System and method for generating full-tooth geometric grid

Abstract

Systems (100) and methods (200) for generating a full tooth model geometry mesh (130) from surface data (114) and volumetric image data (116) are disclosed. A full-tooth geometric grid (130) is generated in a computationally efficient manner. A method (200) of segmenting a full tooth model from volumetric image data (116) and intraoral surface data (114) includes providing volumetric image data (116) of a dental anatomy, providing intraoral surface data (114) of the dental anatomy, and generating enhanced image data including the volumetric image and the intraoral surface data by aligning the surface (114) and the volumetric image data (116). The dental crown (320) in the oral inner surface data (114) is segmented. The method (200) includes selecting a template tooth (322) from a library. The anatomical identifier corresponds to an image data tooth identifier. A template tooth (322) is fitted to the teeth of the enhanced image data. Segmenting teeth in the volumetric image data based on fitting template teeth (610) provides a full tooth model geometry mesh (130).

Description

Background Art

[0001] When planning dental restorations, endodontics, orthodontics, or maxillofacial surgery, having a virtual representation of a patient's actual teeth is helpful. In recent years, computer-based systems and methods that use three-dimensional data (3D data) and 3D models to generate such virtual representations have been developed and implemented. This development is driven by increasingly sophisticated imaging technologies, such as computed tomography (CT), particularly cone-beam computed tomography (CBCT), which also allow for the assessment of non-visible areas of a patient's dentition. While these imaging technologies provide dental professionals with images for easily visually assessing the condition of teeth, 3D data has the potential to go beyond this visual assessment. To this end, the data behind the image can be analyzed, supplementing the image with relevant information about the tooth's condition, which aids dental professionals in planning treatment. In the case of CT or CBCT data, the data can be segmented. However, segmenting one or more teeth from CT or CBCT images into separate entities is challenging due to the similar densities of tooth roots and surrounding bone, as well as contact areas between adjacent teeth. Furthermore, high-intensity streaking artifacts are often present, caused, for example, by dense dental fillings or gaps between adjacent crowns, making it difficult to clearly delineate the crown region of the tooth. Furthermore, the density of the jawbone and teeth is uneven. Both possess different types beneath their respective surfaces. On a macroscopic level, the jawbone is composed of cortical and cancellous bone tissue, while teeth are composed of enamel and dentin. These substances have overlapping densities, making it difficult to clearly distinguish them. Furthermore, on a microscopic level, these tissues have varying degrees of porosity, resulting in weak and unstable margins when visualizing their structures.

[0002] A virtual geometric mesh of the actual tooth can be generated using scan data from intraoral (IO) surface data (such as data obtained by scanning a tooth impression or its model, or by direct intraoral scanning and volumetric image data (such as obtained by CT or CBCT). IO surface data provides high resolution data, but is only applicable to the crown portion of the tooth. Volumetric image data provides full tooth data, but the resolution of the root area may be noisy. There is a challenge in finding a computationally efficient method to generate an accurate full tooth geometric mesh of a patient's teeth based on the scan data. The challenge is to find a computationally efficient method to generate an accurate full tooth geometric mesh of a patient's teeth based on the scan data. Summary of the Invention

[0003] The present disclosure relates to methods and systems for generating a geometric mesh of a full tooth model based on surface data and volumetric image data, the geometric mesh comprising root and crown meshes, each representing a corresponding portion of a patient's dental arch. The methods and systems disclosed herein enable the generation of an accurate full tooth geometric mesh in a computationally efficient manner. In one arrangement, a computer-implemented method for segmenting a full tooth model based on volumetric image data and intraoral surface data of a subject's dental anatomy comprises the following initial steps: (i) providing volumetric image data of the dental anatomy, (ii) providing intraoral surface data of the dental anatomy, and (iii) generating enhanced image data comprising the volumetric image and intraoral surface data by aligning the surface and volumetric image data. Preferably, (iv) the crown of the tooth in the intraoral surface data is segmented. The method further comprises (v) selecting a template tooth from a template tooth library (also referred to as a "root library"). Preferably, the anatomical designation, indicated by, for example, the tooth number of the template tooth (referred to herein as a "tooth identifier" or "identification"), corresponds to the identification of a given tooth in the enhanced image data. Then, (vi) the corresponding template teeth are fitted to the teeth of the enhanced image data, wherein a full tooth model geometric mesh is provided by (vii) segmenting the teeth in the volumetric image data based on the fitted template teeth.

[0004] Typically, fitting the template tooth to the corresponding tooth in the enhanced image data comprises aligning the template tooth with the segmented crown portion of the tooth. Preferably, the fitting further comprises a scaled rigid transformation of the corresponding template tooth relative to the segmented crown portion of the tooth in the enhanced image data.

[0005] In an embodiment, alignment and / or scaled rigid registration of template teeth with segmented crowns in the enhanced image data is performed automatically by a computer, without requiring user input. This embodiment generally includes selecting two or more template teeth from a library of template teeth, wherein the tooth identifier of each selected template tooth corresponds to the identifier of a tooth in the enhanced image data. The selected template teeth are then virtually positioned in the dental arch, for example, by virtually placing the selected template teeth at corresponding locations of teeth with the same tooth identifier in a reference dental arch. Subsequently, the template teeth can be aligned with the corresponding teeth in the enhanced image data by performing a scaled rigid transformation of the virtual dental arch, wherein the crown center positions of the template teeth in the virtual dental arch are mapped to the crown center positions detected for the corresponding teeth in the enhanced image data. In this embodiment, the tooth center positions of the template teeth can be predetermined and stored as part of the template data. Optionally, this alignment is followed by rigid scaling of the template teeth based on the distance between the crown center positions of corresponding adjacent teeth in the enhanced image data. The aligned template teeth can then be further scaled rigidly transformed relative to the segmented crowns of the corresponding teeth in the enhanced image data. The step of aligning and then rigidly scaling the template teeth relative to the segmented crown parts is referred to herein as a rigid initialization step.

[0006] A method for segmenting a full tooth model from volumetric image data and intraoral surface data of a subject's dental anatomy according to the present invention may include determining the outline of a tooth root portion in the enhanced image data. The tooth root portion is represented in the volumetric image data of the enhanced image data. Determining the outline of the tooth root portion includes determining edge points of the tooth in the volumetric image data, starting with edge points that coincide with the apical boundary of the segmented tooth crown portion from the aligned surface data, and propagating these edge points in the apical direction along the root edge of the tooth in the volumetric image data. Propagating edge points in the apical direction along the root edge of the tooth in the volumetric image data is preferably performed in an iterative stepwise process. Typically, the process begins by identifying edge points in the volumetric image data that coincide with the apical boundary of the segmented tooth crown portion, and then, in each step, a new edge point is found in the apical direction starting from the corresponding edge point identified in the previous step. In one arrangement, a step size in the apical direction is individually determined for each such current edge point in any given iteration of the process. The step size can vary between 0 and 5 voxels, for example, between 0.5 and 3 voxels or between 0.5 and 2 voxels. Preferably, a higher step size is used for edge points further away from the presumed root apex than for edge points closer to the presumed root apex. In this way, the edge points are forced to converge to the same plane. Once the edge points have substantially converged to the same plane, the step size is preferably similar for all edge points within that plane.

[0007] In an iteration, a step-by-step process for outlining the root portion of a tooth is guided by a root direction and an inward direction at a current edge point. The inward direction points from the current edge point to the interior of the root. Typically, the inward direction is provided by a gradient direction at a given edge position of the tooth. The root direction provides an indication of the direction towards the root apex position. The root direction of a tooth of the enhanced image data can be obtained from an alignment of a template tooth with a segmented crown portion of the tooth, wherein the tooth identification of the template tooth corresponds to the tooth identification of the tooth of the enhanced image data. In another arrangement, the tooth direction is provided by performing a scaled rigid transformation of the corresponding aligned template tooth relative to the segmented crown surface of the tooth of the enhanced image data. Alternatively, the root direction is an adaptive direction provided by the direction between the crown center of the tooth of the enhanced image data and the center of the current set of edge points in a given iteration.

[0008] Determining the outline of the root portion of the tooth in the enhanced image data may further include testing the convergence of a set of edge points to a common location in the volumetric image data. This convergence may indicate that the set of edge points is close to the root apex location. Therefore, in one embodiment of the method, the apex marker location is derived from the set of edge points that have passed the convergence test.

[0009] After identifying the apex landmark, the outline of the tooth root in the enhanced image data can be determined using the voxel positions of the edge points detected in the iterative step, which resulted in a convergent set of edge points used to identify the apex landmark. The outline of the tooth root in the enhanced image data is also referred to herein as a "tooth mask." According to the method of the present invention, the number of apexes identified for the tooth in the enhanced image data, as described above, can be used to select a template tooth with the same number of apexes, so that the template tooth can be fitted to the tooth.

[0010] A method for segmenting a full tooth model from volumetric image data and intraoral surface data of a subject's dental anatomy according to the method of the present invention comprises fitting a template tooth to a tooth of the enhanced image data. As previously noted, such fitting may comprise aligning the segmented crown portion of the tooth in the enhanced image data on the corresponding template tooth and subsequently applying a scaled rigid transformation. Fitting the corresponding template tooth to the tooth of the enhanced image data may also comprise non-rigidly deforming the template tooth into a deformed template tooth mesh that matches the segmented crown portion and preferably matches the position of the root apex marker of the tooth of the enhanced image data. In one arrangement, the root apex marker may be automatically identified as described above. Alternatively, the apex marker is indicated by the user in a user interface. Such a deformed template tooth mesh provides a tooth model whose crown portion and overall dimensions closely match the crown portion and overall dimensions of the corresponding tooth in the enhanced image data.

[0011] In a further step, the root portion of the deformed template tooth is aligned to the root portion of the tooth in the enhanced image data, more specifically the root portion of the tooth in the volumetric image data. Preferably, before fitting the root portion of the deformed template tooth, its vertices are marked as crown vertices or root vertices based on the proximity between the deformed template tooth mesh and the segmented crown portion of the matching tooth in the enhanced image data. In one embodiment, marking the vertices of the deformed template tooth mesh includes determining, for each vertex on the deformed template tooth mesh, a distance to the nearest point on the segmented crown surface mesh, preferably, such nearest point is not limited to a vertex of the segmented crown portion, but may also be located on any face or edge of said surface. If, for a vertex of the deformed template tooth, the distance is below a given threshold, for example below 0.4, 0.3, 0.2 or 0.1 mm, the vertex is marked as part of the crown portion, otherwise the vertex is marked as part of the root portion. After marking the crown and root vertices of the deformed template tooth, the root portion of the deformed template tooth mesh can be further non-rigidly deformed to align the root portion vertices with the edges of the selected tooth in the volumetric image data. In one arrangement, deforming the root portion of the deformed template tooth mesh to align the root portion vertices with the edges of the selected tooth in the volumetric image data includes a coarse registration step, wherein the root portion vertices of the deformed template tooth mesh are aligned with the outline of the tooth root in the enhanced image data as described above. Preferably, the coarse registration step is followed by a fine registration step, wherein the root portion of the deformed template tooth mesh is further registered to the root edge of the corresponding tooth in the volumetric image data.

[0012] Another object of the present invention is to provide a computer system comprising: a memory and a processor in communication with the memory, the processor being configured with processor-executable instructions to perform operations comprising: (i) receiving volumetric image data of a subject's dental anatomical structure; (ii) receiving intraoral surface data of the dental anatomical structure; (iii) generating enhanced image data by aligning the surface data and the volumetric image data; (iv) segmenting the crown portion of the tooth in the intraoral surface data; (v) selecting a template tooth from a template tooth library, wherein the tooth identification of the template tooth corresponds to the tooth identification of the enhanced image data; (vi) fitting the corresponding template tooth to the tooth in the enhanced image data; and (vii) segmenting the tooth in the volumetric image data based on the fitted template tooth.

[0013] In another object, the present invention provides a non-transitory computer-readable medium storing computer-executable instructions, which, when executed by one or more computer systems, configure the one or more computer systems to perform operations, the operations comprising: (i) receiving volumetric image data of a subject's dental anatomical structure; (ii) receiving intraoral surface data of a dental anatomical structure; (iii) generating enhanced image data by aligning the surface data and the volumetric image data; (iv) segmenting the crown portion of the tooth in the intraoral surface data; (v) selecting a template tooth from a template tooth library, wherein the tooth identification of the template tooth corresponds to the tooth identification of the enhanced image data; (vi) fitting the corresponding template tooth to the tooth in the enhanced image data; and (vii) segmenting the tooth in the volumetric image data based on the fitted template tooth.

[0014] Preferably, the template tooth used in the method and / or system of the present invention includes a crown mesh and a root mesh, wherein the mesh density of the crown mesh is higher than the mesh density of the root mesh. Typically, the mesh density of the crown should correspond to the mesh density of a tooth surface scan typically obtained using an intraoral scan or by scanning a tooth impression or plaster model. The mesh density of the crown mesh portion of the template tooth surface determined based on the average edge length can be between 0.06 and 0.15 mm, for example, between 0.08 and 0.15 mm, for example, between 0.09 and 0.13. On the other hand, the root mesh portion of the template tooth surface should generally reflect the lower resolution of the contour obtained from the CT or CBCT scan data (compared to the tooth surface scan). The mesh density of the root mesh portion of the template tooth surface can be between 0.15 and 0.30, for example, between 0.15 and 0.25, for example, between 0.15 and 0.22. In an embodiment, the average edge length of the root mesh portion is at least 1.1 times, such as at least 1.3 times, such as at least 1.5 times, higher than the mesh density of the root mesh portion. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Embodiments of the present disclosure will be described below, by way of example only, with reference to the accompanying drawings, in which:

[0016] Figure 1 Depicted is a schematic diagram of a mesh generation system according to certain aspects of the present disclosure.

[0017] Figure 2 A mesh generation method according to certain aspects of the present disclosure is shown.

[0018] Figure 3 An IO scan aligned with a CBCT scan is shown.

[0019] Figure 4A Shown is the segmented mesh from the IO scan before separation of the individual crown meshes.

[0020] Figure 4B Shows the mesh after separating the individual crowns Figure 4A grid.

[0021] Figure 5A Shown is a tooth crown mesh from an IO scan.

[0022] Figure 5B Shown Figure 5A Crown mesh superimposed on the template mesh.

[0023] Figure 6 Shown are multiple full tooth geometry meshes after a rigid registration process according to certain aspects of the present disclosure.

[0024] Figure 7 The boundary edge of the 10 scan crown mesh is shown.

[0025] Figure 8A Methods of determining a next position of an edge point on a mesh mask according to certain aspects of the present disclosure are described.

[0026] Figure 8B An axial plane view of a set of edge points according to certain aspects of the present disclosure is shown.

[0027] Figure 9A An early progression of edge points is shown, advancing from the apex away from the crown mesh to define a root cap, wherein the template mesh is used for illustration purposes only, in accordance with certain aspects of the present disclosure.

[0028] Figure 9B shows the further progress of the edge points Figure 9A edge points.

[0029] Figure 9C shows the further progress of the edge points Figure 9B edge points.

[0030] Figure 9D shows the further progress of the edge points Figure 9C edge points.

[0031] Figure 10 A schematic diagram illustrating a step-by-step correction method applicable to edge points that have converged to a plane orthogonal to a tooth root direction, according to certain aspects of the present disclosure.

[0032] Figure 11A A method for evaluating apical convergence of a set of edge points according to certain aspects of the present disclosure is shown.

[0033] Figure 11B Shown Figure 11A Methods applied to different sets of edge points.

[0034] Figure 11C Shown Figure 11A Methods applied to different sets of edge points.

[0035] Figure 12A A method for evaluating apical convergence of a set of edge points according to certain aspects of the present disclosure is shown.

[0036] Figure 12B Shown Figure 12A Methods applied to different sets of edge points.

[0037] Figure 12C Shown Figure 12A Methods applied to different sets of edge points.

[0038] Figure 13A A method for evaluating apical convergence of a set of edge points according to certain aspects of the present disclosure is shown.

[0039] Figure 13B Shown Figure 13A Methods applied to different sets of edge points.

[0040] Figure 14 A filtering method for determining whether an edge point has converged to the root apex of an adjacent tooth is shown.

[0041] Figure 15A The position history of a set of edge points according to certain aspects of the present disclosure is shown.

[0042] Figure 15B The position history of a set of edge points according to certain aspects of the present disclosure is shown.

[0043] Figure 15C The position history of a set of edge points according to certain aspects of the present disclosure is shown.

[0044] Figure 16A Shown are edge point strips after a refined registration process according to certain aspects of the present disclosure.

[0045] Figure 16B Shown are edge point strips after a refined registration process according to certain aspects of the present disclosure.

[0046] Figure 16C Shown are edge point strips after a refined registration process according to certain aspects of the present disclosure.

[0047] Figure 16D Shown are edge point strips after a refined registration process according to certain aspects of the present disclosure.

[0048] Figure 17AA coronal plane view of edge points after a refined registration process according to certain aspects of the present disclosure is shown.

[0049] Figure 17B A sagittal plane view of edge points after a refined registration process according to certain aspects of the present disclosure is shown.

[0050] Figure 17C An axial plane view of edge points after a fine registration process according to certain aspects of the present disclosure is shown.

[0051] Figure 18A A coronal plane view of edge points after a refined registration process according to certain aspects of the present disclosure is shown.

[0052] Figure 18B A sagittal plane view of edge points after a refined registration process according to certain aspects of the present disclosure is shown.

[0053] Figure 18C An axial plane view of edge points after a fine registration process according to certain aspects of the present disclosure is shown.

[0054] Figure 19A A coronal plane view of edge points after a refined registration process according to certain aspects of the present disclosure is shown.

[0055] Figure 19B A sagittal plane view of edge points after a refined registration process according to certain aspects of the present disclosure is shown.

[0056] Figure 19C An axial plane view of edge points after a fine registration process according to certain aspects of the present disclosure is shown.

[0057] Figure 20A A right side view of the patient's jaw is shown.

[0058] Figure 20B According to certain aspects of the present disclosure Figure 20A View of the patient's jaw with superimposition of the full tooth fitting template.

[0059] Figure 21A A left side view of the patient's jaw is shown.

[0060] Figure 21B According to certain aspects of the present disclosure Figure 21A View of the patient's jaw with superimposition of the full tooth fitting template.

[0061] Figure 22 Validation data for two datasets according to certain aspects of the present disclosure are shown. DETAILED DESCRIPTION

[0062] The present disclosure generally relates to methods and systems for generating full tooth geometric meshes. In some arrangements, the full tooth mesh generated by the methods, systems, algorithms or processes of the present disclosure can be used to plan or evaluate dental or maxillofacial surgical treatment plans. In some configurations, the methods or systems disclosed herein may include a fully automated software process that can use volumetric image data, such as obtained by computed tomography (CT) or cone beam computed tomography (CBCT), intraoral (IO) surface data, and a predefined three-dimensional (3D) tooth model including crowns and roots (the predefined 3D tooth model is also referred to herein as a "template tooth," and the library including a set of template teeth is also referred to herein as a "root library") to reconstruct a patient's full tooth mesh (including crown and root surfaces).

[0063] In some arrangements, the method or system may include an algorithm that starts with volumetric image data of a patient's dental arch, or a portion thereof, aligned with IO surface data. The method or system may include one or more of the following steps. The step may use a segmented crown surface from the IO surface data (also referred to herein as a "segmented IO crown" or "segmented crown portion") to calculate the orientation, size, and rotation in which a root should be placed, and use this information to align and preferably scale a template tooth from a root library to one or more segmented IO crowns, preferably to each segmented IO crown. The step may start with the segmented crown boundary edge of the IO surface data and begin searching for the root in the volumetric image data of one or more teeth until a set of possible root apices is found. In some arrangements, the result may be a set of points covering the root and the root apices of one or more teeth in the volumetric data (also referred to herein as a "tooth mask"). Using the points of the segmented crown and mask, a non-rigid crown and root registration step may be applied to the one or more aligned template teeth. A band deformation step may be performed from the gingival margin to the root apices and may be used to fine-tune the shape of the template teeth. In some configurations, the gingival margin can be defined using a looped attire (LAT) generated from a crown margin point, as described herein.

[0064] Figure 1A schematic diagram of a non-limiting illustrative example of a full tooth mesh generation system 100 according to certain aspects of the present disclosure is depicted. In some arrangements, the system 100 may include an intraoral (IO) scanner 110 and a CT scanner or cone beam computed tomography (CBCT) scanner 112. Alternatively or additionally, the system may include a desktop 3D scanner for scanning an impression of the patient's IO anatomy or a plaster model of such an impression. The desktop 3D or IO scanner 110 may be used to obtain surface data of the patient's IO anatomy, while the CT or CBCT scanner 112 may be used to obtain volumetric image data of a maxillofacial region including a portion of a jaw and one or more teeth. Figure 1 As schematically shown, a patient's tooth 10 may span a gum line 11. A crown portion 12 of the tooth 10 may be exposed above the gum line 11. A root portion 14 of the tooth 10 may extend below the gum line 11, toward the patient's jaw. The IO scanner 110 may obtain a high-resolution spatial scan of the crown portion 12 of the tooth 10, but may not provide spatial information about the root portion 14. The CBCT scanner 112 may obtain spatial information about both the crown portion 12 and the root portion 14. However, the volumetric image data 116 of the root portion 14 tends to be noisy compared to the IO surface data 114. As described herein, the system 100 may receive the IO surface data 114 and the corresponding volumetric image data 112 of the tooth 10 and return an accurate full-tooth geometric mesh 130 of the tooth 10. In some embodiments, the system 100 may be fully automatic and may not require human intervention to generate the full-tooth mesh 130 based on the IO surface data 114 and the volumetric image data 116.

[0065] Continue to refer Figure 1 , the system 100 can include a computer 120 having a display 122 and a processing unit 124. The processing unit 124 can be programmed or otherwise configured to perform the computational and analytical steps or methods described herein. The computer 120 can be configured to receive IO surface data 114 from a desktop 3D or IO scanner 110 and volumetric image data 116 from a CT or CBCT scanner 112. The processing unit 124 can be configured to perform the computational and analytical methods described herein to generate a full tooth geometry mesh 130 based on the IO surface data 114 and the volumetric image data 116.

[0066] Figure 2 A schematic overview of a full tooth mesh generation method 200 is depicted. In some embodiments, the full tooth mesh generation system 100 ( Figure 1) may be configured to perform a full tooth mesh generation method 200. The full tooth mesh generation method 200 may receive IO surface data and volume image data of a patient's teeth as input. The method 200 may include an alignment step 210, wherein the IO surface data and the volume image data are aligned with each other. The IO surface data aligned with the volume image data may be referred to as enhanced image data. The method 200 may include a rigid initialization step 220, wherein one or more template tooth meshes are aligned with one or more of the IO surface data 114, preferably with all crown portions. In some arrangements, the rigid initialization step 220 may include detecting the approximate position, size, and orientation of one or more tooth roots by performing said alignment of one or more template teeth of the root library with the crown portions of the IO surface data in the enhanced image data. The method 200 may include a root apex detection step 230, wherein a set of edge points is tested for convergence as the edge points advance along the root portion of the volume image data. In some configurations, the apex detection step 230 may include performing an iterative search for the root apex by segmenting the root in the volumetric image data from voxels that coincide with the aligned crown mesh boundary edge until one or more root apexes of the root are found. The method 200 may include a coarse registration step 240, wherein the point history of convergent edge points is used to form a mask to which the template tooth mesh can be deformed. In some arrangements, the coarse registration step 240 may include performing a non-rigid deformation of the crown and root of the aligned template tooth on the segmented IO crown and the set of points (mask) found in the apex detection step 230, respectively. The method may include a fine registration step 250, wherein a fine registration of the root apex of the aligned and deformed template tooth with the volumetric image data is performed. In some configurations, the fine registration step 250 may include performing a progressive or zonal optimization of the root surface starting from the crown boundary edge to improve the registration accuracy of the root portion of the template tooth. After the template teeth are registered with the IO surface data and volume image data of the given tooth, a fully segmented tooth model can be obtained based on the registered template teeth.

[0067] Figure 3A non-limiting exemplary embodiment of aligning volumetric image data obtained from a CBCT scan 302 with IO surface data obtained from an IO scan 304 is shown. In some arrangements, the CBCT scan 302 and the IO scan 304 are aligned with each other by aligning the CBCT crown portion 306 of the CBCT scan 302 with the IO crown portion 308 of the IO scan 304, or vice versa. In some arrangements, the method and system may include an algorithm whose input data is enhanced image data, which includes volumetric image data (e.g., CBCT scan 302 of the same patient) aligned with IO surface data (e.g., IO scan 304). The algorithm may search for and segment tooth roots within the CBCT scan 302 by extracting a set of points in 3D space for each tooth root, including the location of the root apex. Thereafter, a template tooth from a root library is registered based on the segmented crowns and the CBCT root points identified during segmentation.

[0068] Figure 4A and Figure 4B Trimming features that may be included in the rigid initialization step 220 are shown. Figure 4A A segmented IO surface mesh 300 obtained from an IO scan 304 is shown. The depicted segmented IO surface mesh 300 shows a plurality of tooth crowns 312 and a portion of adjacent tissue 314. In some variations, the segmented IO surface mesh 300 can include information regarding a tooth identification (ID, e.g., tooth number) for each tooth. In some arrangements, the curvature of the crown surface of the segmented IO crown mesh 300 can be analyzed to define a trimmed crown mesh surface that excludes interproximal regions 316. Figure 4B The segmented IO crown mesh 300 is shown after the curvature of the crown surface has been analyzed and the interproximal region 316 has been removed. In some arrangements, portions of the IO surface data representing soft tissue are identified as separate tissue and can be identified or removed. In some arrangements, the rigid initialization step 220 can include selecting a template tooth from a root library based on the tooth ID of the segmented IO crown mesh. The mesh of the selected template tooth can then be aligned with the segmented IO crown mesh.

[0069] Preferably, the registered template teeth have the same tooth ID as the segmented IO crown mesh to which they are aligned. The tooth type of each tooth in the enhanced image that needs to be segmented can be input by the user or determined automatically. PCT application WO2023194500 discloses a method for automatically identifying the crown center position and tooth ID (tooth number) of each tooth represented in volumetric image data. The automatically detected tooth number can then be used to select a suitable template tooth for aligning the template tooth with the segmented IO crown portion 320 in the enhanced data. Preferably, the template teeth are selected from a root library that includes one or more template teeth for each anatomical tooth type. In addition, preferably, each of the template teeth includes a crown mesh connected to the root mesh, wherein the density of the crown mesh is higher than the mesh density of the root mesh. Typically, the mesh density of the crown should correspond to the mesh density of a dental surface scan typically obtained using an intraoral scan or by scanning a dental impression or plaster model. The mesh density of the crown mesh portion of the template tooth surface, determined by the average edge length, may be between 0.06 and 0.15 mm, for example, between 0.08 and 0.15 mm, for example, between 0.09 and 0.13. On the other hand, the root mesh portion of the template tooth surface should generally reflect the lower resolution of the contours obtained from the CT or CBCT scan data (compared to the dental surface scan). The mesh density of the root mesh portion of the template tooth surface may be between 0.15 and 0.30, for example, between 0.15 and 0.25, for example, between 0.15 and 0.22. In an embodiment, the average edge length of the root mesh portion is at least 1.1 times, for example, at least 1.3 times, for example, at least 1.5 times, higher than the mesh density of the root mesh portion.

[0070] Figure 5A and Figure 5B Alignment features are shown that may be included in the rigid initialization step 220. For each vertex of the segmented crown mesh 320, a corresponding point may be found on the template tooth mesh 322. These correspondences may be used for iterative alignment and scaling using, for example, an iterative closest point (ICP) algorithm. Figure 5A Corresponding points from the crown mesh 320 are shown. Figure 5B A tooth template grid 322 is shown aligned with a crown grid 320 .

[0071] Figure 6 A non-limiting exemplary rigid scaled transformed mesh 400 is shown that may be obtained by performing the rigid initialization step 220. The rigid initialization step 220 may include an algorithm that performs a rigid scaled transform to determine an initial position of the template tooth mesh 322 relative to the segmented crown mesh 320. In some variations, the rigid scaled transform may provide an initial root direction at which to begin a search algorithm to detect the root apex in the apex detection step 230.

[0072] In one arrangement, the alignment and / or scaled rigid registration of the template teeth with the IO crowns segmented in the enhanced image data is performed automatically by a computer without user input. In an embodiment, the template tooth model is automatically registered to the teeth in the enhanced image data using the crown center positions determined for the teeth in the enhanced image data and the corresponding template teeth, respectively. In this embodiment, the tooth center positions of the template teeth can be predetermined and stored as part of the template data. WO2023194500 discloses a method for automatically identifying the crown center positions and tooth numbers of teeth represented in volumetric image data of a patient's maxillofacial region. Alternatively, the center position of the IO segmented crown can be determined by averaging the positions of all its vertices. Based on the tooth number detected or determined for the tooth in the volumetric data of the enhanced image data, the corresponding template tooth is selected and positioned in a virtual dental arch, such an average dental arch being obtained by averaging the dental arches and the corresponding tooth positions determined for a plurality of individual dental arches. For example, if the tooth numbers of all teeth in the upper jaw are present in the volume data, a set of model teeth is selected that includes template teeth for each upper tooth, wherein these template teeth are located in the dental arch corresponding to the dental arch of the virtual upper jaw. Subsequently, a scaled rigid transformation is applied to the virtual upper dental arch to map the initial crown center positions of the template teeth to the crown center positions detected for the corresponding teeth in the enhanced image data. Thereafter, the dimensions of each template tooth can be updated based on the distance between the adjacent tooth center positions. Preferably, this scaled rigid transformation is performed by rigid grid-based registration of each template tooth with the segmented 10 crown portion 320 of the corresponding tooth in the enhanced image data. This additional grid-based registration step can provide a more accurate orientation of the corresponding template tooth. Typically, this further grid-based registration includes applying an iterative closest point (ICP) algorithm, wherein the closest point correspondences between vertices of the segmented crown surface and the vertices of the respective corresponding template teeth are determined. Preferably, the closest point correspondences are determined for all vertices in the segmented crown surface and the vertices of the respective corresponding template teeth, and then a point-to-point cost function is used to establish an optimal rigid transformation between the correspondences. For example, a Tukey-based cost function is used, where the Tukey distance parameter is set to 1 mm.

[0073] The combination of the scaled rigid initialization of the template tooth model in the rigid initialization step 220 and the aligned segmented crown mesh 320 can provide a wealth of information about the crown surface and orientation of each tooth in the enhanced image data. Since the tooth roots are not present in the 10 surface data, information about the root level of each tooth is extracted from the volumetric image data. To address the situation where the volumetric image data may have lower resolution and be noisier than the 10 surface data, the system 100 or method 200 may include an algorithm that begins by determining a rough outline of the tooth root (referred to herein as a "tooth mask") while attempting to detect one or more root apexes of the tooth. The location of these detected root apexes will give a rough idea of the root's orientation and length, and may also provide an indication of the number of root apexes (i.e., the number of root segments) the tooth has. Correctly detecting the number of root apexes will allow the selection of the most appropriate template tooth in the subsequent registration step, i.e., a template tooth with the same tooth number and number of root apexes as the tooth selected in the image data. The rough outline of the tooth root in the volumetric image data, or tooth mask, will be used to further initialize the root shape of the aligned template tooth.

[0074] refer to Figure 7 The overall strategy in the apex detection step 230 is to start from the edge points 330 of the (accurately registered) crown mesh 320 and gradually propagate these points along the root edge towards the root apex in the volume image data. When approaching the root apex, the edge points 330 should be appropriately clustered to check their convergence. Once all edge points 330 have converged or the maximum number of steps has been reached, it can be checked which converged clusters can be marked as reliable. Based on these reliable converged clusters, clustered marked positions can be constructed, and the positions of the voxels that are accessed in intermediate steps to obtain such clusters can constitute the tooth cover of the tooth root.

[0075] Figure 7 Non-limiting exemplary starting conditions for propagating edge points 330 in the apex detection step 230 are shown. Apex detection can be initiated by the Cartesian coordinates of edge points 330 (crown boundary vertices) of the segmented 10 crown mesh 320 in the enhanced image data. All voxels that the edge points 330 fall into can then be retrieved and filtered to avoid duplicate voxels (e.g., multiple edge points 330 may be mapped to the same voxel).

[0076] The movement of each edge point 330 in the apex detection step 230 can be iteratively converged using the root direction, inward direction, and step size described herein to find one or more possible root apexes. The movement of the edge point 330 can be divided into two main stages. The first stage can be designed to move the edge point 330 sufficiently below the crown of the tooth, avoiding any possible bright spots that may be caused by fillings or metal braces on the top of the crown. In some arrangements, an initial rigid scaling transform estimate obtained from a rigid registration of the crown mesh can be used for the root direction. In some arrangements, a center direction can be used for the inward direction to limit the chance of the edge point deviating from adjacent teeth or nearby structures. In some variations, an adaptive step size can be used, where a larger step size value can be scaled by the voxel dimension and considered for the lowest point, while a smaller step size is applied to the highest point, forcing the lower points to converge to the same plane as the higher points.

[0077] The second stage of movement of the edge point 330 can be configured to handle a variety of root shapes, as well as roots that are split into multiple segments, requiring greater adaptability. An adaptive root direction can be used to capture the actual root direction at each step. In some configurations, the adaptive root direction can be defined by the center of the edge point 330 and the center determined for the segmented IO crown mesh 320 in the enhanced image data. The center position of the segmented IO crown mesh can be determined by averaging the positions of all vertices of the segmented IO crown mesh. For the inward direction, the gradient direction of the volumetric image data at the location of the edge point 330 can be used, providing more local information and may be more suitable for handling multiple root segments. Similarly, an adaptive step size can be used to maintain planar alignment of the edge point 330. In some variations, the same approach can be used, but the step size value can be weighted using the inward direction.

[0078] Figure 8A and Figure 8B A non-limiting exemplary process for determining the movement of edge points 330 is shown. In order to move the edge points 330 towards the apex of the root, an estimated root direction 332 can be used for each edge point 330. The initial root direction can be defined by the initial rigid registration, however, when the edge points 330 are coplanar, an adaptive root direction can be used. The adaptive root direction can be defined as the direction of the detection center of the crown mesh segmented from the enhanced image data towards the average value of all edge points 330 (edge point center 331). To further guide the apical search from the crown edge to the root apex, an inward direction 334 can also be defined, which is approximately perpendicular to the local edge surface 336 in the current edge point 330. Figure 8AA volumetric image data slice intersection view is shown, demonstrating different components of the movement of the edge point 330 during each iteration. By combining this inward direction 334 with the root direction 332, a plane generally orthogonal to the circumferential edge of the root near the current edge point 330 can be defined. Figure 8B An axial slice intersection view is shown, which demonstrates the plane and different inward directions 334 at different edge points 330. Once the root direction 332 and the inward direction 334 are determined for the non-converged edge point 330, the algorithm can begin to find its next position in the plane using the reference frame defined by the two directions (root direction 332 and inward direction 334). Figure 8A As shown, a set of possible stepping directions 328 may be calculated in the plane.

[0079] Figures 9A-9D A non-limiting example of a process for determining an adaptive step that centers the edge points 320 on a plane orthogonal to the direction of the tooth root is shown. Figures 9A-9D The template tooth grid 322 is shown for reference only. Once the orientation is determined, the step length can be calculated. The step length can be determined for each edge point 330 so that the edge points 330 are gradually aligned in an approximate plane orthogonal to the root orientation (see Figures 9A-9D ). For each step all non-converged points 330 are moved to their next voxel position, duplicates can be removed, and the position history (tooth mask) of each edge point 330 can be retained for use in subsequent steps.

[0080] Figure 10 Correction steps are shown that may be used to prevent edge points 330a, 330b from deviating from coplanar alignment. Figure 10 It is schematically shown how a correction step size can be applied to a first edge point 330a and a second edge point 330b that are coplanar in a plane 333 orthogonal to the root direction 332. Figure 10 In the illustrated diagram, the projection step 340a of the first edge point 330a is extended to a correction step 342a to maintain planar alignment of the first edge point 330a with the second edge point 330b in a plane orthogonal to the root direction 332 .

[0081] At the end of each iteration of the second apex detection phase, a convergence check can be initiated for all unconverged edge points 330. The convergence status of an edge point 330 may become "true" due to successful convergence or stagnation. An edge point 330 may be considered stagnant when the edge point direction calculated using the root direction 332 and the inward direction 334 causes the edge point 330 to move almost parallel to the expected edge point plane near the atypical local edge structure in the volume image data. In some configurations, stagnation of an edge point 330 can be detected by calculating a height estimate of the edge point 330. If the height estimate is greater than a threshold, then the edge point 330 can be marked as converged. The check for successful convergence aims to find a cluster of edge points 330 projected onto a two-dimensional (2D) plane and detect whether the projected edge points 330 are converging towards the apex. To assess the convergence of the edge point 330, a threshold (e.g., set to 1) can be applied to the standard deviation of the projected distance to the plane to check whether the adaptive step size has been sufficiently aligned with the edge point 330. If the standard deviation is above the threshold, the edge points 330 may be considered not yet aligned and the process may proceed to the next iteration. Otherwise, the process may continue clustering the projected edge points 330 and checking the convergence of each cluster of edge points 330.

[0082] Figures 11A-11C A visualization of a method of finding edge points 330 that converge to the root apex of a tooth is shown. Clustering of projected edge points 330 can be based on the detection of circular patterns. For each projected edge point 330, an approximate circle 352 is determined based on the edge point 330 being present in a small 2D Euclidean neighborhood, typically with a diameter of at most 3 mm, preferably at most 2 mm. Projected edge points 330 that are close enough to the resulting circle 352 can then be classified as being part of an initial cluster corresponding to the edge point 354 under consideration. The final clusters can be extracted based on the cluster size. The cluster circles 330 are calculated by estimating the local circle fit of each projected edge point 330. First, a Euclidean neighborhood with a radius (e.g., a 2 mm radius) can be determined around the edge point 324 under consideration. As Figures 11A-11C As shown, circle fitting can be initiated by determining the center 356 of the circle, which should be located at the approximate intersection of the directions defined by the projected gradient 358 at the edge point 330. This process can be divided into two steps. First, a set of initial clusters can be calculated based on the proximity of all projected edge points 330 to the center 356. Figures 11A-11CThree different initial clusters are shown for a given step of the algorithm. Next, obtaining a list of initial processes, an iterative process can be performed by first taking the largest cluster, and for each edge point 330 in the cluster, edge points 330 can be removed from the other initial clusters. The process can continue searching for the largest cluster of the updated list and repeating the previous steps until each edge point 330 is part of at most one final cluster. Empty clusters can be discarded.

[0083] Figures 12A-12C A non-limiting example of clusters 360 of edge points 330 that have passed the convergence check based on cluster size is shown. After each iteration, all non-converged edge points 330 can be checked for convergence and divided into final clusters. The convergence of each final cluster 360 can be evaluated based on three conditions. First, the cluster size condition can test whether the cluster 360 is very large. In one arrangement, the identified cluster with the largest number of edge points 330 is considered very large, while the second identified cluster is considered large enough only if it includes at least 50% of the number of edge points 330 in the largest cluster. If any subsequent cluster (third or fourth) includes at least 35% of the number of edge points 330 in the largest cluster, it is considered very large. Second, the radius size condition can test whether the radius associated with the final cluster 360 is small enough. Third, an inner product test can be determined between the cluster average gradient and the root direction 332. In some configurations, if the result of the inner product is less than a threshold, the result can be considered a pass. The threshold value can, but need not, be adjusted based on the tooth type being considered. For example, the threshold value can be set to a first value (e.g., -0.3) for a first tooth type (e.g., incisors and canines), a second threshold value (e.g., -0.35) for a second tooth type (e.g., premolars), and a third threshold value (e.g., -0.4) for a third tooth type (e.g., molars). If the final cluster 360 passes all three conditions, the cluster 360 can be considered a converged cluster 360. Figures 12A-12C The final clusters 360 at the end of different apex detection iterations are shown.

[0084] Figure 13A and Figure 13B A schematic diagram depicting the visualization of various directions in the calculation of potential root apex markers is depicted. Edge points 330 in a cluster 360 are represented by points arranged on a common circle 361, with cluster centers 362 centered relative to the edge points 330. The gradient direction 364 at each edge point 330 is shown by an arrow emanating from the edge point 330. The average gradient 366 of these edge points 330 is shown emanating from the point at the cluster center 362. The root direction 332 is also shown in FIG. Figure 13A and Figure 13BThe orthogonal gradient average 368 is shown as emanating from the point at the cluster center 362 in the root apex direction. Figure 13B The dotted arrow in FIG. 3 is shown as an arrow head emitting from the direction of the tooth root 332. Figure 13B As shown, the modified apical direction 370 may be the vector sum of the root direction 332 and the orthogonal gradient average 368 .

[0085] In order to determine the most reliable apical landmark for the current tooth, the algorithm can first evaluate which of the converged clusters 360 can be classified as sufficiently reliable. Each cluster 360 will be assigned an apical probability that reflects the chance that it represents the root apex. To calculate this probability, the following factors can be used: i.) the distance of the cluster 360 to the center of the segmented crown detected in the enhanced image data; ii.) the inner product of the gradient mean and the root direction. For each converged cluster 360, the apical probability can be given by the product of the two factors. In some arrangements, the potential apical landmark position that each converged cluster 360 will produce the root apex can also be calculated. Because the clusters 360 are detected as circular contours, they often converge slightly early, resulting in the cluster center being slightly closer to the crown than the actual apical position. In this way, the offset can be used to calculate the potential landmark position.

[0086] To prevent clusters 360 from providing invalid information to the next step in the algorithm, a filtering operation can be added, which can be based on the location of the corresponding potential apical markers of the clusters 360. First, a small neighborhood of voxel positions around the potential markers can be determined. This neighborhood can be a small cube of voxel positions centered on the voxel where the potential marker is located. If any of these voxel positions are outside the range of the volume image data, or if the intensity is equal to the background intensity, the marker and the corresponding converged cluster 360 are considered unreliable.

[0087] Figure 14 A non-limiting exemplary embodiment of apex detection of an upper canine 500 is shown. While the apex probability reflects the chance that the converged cluster 360 represents the apex 502, the selection probability reflects whether the found apex 502 belongs to the current tooth 500 or to an adjacent tooth 504. This additional probability is particularly relevant for teeth with only one root segment, which are typically narrow and whose apex may be close to the apex of its adjacent root segment (e.g., see Figure 14). For these teeth, the probability is defined by adding an additional factor to the apical probability, which indicates that the center of the segmented crown detected in the IO surface data should not be far away from the line defined by the cluster center and the root direction. For teeth with multiple root segments, the selection probability can be the same as the apical probability. Finally, the potential marker with the highest selection probability can be the main candidate for the most reliable apical marker. The final check includes a reliability check of the corresponding apical probability (values greater than 70% are empirically estimated to be sufficient).

[0088] Figures 15A-15C Non-limiting exemplary visualizations of apex detection results for different tooth types are shown. The segmented crown mesh 320 is shown as the darkest (i.e., black). Reliable edge point masks 380 are shown in light gray. Unconverged edge points and their callback positions 382 are shown in medium to dark gray and are displayed as vertical lines. Edge points that converged to unreliable clusters and their callback positions 384 are shown in Figure 15C 386 are shown only in light grey. The most reliable apex marker 386 is represented by a yellow square. The traced path of edge points contained within the converged clusters is a very useful approximation of the root edge in the volume image data, which can be used to exploit subsequent non-rigid registration (deformation). In order to extract a rough outline or mask of the root, the first step may be to apply a thresholding operation on the corresponding apex probabilities to determine which of the converged clusters are considered reliable. The same criteria as for the apex selection probability can be applied, with a probability greater than 70%. The final edge points that are members of any reliable converged cluster are collected and their previous positions are added using the callback information. After removing any duplicates that occur from the group, a mask of reliable edge points is obtained. Note that the mask may be empty when no clusters have converged or no cluster is considered reliable. In Figures 15A-15C A view of the dental shield 380 can be seen in FIG.

[0089] As described herein, once all the resulting components of the apex detection step 230 have been collected, the actual segmentation of the tooth can begin by deforming the shape of the template tooth. The resulting transformation obtained in the rigid initialization step 220 can be used to initialize the tooth template shape. Next, the tooth template shape can be non-rigidly deformed to match the segmented 10 crown scan. The method can begin by temporarily deforming the crown mesh 320 toward the template tooth 322 to determine the correspondence between the two meshes. The most reliable apex landmark 386 obtained from the apex detection step 230 can be used (after rigid initialization) to determine the closest apex landmark of the template shape, and this correspondence can be added to the crown-level correspondence. Based on the final set of correspondences, the template tooth mesh 322 can be deformed toward the segmented 10 crown mesh using ICP. This non-rigidly deformed template tooth now provides a closed tooth mesh that accurately matches the segmented crown surface, including a template root mesh portion whose length approximates the actual root length due to the inclusion of the apex landmark. Therefore, when the apex landmark correspondence is included, the deformation will approximately initialize the length and orientation of the root. Even when the tooth root consists of multiple root segments, segments without labeled correspondences can be roughly initialized by the deformable model.

[0090] After non-rigid initialization of the crown surface, an optimization method is used to reposition the vertex positions in the root portion of the deformed template tooth to the root edge identified in the volumetric image. In the coarse non-rigid root registration step 240 of the tooth segmentation algorithm, the registration of the root vertices of the template shape with the reliable edge point mask can be performed. The goal is to gradually deform the root vertices in the template tooth 322 towards the mask 380, starting from close to the crown and gradually moving towards the root apex. During the upcoming deformation, it may be desirable to move only the vertices marked as roots in the mesh, while keeping the vertices marked as crowns fixed. Smart sampling can be used to select only a subset of the crown vertices to reduce the required computational time. For computational efficiency and independence of voxel size, the same spatially uniform sampling can be applied to the mask 380.

[0091] In an embodiment, marking the vertices of the deformed template tooth mesh includes determining the distance to the nearest point on the segmented crown surface mesh for each vertex on the deformed template tooth mesh. Preferably, the nearest point is not limited to vertices of the segmented 10 tooth crown surface, but may also be located on any face or edge of the surface. If the distance for a vertex of the deformed template tooth falls below a given threshold, such as 0.4, 0.3, 0.2, or 0.1 mm, the vertex is marked as part of the crown surface; otherwise, the vertex is marked as part of the root surface. Note that deformed template tooth vertices in interstitial crown regions (not present in the segmented crown surface) will be marked as root vertices. For teeth with deep recesses in the crown surface, such as molars, the non-rigid deformation of the template tooth may not fully match the segmented crown surface in some locations, resulting in a few isolated vertices of the deformed template tooth having a distance from the segmented crown surface exceeding the distance threshold. These deformed template tooth vertices will be marked as roots. Therefore, an operation is preferably applied to the vertices marked as roots to retain only the largest connected portions. In the described operation, the labeling of the apex that was originally labeled as the root but not part of the largest connected portion was changed to the crown.

[0092] Figures 16A-16D Depicted is a visualization of different sets of edge points that can be used for non-rigid root initialization. In order to register the root portion of the (deformed) template tooth mesh 322 with a crown-to-apex progressive tooth mask, a set of continuous, possibly overlapping bands can be defined based on the Euclidean distance between the mask points 380 and the crown center of the current tooth detected in the 10 scan. Figures 16A-16D A visualization of these bands is shown. Figures 16A-16D In FIG, the fixed crown apex 394 is represented by a gray dot and is generally along the crown. Figures 16A-16D, the active band 390 is the mask points 380 shown as grey dots and appears as a patch outlined by a dotted line that appears sequentially from the crown to the root apex. The mask points 380 that are part of the further subsampled set 396 are represented by light grey dots. In some arrangements, a non-rigid registration method can now be used to deform the root shape towards the subsampled mask. The method will perform n iterations, where the kth iteration is associated with a band that will be called the active band 390 for that iteration. Based on the active band 390, correspondences between vertices in the template tooth mesh 322 and the (subsampled) mask points 380 are determined. These correspondences are then used for a thin plate spline deformation that is applied to all root vertices in the root shape. After the final deformation in the nth iteration, the resulting root shape will be the result of the mask registration. During iteration k of the non-rigid registration method, two sets of correspondences between the subsampled mask and the current template tooth shape are determined. The first set is centered on mask points 380 in the active band 390, while the second set is constructed very similarly using mask points 380 in the subsampled mask.

[0093] Typically, after a coarse registration of the root vertices of the template teeth on a rough outline (tooth mask) of the volumetric image data, a finer registration 250 of the root vertices of the template teeth with the volumetric image data itself is performed. Alternatively, the fine registration 250 is performed immediately after a non-rigid registration of the template teeth with the segmented crown surface, i.e., without a previous coarse registration with the tooth mask. Preferably, this fine repositioning of the root landmark vertices includes pre-processing of the volumetric image data of the one or more teeth to be segmented. In an embodiment, the pre-processing includes cropping the volumetric image data into a bounding box surrounding the registered deformed template teeth. This bounding box may tightly surround the deformed template teeth. Preferably, the tightly defined bounding box is extended on all six sides by one or more voxels, which define an area covering the expected area of the image where the teeth are located. For example, the tightly defined bounding box may be extended on all six sides by [4 / m] voxels, where m is the minimum voxel size in all its dimensions. In cases where the expanded bounding box falls outside the volume of the original volumetric image, the edges of the expanded bounding box outside the volumetric image are reduced to coincide with the volumetric image data boundaries. Subsequently, the cropped image is smoothed using, for example, a Gaussian filter. Thereafter, the image intensity of the smoothed image can be normalized between 0 and 1. Preferably, the optimization method is used to reposition the root landmark vertices in the deformed template tooth to the tooth edges identified in the smoothed and normalized cropped image that includes the deformed template tooth. The resulting image can be used to evaluate the cost function in the upcoming optimization method. Similar to hood registration, fine registration can also use a progressive zonal approach. In this case, the root vertices can be clustered into continuous, possibly overlapping zonal bands from crown to apex. During each iteration k of fine registration, an optimal deformation of the root can be determined based on the currently active zonal band. This optimal deformation can then be applied to all root vertices located within the zonal band or away from the tooth crown. More specifically, due to the non-rigid deformation of the template tooth, the fine repositioning of the root landmark vertices of the deformed template tooth leverages the already precise positioning of its crown landmark vertices to match the segmented crown surface. In an embodiment, the repositioning of the root marking vertices comprises assigning each of said root marking vertices to one of a series of consecutive bands based on a distance between said root marking vertex and a position of the tooth apex calculated based on one or more identified apex positions of said tooth. The banding method according to this embodiment allows for a gradual repositioning of the root marking vertices using an optimization method that takes into account the specific positioning of the crown marking vertices and propagates this certainty from the area near the crown towards the root apex. Preferably, said series of bands on the deformed model tooth surface partially overlap. Once the consecutive bands are defined, the optimization method is applied to each band before moving on to the next band. For each band The optimization method determines the best deformation, which is applied to the All template root vertices within and below this band (i.e. closer to the tooth vertex position) are recalculated. After applying the deformation, the vertex normals of the template teeth are recalculated and the method continues with the next band. The deformation itself can be performed using a thin plate spline deformation with variable stiffness. The stiffness can be relaxed since it is possible to allow for more local deformation of the template shape due to the improved starting position.

[0094] A cost function that can be used to evaluate a particular deformation can include three separate terms: a term based on the gradient norm, a term based on the inner product between the image gradient and the vertex normal, and an intensity uniformity cost term. This cost function profile can encourage optimization to improve template fit when approaching a favorable solution, while avoiding excessive deformations when placed in suboptimal positions in locally noisy volumetric image data. Finally, the optimal deformation during each iteration k can be found by minimizing the cost function using a trust region optimization method. After optimizing the position of the root marker vertex in the last (apical) zone, the registered template tooth is typically accurately matched to the segmented crown surface and tooth margin identified in the volumetric image, thereby providing an accurate segmentation of the selected tooth.

[0095] After the automatic tooth segmentation is completed, the reconstructed tooth root can have the position, size, orientation and shape of the tooth root in the volume image data. Figures 17A-19C The profiles of different views of an incisor 500, a premolar 502, and a molar 504 are shown, with an IO scan cross section 510 with CBCT slices shown in dotted lines and a registration template cross section 512 with CBCT slices shown in dashed-dotted lines. Figures 17A-17C Coronal, sagittal, and axial plane views, respectively, of a point map determined by automatic tooth segmentation of an incisor 500 in accordance with certain aspects of the present disclosure are shown. Figures 18A-18C Shown are coronal, sagittal, and axial plane views, respectively, of a point map determined by automatic tooth segmentation of premolars according to certain aspects of the present disclosure. Figures 19A-19C Shown are coronal, sagittal, and axial plane views, respectively, of a point map determined by automatic tooth segmentation of molars according to certain aspects of the present disclosure.

[0096] Figure 20A and Figure 20B Shown is a 3D visualization of automatic teeth segmentation results according to certain aspects of the present disclosure. Figure 20A A right view CBCT image 600 of a patient without the fitted template teeth 610 of the present disclosure is shown. Figure 20B A right view CBCT image 600 of a patient is shown with a fitting template tooth 610 of the present disclosure overlaid on the CBCT image 600 .

[0097] Figure 21A and 21BDepicted are 3D visualizations of automatic teeth segmentation results according to certain aspects of the present disclosure. Figure 21A A left view CBCT image 600 of a patient without the fitted template teeth 610 of the present disclosure is shown. Figure 21B A left view CBCT image 600 of a patient is shown with a fitting template tooth 610 of the present disclosure overlaid on the CBCT image 600 .

[0098] Figure 22 Validation results for two datasets are shown. Dataset 1 contains cases commonly seen in orthodontic treatment. Dataset 2 contains common cases that may have braces or crown fillings that affect the quality of the CBCT scan image. Success rates are shown for the full dataset (Dataset 1 + Dataset 2) and for a filtered dataset with out-of-range cases removed. For the filtered data, a comparison of success rates for teeth with and without braces is also shown. Figure 22 Also shown are filtered data comparing the success rates for teeth with and without crown fillings. Figure 22 The bottom table in Figure 2 shows a comparison of the success rates of different tooth types for filtering data. In the table, the abbreviations U and L represent upper or lower teeth, and the abbreviations I, C, P, and M represent incisors, canines, premolars, and molars, respectively.

[0099] Other variations and terms

[0100] While certain embodiments have been described, these embodiments are presented by way of example only and are not intended to limit the scope of protection. Indeed, the novel methods and systems described herein may be embodied in a variety of other forms. Those skilled in the art will appreciate that the present disclosure extends beyond the specifically disclosed embodiments to other alternative embodiments or uses and obvious modifications and equivalents thereof, including embodiments that do not provide all of the features and advantages set forth herein. Furthermore, various omissions, substitutions, and changes in the form of the methods and systems described herein may be made. Those skilled in the art will appreciate that, in some embodiments, the actual steps taken in the processes illustrated or disclosed may differ from those shown in the accompanying drawings. Depending on the embodiment, some of the steps described above may be eliminated; others may be added. Therefore, the scope of the present disclosure is not intended to be limited by the specific disclosure of the preferred embodiments herein, but rather by the claims set forth herein or in the future. The language of the claims is to be interpreted broadly based on the language used in the claims and is not limited to the examples described in the patent specification during the application process, which are to be construed as non-exclusive.

[0101] Features, materials, characteristics or combinations described in conjunction with a particular aspect, embodiment or example are to be understood to be applicable to any other aspect, embodiment or example described herein unless incompatible therewith. All features disclosed in this specification (including any accompanying claims, abstract and drawings), or all steps of any method or process so disclosed, may be combined in any combination, except combinations in which at least some of such features or steps are mutually exclusive. Protection extends to any novel feature or any novel combination of features disclosed in this specification (including any accompanying claims, abstract and drawings), or to any novel feature or any novel combination of steps of any method or process so disclosed.

[0102] Conditional language, such as "can," "could," "might," or "may," unless otherwise specifically stated or understood otherwise in the context of use, is generally intended to convey that certain embodiments include certain features, elements, or steps, while other embodiments do not. Thus, such conditional language is generally not intended to imply that one or more embodiments require a feature, element, or step in any way. The terms "comprising," "including," "having," and the like are synonymous and are used inclusively in an open manner and do not exclude additional elements, features, actions, operations, and the like. Additionally, the term "or" is used in its inclusive sense (rather than its exclusive sense), so that when used, for example, to connect a series of elements, the term "or" means one, some, or all of the elements in the list. Additionally, the term "each," as used herein, in addition to having its ordinary meaning, may also mean any subset of a set of elements to which the term "each" applies.

[0103] Conjunctive language, such as phrases such as "at least one of X, Y, and Z," unless expressly stated otherwise, is generally understood, depending on the context, to convey that an item, term, etc., can be X, Y, or Z. Thus, such conjunctive language is generally not intended to imply that certain embodiments require the presence of at least one X, at least one Y, and at least one Z.

[0104] As used herein, language of degree, such as the terms "approximately," "about," "generally," and "substantially," as used herein, refers to a value, amount, or characteristic that is close to a stated value, amount, or characteristic that still performs a desired function or achieves a desired result. For example, the terms "approximately," "about," "generally," and "substantially" may refer to an amount that is less than 10% of the stated amount. As another example, the terms "substantially parallel" and "substantially parallel" may refer to a value, amount, or characteristic that deviates from perfect parallelism by less than 15 degrees.

Claims

1. A computer-implemented method for segmenting a tooth model based on volumetric image data and intraoral surface data of a subject's dental anatomy, comprising the following steps: Provides volumetric image data of tooth anatomy; Provide intraoral surface data of tooth anatomy; generating enhanced image data by aligning the surface data and the volumetric image data; Segment the crown of the tooth in the intraoral surface data; selecting a template tooth from a template tooth library, wherein a tooth identifier of the template tooth corresponds to a tooth identifier of a tooth in the enhanced image data; fitting a corresponding template tooth to the tooth of the enhanced image data; and Teeth in the volumetric image data are segmented based on the fitted template teeth.

2. The computer-implemented method of claim 1 , wherein: Fitting the corresponding template tooth to the tooth of the enhanced image data includes aligning the corresponding template tooth with the segmented crown portion of the tooth of the enhanced image data.

3. The computer-implemented method of claim 2, wherein: Fitting the corresponding template tooth to the tooth of the augmented image data further comprises a scaled rigid transformation of the corresponding template tooth relative to the segmented crown portion of the tooth in the augmented image data.

4. The computer-implemented method of claim 1 , further comprising selecting two or more template teeth from a template teeth library, in, a tooth identification of each of the selected template teeth corresponding to the tooth identification of the tooth of the enhanced image data, wherein a crown center position is determined for each of the selected template teeth; Virtually positioning the selected template teeth into the dental arch; and A scaled rigid transformation of the virtual dental arch is performed to map the crown center position of the template tooth to the crown center position determined for the segmented crown portion of the corresponding tooth in the enhanced image data. 5 . The computer-implemented method of claim 4 , further comprising strictly scaling the template teeth based on a distance between crown center locations of corresponding adjacent teeth in the enhanced image data.

6. The computer-implemented method of claim 4 or 5, further comprising a scaled rigid transformation of the segmented crown portion of the template tooth relative to the corresponding tooth of the augmented image data to which the template tooth is mapped.

7. The computer-implemented method of claims 1 to 6, further comprising determining a contour of a root portion of a tooth in the enhanced image data, in, Determining the contour includes determining edge points of the tooth in the volume image data, starting from edge points that coincide with the apical boundary of the segmented crown portion of the aligned surface data, and propagating these edge points in the apical direction along the root edge of the tooth in the volume image data.

8. The computer-implemented method of claim 7, wherein: Propagating edge points in the apical direction along the root edge of the tooth in the volumetric image data is an iterative step-by-step process guided by the root direction and the inward direction at the current edge point at a given iteration, wherein the root direction provides an indication of the direction towards the root apex, and The inward direction is provided by the gradient direction at a given edge position of the tooth.

9. The computer-implemented method of claim 8, wherein: Providing a root orientation of the tooth of the enhanced image data includes aligning the template tooth with the tooth of the enhanced image data, wherein a tooth identification of the template tooth corresponds to a tooth identification of the tooth of the enhanced image data.

10. The computer-implemented method of claim 9, wherein: Providing the root orientation of the tooth of the enhanced image data also includes a scaled rigid transformation of the corresponding template tooth relative to the segmented crown surface of the tooth of the enhanced image data.

11. The computer-implemented method of claim 8, wherein: The root direction of a tooth in the enhanced image data is an adaptive direction provided by the direction between the crown center of the tooth and the center of the current set of edge points in a given iteration.

12. The computer-implemented method of claims 7 to 11, wherein: At each iteration, the set of edge points is tested for convergence to the same location.

13. The computer-implemented method of claim 12, wherein: The apical marker position was derived from the set of edge points that passed the convergence test.

14. The computer-implemented method of claim 13, wherein: The outline of the root of the tooth in the enhanced image data is determined by the voxel positions of the edge points detected in the iterative steps resulting in a converged set of edge points.

15. The computer-implemented method of claims 1 to 14, wherein: Fitting the corresponding template tooth to the tooth of the augmented image data further comprises deforming the template tooth into a deformed template tooth mesh that matches the segmented crown portion and preferably matches the root apex marking position of the tooth of the augmented image data.

16. The computer-implemented method of claim 15, wherein: Vertices of the deformed template tooth are labeled as crown vertices or root vertices based on proximity between the deformed template tooth mesh and the segmented crown portion of the matching tooth of the enhanced image data.

17. The computer-implemented method according to any one of claims 15 to 16, wherein: Fitting the corresponding template teeth to the teeth of the enhanced image data further includes deforming a root portion of the deformed template tooth mesh to align vertices of the root portion with edges of the selected tooth in the volumetric image data.

18. The computer-implemented method of claim 17, wherein: Deforming the root portion of the deformed template tooth mesh to align vertices of the root portion with edges of the selected tooth in the volumetric image data comprises a coarse registration step, wherein the vertices of the root portion of the deformed template tooth mesh are aligned with the contour of the root of the tooth in the enhanced image data determined according to claims 7 to 12.

19. The computer-implemented method of claim 18, wherein: The coarse registration step is followed by a fine registration step, wherein the root portion of the deformed template tooth mesh is further registered to the root margin of the corresponding tooth in the volumetric image data of the enhanced image data.

20. The computer-implemented method of any preceding claim, wherein: The template tooth includes a tooth crown mesh and a tooth root mesh, and the mesh density of the tooth crown mesh is higher than the mesh density of the tooth root mesh.

21. A computer system comprising: a memory and a processor in communication with the memory, the processor being configured with processor-executable instructions to perform operations comprising: receiving volumetric image data of a subject's dental anatomy; receiving intraoral surface data of a tooth anatomical structure; generating enhanced image data by aligning the surface data and the volumetric image data; Segment the crown of the tooth in the intraoral surface data; selecting a template tooth from a template tooth library, wherein a tooth identifier of the template tooth corresponds to a tooth identifier of a tooth in the enhanced image data; fitting a corresponding template tooth to the tooth of the enhanced image data; and Teeth in the volumetric image data are segmented based on the fitted template teeth.

22. A non-transitory computer-readable medium storing computer-executable instructions that, when executed by one or more computer systems, configure the one or more computer systems to perform operations comprising: receiving volumetric image data of a subject's dental anatomy; receiving intraoral surface data of a tooth anatomical structure; generating enhanced image data by aligning the surface data and the volumetric image data; Segment the crown of the tooth in the intraoral surface data; selecting a template tooth from a template tooth library, wherein a tooth identifier of the template tooth corresponds to a tooth identifier of a tooth in the enhanced image data; fitting a corresponding template tooth to the tooth of the enhanced image data; and Teeth in the volumetric image data are segmented based on the fitted template teeth.

Images