its measurement accuracy is higher than that of cone-beam CT, the Euclidean distance is directly calculated from these coordinates to obtain the first width distance. The first width distance represents the transverse width after arch expansion measured based on intraoral scan data and can be used as a reference benchmark. This step obtains the measurement value of the same physical quantity from two independent data sources (cone-beam CT and intraoral scan), providing redundant information for subsequent anomaly detection.

[0050] During follow-up after arch expansion surgery, a set of width distance data was recorded at each follow-up visit. Let D1 be the second width distance at the previous follow-up time (e.g., 1 month post-surgery), and D2 be the second width distance at the current follow-up time (e.g., 3 months post-surgery). The first change is the difference between D2 and D1, reflecting the trend of width change over time based on cone-beam CT data. Under normal circumstances, the width may remain stable or experience a slight recurrence (width reduction) after arch expansion surgery; therefore, the first change should be a small amount close to zero or negative. Similarly, let D3 be the first width distance measured by intraoral scanning at the previous follow-up time, and D4 be the first width distance at the current time. The second change is the difference between D3 and D4. Since intraoral scanning measurements are not affected by metal artifacts, their trend reflects the true width change. This step converts absolute distance measurements into changes, eliminating systematic biases caused by factors such as patient head position and scanning parameters between different follow-up times, making cone-beam CT data comparable to intraoral scanning data.

[0051] This step sets two thresholds to determine if there are any abnormalities in the cone-beam CT data. The first and second thresholds are obtained as follows: The first threshold is set based on the physiological fluctuation range of width after normal arch expansion surgery. By statistically analyzing the cone-beam CT data of a group of patients with good postoperative stability (i.e., no abnormal recurrence or measurement error), the distribution of width change between two adjacent follow-up visits is calculated, and the 95th percentile of this distribution is taken as the first threshold. For example, if the absolute value of width change is less than 0.5 mm in 95% of normal patients, the first threshold can be set to 0.5 mm. The second threshold is set based on the repeatability accuracy of the intraoral scanner. The repeatability accuracy of the intraoral scanner is usually provided by the equipment manufacturer (e.g., ±0.05 mm). Considering the variability in clinical operation, the second threshold can be set to 2 to 3 times this accuracy (e.g., 0.15 mm).

[0052] The logic for anomaly determination is as follows: When the absolute value of the first change (width change measured by cone-beam CT) is greater than the first change threshold, while the absolute value of the second change (width change measured by intraoral scan) is less than the second change threshold, it indicates that the intraoral scan data shows a stable width (in line with physiological expectations), while the cone-beam CT data shows a significant change in width. Since intraoral scan data is reliable, this inconsistency can only be explained by an abnormality in the registration or calibration of the cone-beam CT data (e.g., due to metal artifacts causing incorrect marker detection, resulting in inconsistent coordinates between two follow-up visits). Conversely, if both changes are simultaneously large or simultaneously small, it may be a genuine physiological change or normal measurement fluctuation, and is not considered abnormal.

[0053] This invention automatically identifies anomalies in cone-beam CT data by comparing the changing trends of two independent data sources. This judgment method utilizes the physiological characteristic of slow width changes after arch expansion surgery (width changes are usually very small within adjacent follow-up intervals) and the high reliability of intraoral scan data to achieve online verification of cone-beam CT data quality. Without this judgment, when cone-beam CT data contains errors due to artifacts, the system will directly use the erroneous data for stability assessment, which may lead to false alarms (e.g., judging normal stability as recurrence, or judging recurrence as stability).

[0054] When the system determines that the registration or calibration of cone-beam computed tomography (CBCT) data is abnormal, remedial measures are required to ensure the effectiveness of subsequent stability assessments. The specific procedure is as follows: The second width distance calculated based on CBCT data is abandoned, and instead, the first width distance directly measured from intraoral scan data is used as the width measurement value at the current follow-up time for subsequent recurrence rate calculations and early warning judgments. Simultaneously, the system output clearly indicates that the data source for this width distance is the intraoral scan, so that doctors understand how the data was generated. If doctors have doubts about the results of the automatic rollback, they can manually review the CBCT images or perform a new intraoral scan. In the event of unreliable CBCT data, the system can automatically switch to a backup data source (intraoral scan), ensuring the continuity and reliability of stability assessment data. Without this rollback mechanism, when CBCT data is abnormal, the system will be unable to output valid assessment results or will output incorrect results. By rolling back to use intraoral scan data, the system can still perform stability assessments based on arch width. Furthermore, indicating the data source enhances the system's transparency, facilitating manual review by doctors when necessary. This design enables the method to provide high-precision coordinates of bony landmarks in the vast majority of cases (when cone-beam CT data is normal), and to provide usable dental arch width data in the rare abnormal cases, thus ensuring the robustness of the entire assessment and early warning system.

[0055] It should be noted that when the current step determines that the artifacts are severe and abandons fine-tuning, the second width distance is equal to the first width distance. At this time, the change in the first width is equal to the change in the second width. In this embodiment, the anomaly determination condition is not met, and the system uses the measurement value normally. The anomaly detection in this embodiment is mainly aimed at the situation where the registration or calibration process does not trigger the determination of severe artifacts, but the result is still unreliable.

[0056] like Figure 2 As shown, the present invention also provides a stability assessment and early warning system for maxillary expansion surgery, comprising: The data acquisition module is used to acquire maxillary cone-beam CT data and intraoral scan data at the same follow-up time. Anchor region extraction module is used to extract anti-artifact bone anchoring zone point cloud based on the maxillary cone-beam CT data; The spatial registration module is used to extract the palatal mucosa surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring area point cloud based on the intraoral scanning data, and to rigidly register the anti-artifact bone anchoring area point cloud with the palatal mucosa surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system. The marker detection module is used to obtain key markers for the efficacy of vasodilation based on the intraoral scanning data, and to obtain their coordinates in the intraoral scanning coordinate system. The coordinate projection module is used to project the coordinates onto the cone-beam CT coordinate system according to the cross-modal space mapping matrix, so as to obtain the cross-modal mapping coordinates of the key landmark points of the arch expansion treatment in the cone-beam CT coordinate system; The local calibration module is used to determine a local region in the cone-beam CT data based on the cross-modal mapping coordinates, and to perform position calibration on the cross-modal mapping coordinates based on the cone-beam CT data, and output the final marker coordinates; The evaluation and early warning module is used to output stability evaluation conclusions and early warning information based on the final marker point coordinates.

[0057] The local calibration module includes: The neighborhood delineation unit is used to select a cubic neighborhood of a predetermined size in the cone-beam CT volume data, centered on the cross-modal mapping coordinates, based on the cross-modal mapping coordinates. A variance calculation unit is used to calculate the gray-level variance within the neighborhood of the cube. The anomaly detection and coordinate determination unit is used to determine whether the grayscale variance is lower than a first preset variance threshold or higher than a second preset variance threshold. If the grayscale variance is lower than the first preset variance threshold or higher than the second preset variance threshold, it is determined that the area is severely damaged by metal artifacts, and the cross-modal mapping coordinates are directly used as the final marker point coordinates. If the gray-level variance is not lower than the first preset variance threshold or not higher than the second preset variance threshold, then the gray-level gradient modulus of each voxel in the neighborhood of the cube is calculated. Based on the gray-level gradient modulus, a quadratic surface is fitted in the neighborhood of the cube, and the local maximum point of the gray-level gradient modulus is searched. The coordinates of the local maximum point are used as the coordinates of the final marker point.

[0058] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for assessing and warning of stability after maxillary expansion surgery.

[0059] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for assessing and warning of stability after maxillary expansion surgery.

[0060] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0061] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for assessing and providing early warning of stability after maxillary expansion surgery, characterized in that, include: Acquire maxillary cone-beam computed tomography (CBCT) data and intraoral scan data at the same follow-up time. Anti-artifact bone anchoring zone point cloud was extracted based on the aforementioned maxillary cone-beam CT data; Based on the intraoral scanning data, the palatal mucosa surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring area point cloud is extracted, and the anti-artifact bone anchoring area point cloud and the palatal mucosa surface morphology point cloud are rigidly registered to obtain a cross-modal space mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system. Based on the intraoral scanning data, key landmarks for the efficacy of arch dilation are obtained, and their coordinates in the intraoral scanning coordinate system are obtained. The coordinates are projected onto the cone-beam CT coordinate system based on the cross-modal spatial mapping matrix to obtain the cross-modal mapping coordinates of the key landmark points of the arch expansion treatment in the cone-beam CT coordinate system; Based on the cross-modal mapping coordinates, a local region is determined in the cone-beam CT data, and the position of the cross-modal mapping coordinates is calibrated based on the cone-beam CT data to output the final marker coordinates; Based on the final marker coordinates, output stability assessment conclusions and early warning information.

2. The method for assessing and warning of stability after maxillary expansion surgery according to claim 1, characterized in that, The step of extracting anti-artifact bony anchoring zone point cloud based on the maxillary cone-beam CT data includes: Obtain the horizontal plate region of the palatine bone in the middle and posterior part of the hard palate and the bilateral nasal floor plate regions as target anchoring areas; Based on the target anchoring area, a seed point position is set in the cone-beam CT data. The seed point position satisfies the following constraints: in the horizontal direction, it is limited to the left and right predetermined widths of the mid-palatal suture; in the sagittal direction, it is limited to the predetermined proportion range after the full length of the mid-palatal suture; and in the vertical direction, it is limited to the preset bone tissue grayscale threshold range of the hard palate cortex. Based on the seed point location, a region growing algorithm is used to segment the anti-artifact bone anchoring area point cloud, starting from the seed point and using conditions of being greater than the preset bone tissue grayscale threshold and having a smooth gradient.

3. The method for assessing and warning of stability after maxillary expansion surgery according to claim 1, characterized in that, The step of rigidly registering the anti-artifact bone anchoring area point cloud with the palatal mucosa surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system includes: The iterative nearest point algorithm is used to iteratively solve for the rotation matrix and translation vector that minimizes the average Euclidean distance between the point cloud of the anti-artifact bone anchoring zone and the point cloud of the palatal mucosa surface morphology. Construct a cross-modal space mapping matrix from the oral scan coordinate system to the cone-beam CT coordinate system based on the rotation matrix and translation vector.

4. The method for assessing and warning of stability after maxillary expansion surgery according to claim 1, characterized in that, The step of obtaining key landmarks for arch vasodilation treatment based on the intraoral scan data and obtaining their coordinates in the intraoral scan coordinate system includes: A three-dimensional mesh model is constructed based on the intraoral scan data, and the maximum and minimum principal curvatures of each vertex on the mesh model are calculated. The maximum and minimum principal curvatures are obtained by calculating the eigenvalues ​​of the Hessian matrix after fitting a quadratic surface to the local neighborhood of the vertex of the mesh model. Calculate the shape index of each vertex based on the maximum and minimum principal curvatures; Based on the shape index, vertex regions with a shape index less than a first preset shape index threshold are marked as central fossa candidate regions, and vertex regions with a shape index greater than a second preset shape index threshold are marked as cusp candidate regions. Obtain the expected positional sequence information of teeth on the dental arch and the preset geometric template of key landmark points for arch expansion treatment; Based on the expected positional order information, the point closest to the candidate central fossa and the candidate cusp is selected as the key landmark point for arch expansion treatment in each candidate central fossa and the candidate cusp, and its coordinates in the intraoral scanning coordinate system are obtained.

5. The method for assessing and warning of stability after maxillary expansion surgery according to claim 1, characterized in that, The steps of determining a local region in the cone-beam CT data based on the cross-modal mapping coordinates, calibrating the position of the cross-modal mapping coordinates based on the cone-beam CT data, and outputting the final marker coordinates include: Based on the cross-modal mapping coordinates, a cubic neighborhood of a predetermined size is taken in the cone-beam CT volume data with the cross-modal mapping coordinates as the center; Calculate the gray-level variance within the neighborhood of the cube; Determine whether the grayscale variance is lower than a first preset variance threshold or higher than a second preset variance threshold: If the grayscale variance is lower than the first preset variance threshold or higher than the second preset variance threshold, it is determined that the area is severely damaged by metal artifacts, and the cross-modal mapping coordinates are directly used as the final marker point coordinates. If the gray-level variance is not lower than the first preset variance threshold and not higher than the second preset variance threshold, then the gray-level gradient magnitude of each voxel in the neighborhood of the cube is calculated. Based on the gray-level gradient magnitude, a quadratic surface is fitted in the neighborhood of the cube, and the local maximum point of the gray-level gradient magnitude is searched. The coordinates of the local maximum point are used as the coordinates of the final marker point.

6. The method for assessing and warning stability after maxillary expansion surgery according to claim 1, characterized in that, It also includes the following steps: Based on the final landmark coordinates, calculate the second width distance between the key landmarks of the left and right symmetrical arch vasodilation efficacy at the current follow-up time, and obtain the first width distance between the key landmarks of the left and right symmetrical arch vasodilation efficacy based on the intraoral scan data. Obtain the historical width distance corresponding to the previous follow-up time, and calculate the first width change based on the historical width distance and the second width distance; obtain the historical oral scan width distance corresponding to the previous follow-up time, and calculate the second width change based on the historical oral scan width distance and the first width distance; Compare the first width change with the first change threshold, and compare the second width change with the second change threshold: If the change in the first width is greater than the first change threshold and the change in the second width is less than the second change threshold, it is determined to be abnormal. The first width distance is used to replace the second width distance, and the data source is marked as intraoral scanning in the output result.

7. A stability assessment and early warning system for maxillary expansion surgery, characterized in that, include: The data acquisition module is used to acquire maxillary cone-beam CT data and intraoral scan data at the same follow-up time. Anchor region extraction module is used to extract anti-artifact bone anchoring zone point cloud based on the maxillary cone-beam CT data; The spatial registration module is used to extract the palatal mucosa surface morphology point cloud corresponding to the anatomical position of the anti-artifact bone anchoring area point cloud based on the intraoral scanning data, and to rigidly register the anti-artifact bone anchoring area point cloud with the palatal mucosa surface morphology point cloud to obtain a cross-modal spatial mapping matrix from the intraoral scanning coordinate system to the cone-beam CT coordinate system. The marker detection module is used to obtain key markers for the efficacy of vasodilation based on the intraoral scanning data, and to obtain their coordinates in the intraoral scanning coordinate system. The coordinate projection module is used to project the coordinates onto the cone-beam CT coordinate system according to the cross-modal space mapping matrix, so as to obtain the cross-modal mapping coordinates of the key landmark points of the arch expansion treatment in the cone-beam CT coordinate system; The local calibration module is used to determine a local region in the cone-beam CT data based on the cross-modal mapping coordinates, and to perform position calibration on the cross-modal mapping coordinates based on the cone-beam CT data, and output the final marker coordinates; The evaluation and early warning module is used to output stability evaluation conclusions and early warning information based on the final marker point coordinates.

8. The stability assessment and early warning system after maxillary expansion surgery according to claim 7, characterized in that, The local calibration module includes: The neighborhood delineation unit is used to select a cubic neighborhood of a predetermined size in the cone-beam CT volume data, centered on the cross-modal mapping coordinates, based on the cross-modal mapping coordinates. A variance calculation unit is used to calculate the gray-level variance within the neighborhood of the cube. The anomaly detection and coordinate determination unit is used to determine whether the grayscale variance is lower than a first preset variance threshold or higher than a second preset variance threshold. If the grayscale variance is lower than the first preset variance threshold or higher than the second preset variance threshold, it is determined that the area is severely damaged by metal artifacts, and the cross-modal mapping coordinates are directly used as the final marker point coordinates. If the gray-level variance is not lower than the first preset variance threshold or not higher than the second preset variance threshold, then the gray-level gradient modulus of each voxel in the neighborhood of the cube is calculated. Based on the gray-level gradient modulus, a quadratic surface is fitted in the neighborhood of the cube, and the local maximum point of the gray-level gradient modulus is searched. The coordinates of the local maximum point are used as the coordinates of the final marker point.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images