Method and system for tooth pose estimation

Abstract

According to an embodiment, a computer-implemented method for tooth pose estimation is disclosed. The method comprises obtaining a virtual 3D representation (100) representing the patient's dentition, segmenting the virtual 3D representation (100) to obtain at least a first segmented tooth representation (101) and a second segmented tooth representation (105), where the at least first and second segmented tooth representations (101, 105) represent adjacent teeth in the patient's dentition. Furthermore, the method comprises determining an initial tooth pose (302) of the first segmented tooth representation (101) using the geometric parameters of the first segmented tooth representation (101) and the geometric parameters of the second segmented tooth representation (105). A normalized tooth representation (301) of the first segmented tooth representation (101) is obtained by transforming the first segmented tooth representation (101) using the initial tooth pose (302) of the first segmented tooth representation (101). The normalized tooth representation (301) of the first segmented tooth representation (101) is then input into a trained neural network (400). An output is generated from the trained neural network (400), which includes the correct tooth pose of the first segmented tooth representation (101).

Description

Technical Field

[0001] The present disclosure relates to computer-implemented methods and systems for tooth pose estimation in a virtual 3D representation of a patient's dentition. Background Art

[0002] The development of intraoral scanning technology has played an important role in the transformation to modern digital dentistry. The use of intraoral 3D scanners allows dental professionals to accurately and quickly capture the patient's intraoral situation, which can then be visualized as a virtual 3D model or virtual 3D representation on a display. The virtual 3D representation of the patient's teeth thus obtained can be used as a digital impression, which has many advantages over traditional physical impressions. The overall efficiency of dental surgery is thus improved, because, for example, the created digital impression can be used for various applications, such as dental restoration design or orthodontic treatment planning. In addition, the virtual 3D representation can also be used for diagnostic purposes in order to visualize the presence of dental conditions such as caries, plaque, gingivitis or tooth wear to the user (e.g. dentist).

[0003] When using virtual 3D representations in any of the above applications, accuracy can be very important. Therefore, it is necessary to correctly determine the tooth pose of each individual tooth representation within the virtual 3D representation of the patient's tooth group. The tooth pose can be understood as a local coordinate system for the tooth representation, wherein the local coordinate system includes a tooth-specific origin and three mutually orthogonal axes that intersect at the origin. The tooth pose can be set manually, which is a laborious and time-consuming process and involves manually annotating the coordinate system axes of each tooth representation. Other methods for determining tooth pose may include the use of mathematical algorithms, such as principal component analysis (PCA), which may take many seconds to output the tooth pose. The required computation time makes current methods less attractive for performance-sensitive applications, for which the tooth pose needs to be estimated within a limited time frame (e.g., less than one second). In addition, if the patient's teeth are severely worn, the method of tooth pose estimation using the principal component analysis algorithm may be prone to errors.

[0004] It would be desirable to provide an alternative method for tooth pose estimation that is capable of computing tooth pose in an efficient manner, eg, in less time than current algorithms allow. Summary of the invention

[0005] The invention is set out in the appended claims.

[0006] In one embodiment, a computer-implemented method for tooth posture estimation is disclosed, wherein the method comprises:

[0007] - obtaining a virtual 3D representation representing the patient's dentition;

[0008] - segmenting the virtual 3D representation to obtain at least a first segmented tooth representation;

[0009] - inputting at least a first segmented tooth representation into a trained neural network;

[0010] - generating an output from the trained neural network, wherein the output comprises a correct tooth pose of at least a first segmented tooth representation.

[0011] The term "tooth pose" may be understood as corresponding to a local coordinate system of a tooth representation. In the present disclosure, the terms "tooth pose" and "local coordinate system" have the same meaning and may be used interchangeably. Thus, a tooth pose may include three mutually orthogonal axes intersecting at a tooth-specific origin. The local coordinate system may be placed within the tooth representation. The tooth-specific origin may be placed at a center point of the tooth representation. The center point may be a geometric parameter, such as a centroid or center of mass of the tooth representation. The local coordinate system may be defined based on the morphological properties of each tooth representation.

[0012] For example, the first axis of the tooth pose can be located in the direction of the adjacent tooth representation along the mesiodistal direction of the tooth representation. When viewing the patient's teeth in a dental quadrant, if the tooth pose is associated with a tooth on the left side of the maxillary (upper) dental arch or a tooth on the right side of the mandibular (lower) dental arch, the first axis of the tooth pose can point in the distal direction. If the tooth pose is associated with a tooth on the right side of the maxillary dental arch or a tooth on the left side of the mandibular dental arch, the first axis of the tooth pose can point in the mesial direction.

[0013] For example, the second axis of the tooth posture can be along the crown direction of the tooth representation and can be orthogonal to the occlusal surface or bite surface of the tooth representation. The crown direction can be understood as the direction from the root to the crown of the tooth, in the direction of the crown of the tooth. The third axis of the tooth posture can be orthogonal to the first and second axes and can be outward along the vestibular or facial direction. The tooth posture defined in this way can also be called the correct tooth posture, the target tooth posture or the correct local coordinate system.

[0014] A method according to an embodiment includes obtaining a virtual 3D representation representing a patient's dentition. The virtual 3D representation of the patient's dentition can be constructed based on scan data collected during an intraoral scanning process, in which an intraoral scanner is used to scan the patient's oral condition including teeth and gums. As an alternative to the intraoral scanner, a so-called laboratory scanner can be used to scan a plaster model of the patient's dentition. The captured scan data can be used to construct a virtual 3D representation that accurately represents the scanned teeth and gums. The virtual 3D representation can be stored in a memory device of a computer system and accessed from the memory device of the computer system. Typically, the virtual 3D representation can be displayed on a display screen in the form of a 3D mesh (e.g., a triangular mesh). Alternatively, the virtual 3D representation can be displayed on a display screen in the form of a point cloud, a volume representation, or any other suitable 3D representation. The virtual 3D representation representing the patient's dentition can be referred to as a virtual 3D representation.

[0015] Furthermore, the method comprises segmenting the virtual 3D representation to obtain at least a first segmented tooth representation and a second segmented tooth representation. At least the first and second segmented tooth representations may include adjacent teeth in the patient's dentition. The segmentation process allows for the identification of different dental objects, such as individual teeth and / or surrounding gums, in the virtual 3D representation representing the patient's dentition having a plurality of teeth. The term dentition may include the arrangement and morphology of teeth in at least one of the upper and lower jaws. Thus, the output of the segmentation process may include the identification of individual tooth representations, typically in the form of a tooth mesh or a tooth point cloud.

[0016] At least the first and second segmented tooth representations are preferably representations of adjacent teeth in the patient's dentition. This is useful in the tooth posture estimation method because the initial tooth posture of at least the first segmented tooth representation can be determined or estimated based on the geometric parameters of the first segmented tooth representation and the geometric parameters of the second segmented tooth representation. Therefore, the initial tooth posture is simply calculated and can be a very good estimate of the correct tooth posture. The term "adjacent teeth" should be understood as two teeth placed adjacent to each other in the upper jaw or two teeth placed adjacent to each other in the lower jaw. Most commonly, adjacent teeth have consecutive numbers in the Universal Numbering Notation or System (UNN), in which numbers 1-32 are assigned to human teeth. In some cases, adjacent teeth are not marked with consecutive numbers in the UNN symbol, for example if the teeth between the current adjacent teeth have been removed before.

[0017] According to an embodiment of the present disclosure, the method further includes determining an initial tooth posture of the first segmented tooth representation using geometric parameters of the first segmented tooth representation and geometric parameters of the second segmented tooth representation. In the case where the tooth posture of the first segmented tooth representation is accurately determined using a trained neural network, the step of determining the initial tooth posture of the first segmented tooth representation can be regarded as a preprocessing step for creating an input of the trained neural network, which enables the trained neural network to generate a more accurate output in a faster manner compared to a tooth posture estimation method without the preprocessing step.

[0018] In one embodiment, the geometric parameters of the first segmented tooth representation and the geometric parameters of the second segmented tooth representation may be the centroid of the first segmented tooth representation and the centroid of the second segmented tooth representation, respectively. The centroid of the segmented tooth representation is a point in 3D space located at the geometric center of the segmented tooth representation.

[0019] In an alternative embodiment, the geometric parameters of the first segmented tooth representation and the geometric parameters of the second segmented tooth representation may be the center of gravity of the first segmented tooth representation and the center of gravity of the second segmented tooth representation, respectively. The center of gravity of the segmented tooth representation is a point in 3D space that is located at the center of gravity of the segmented tooth representation. If the segmented tooth representation represents molars and / or premolars, the cusps located at the top of the teeth may cause the center of gravity to be located closer to the top of the tooth representation than the center of mass of the tooth representation. The advantage of using the center of mass over the center of gravity may be reflected in that the center of mass is more robust to different tooth morphologies.

[0020] Alternatively, the geometric parameter of the first segmented tooth representation may be the center of gravity of the first segmented tooth representation, and the geometric parameter of the second segmented tooth representation may be the center of mass of the second segmented tooth representation, or the geometric parameter of the first segmented tooth representation may be the center of mass of the first segmented tooth representation, and the geometric parameter of the second segmented tooth representation may be the center of gravity of the second segmented tooth representation. In yet another embodiment, the geometric parameter of the first or second segmented tooth representation may be a point on the longitudinal axis of the first or second segmented tooth representation, wherein the point may be at a certain distance from the occlusal surface of the first or second segmented tooth representation.

[0021] The geometric parameter may be a point associated with the center of the segmented tooth representation. Thus, the geometric parameter of a first segmented tooth representation may be a point associated with the center of the first segmented tooth representation, such as the center of mass or center of gravity of the first segmented tooth representation. The geometric parameter of a second segmented tooth representation may be a point associated with the center of the second segmented tooth representation, such as the center of mass or center of gravity of the second segmented tooth representation.

[0022] Determining the initial tooth posture of the first segmented tooth representation may include determining an initial local coordinate system of the first segmented tooth representation. This in turn means determining the origin of the initial local coordinate system and determining the three axes of the initial local coordinate system. The origin of the initial local coordinate system of the first segmented tooth representation may be located in the geometric parameters of the first segmented tooth representation, such as at the center of mass of the first segmented tooth representation. The center of mass of the second segmented tooth representation representing the teeth adjacent to the teeth represented by the first segmented tooth representation may be used to determine the first axis of the initial local coordinate system of the first segmented tooth representation. For example, the difference between the geometric parameters (such as the center of mass) of the first and second segmented tooth representations may be calculated to determine the first axis of the initial local coordinate system of the first segmented tooth representation. Therefore, the first axis of the initial local coordinate system of the first segmented tooth representation will point to the center of mass of the second tooth representation. Instead of the center of mass of the first and second segmented tooth representations, or in addition to the center of mass of the first and second segmented tooth representations, another geometric parameter may be used to determine the first axis of the initial local coordinate system of the first segmented tooth representation, such as the center of mass of the first and second segmented tooth representations.

[0023] If the teeth represented by the first and second segmented tooth representations are perfectly aligned relative to each other, the first axis of the initial tooth posture of the first segmented tooth representation will closely match the first axis of the correct tooth posture of the first segmented tooth representation. The teeth in the maxilla or mandible are generally considered to be aligned if the placement of the teeth in the maxilla or mandible follows an ideal dental arch line (which can be U-shaped). If the teeth represented by the first and second segmented tooth representations are not aligned relative to each other, the first axis of the initial tooth posture of the first segmented tooth representation will not match the first axis of the correct tooth posture of the first segmented tooth representation. However, in this case, the first axis of the initial tooth posture can be a good estimate of the first axis of the correct tooth posture.

[0024] Once the first axis of the initial tooth posture of the first segmented tooth representation is determined, the second axis and the third axis of the initial tooth posture can be determined. For example, the second axis can point from the center of mass of the first segmented tooth representation to the occlusal surface of the first segmented tooth representation. The second axis of the initial tooth posture of the first segmented tooth representation can be calculated as the mean facet normal of all facets of the tooth representation in the virtual 3D representation. The facet normal is a vector orthogonal to the direction of a given facet. Once the second axis of the initial tooth posture of the first segmented tooth representation is determined, the third axis of the initial tooth posture of the first segmented tooth representation can be found as the cross product of the first axis and the second axis.

[0025] In an embodiment, the method further comprises obtaining a standardized tooth representation of the first segmented tooth representation by transforming the first segmented tooth representation using the initial tooth pose of the first segmented tooth representation. The transformation may comprise rotating the first segmented tooth representation according to an initial local coordinate system or initial tooth pose of the first segmented tooth representation, and translating the first segmented tooth representation according to a negative value of the coordinates of the center of mass of the first segmented tooth representation. The result of the standardization is reflected in bringing the first segmented tooth representation into the origin of the global coordinate system with coordinates (0, 0, 0), and the axes of the initial tooth pose of the first segmented tooth representation overlap with the corresponding axes of the global coordinate system.

[0026] Obtaining a standardized tooth representation of the first segmented tooth representation is useful because each vertex of the first segmented tooth representation is placed around the origin of the global coordinate system with coordinates (0, 0, 0). If the trained neural network processes coordinates of multiple segmented tooth representations, all segmented tooth representations are placed around the origin of the global coordinate system, which makes the operations performed by the trained neural network faster and the overall performance of the trained neural network more robust.

[0027] In an embodiment, a standardized tooth representation of the first segmented tooth representation is used as an input to a trained neural network. The input to the trained neural network may include an approximation of the tooth pose of the first segmented tooth representation, and the output of the trained neural network provides the correct tooth pose of the first segmented tooth representation. The trained neural network may be a deep learning network capable of directly processing three-dimensional (3D) point clouds. An example of such a network architecture is a PointNet neural network developed by Stanford University that is invariant to the order of points on the input ("PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation", Charles R. Qi, Hao Su, Kaichun Mo, Leonidas J. Guibas). Alternatively, PointNet++ or a PointNet-like network architecture may be used for the purpose of tooth pose estimation. Alternatively, the trained neural network may process two-dimensional data, where the two-dimensional data may be obtained by converting a three-dimensional point cloud (e.g., by a so-called mesh flattening algorithm) into two-dimensional data.

[0028] After obtaining the correct tooth pose of the first segmented tooth representation, a segmented virtual 3D representation representing the patient's dentition can be displayed, which has the correct tooth pose of the first segmented tooth representation. The display can be performed, for example, on a display screen. Additionally or alternatively, the display can include displaying only the first segmented tooth representation, which has the correct tooth pose of the first segmented tooth representation.

[0029] In one embodiment, the method may include displaying to a user a first segmented tooth representation having one axis, such as a second axis of the tooth pose of the first segmented tooth representation. The second axis may be along the crown direction of the first segmented tooth representation and may be orthogonal to the occlusal plane or occlusal surface of the first segmented tooth representation. The second axis of the tooth pose of the first segmented tooth representation may be used to design orthodontic treatment. In particular, it may be used to determine how the teeth represented by the first segmented tooth representation may be rotated in the jaw. This may be particularly beneficial for dental professionals to predict the outcome of orthodontic treatment.

[0030] In one embodiment, the method may include quantifying the mutual alignment of two adjacent teeth using at least one of the axes (e.g., the first axis of the tooth posture of the first segmented tooth representation and the first axis of the tooth posture of the second segmented tooth representation). For example, the first axis of the tooth posture of the first segmented tooth representation intersects with the first segmented tooth representation to determine the mesial point and the distal point of the first segmented tooth representation. This may be helpful for dental professionals to determine how the tooth represented by the first segmented tooth representation is aligned relative to at least one of its adjacent teeth. That is, if the tooth represented by the first segmented tooth representation is aligned with at least one of its adjacent teeth, the mesial point or the distal point of the first segmented tooth representation is substantially close to the distal point or the mesial point of the segmented tooth representation representing at least one adjacent tooth, respectively.

[0031] In one embodiment, the geometric parameter of the first segmented tooth representation can be the centroid of the first segmented tooth representation, and the geometric parameter of the second segmented tooth representation can be the centroid of the second segmented tooth representation. The centroid represents the geometric center of a given volume. Therefore, the centroid of the first segmented tooth representation is located at the geometric center of the first segmented tooth representation, and the centroid of the second segmented tooth representation is located at the geometric center of the second segmented tooth representation.

[0032] The centroid of the first segmented tooth representation can be identified as the center of a bounding box placed around the first segmented tooth representation. A bounding box is a three-dimensional box or a minimum-sized cuboid that encloses the first segmented tooth representation. The centroid can be determined by identifying a first point of the bounding box with the lowest x, y, z coordinate value and a second point of the bounding box with the highest x, y, z coordinate value and calculating the average of the first point and the second point. In a similar manner, the centroid of the second segmented tooth representation can be identified.

[0033] In an embodiment according to the present disclosure, determining an initial tooth pose of the first segmented tooth representation using geometric parameters of the first and second segmented tooth representations may include determining first, second, and third axes of the initial tooth pose of the first segmented tooth representation. The first, second, and third axes of the initial tooth pose are orthogonal to each other and intersect at a center point of the first segmented tooth representation. The center point may be a geometric parameter, such as a center of mass or a center of gravity of the first segmented tooth representation.

[0034] The first axis of the initial tooth posture of the first segmented tooth representation may point to the center point of the second segmented tooth representation, which in one embodiment may be the center of mass of the second segmented tooth representation. The second axis of the initial tooth posture of the first segmented tooth representation may be along the crown direction of the first segmented tooth representation and may be orthogonal to the occlusal plane or occlusal surface of the first segmented tooth representation. The crown direction may be understood as the direction from the root to the crown, in the direction of the crown. The third axis of the initial tooth posture of the first segmented tooth representation may be orthogonal to the first and second axes, and may be oriented from the jaw outward along the vestibular or facial direction.

[0035] The first axis of the initial tooth pose of the first segmented tooth representation can be calculated as the difference between the center of mass of the first segmented tooth representation and the center of mass of the second segmented tooth representation. In this way, an approximation of the first axis of the correct tooth pose of the first segmented tooth representation can be found.

[0036] The first axis of the correct tooth posture of the first segmented tooth representation may be along the mesiodistal direction of the first segmented tooth representation.

[0037] The second axis of the initial tooth posture of the first segmented tooth representation is orthogonal to the occlusal surface of the first segmented tooth representation. Generally speaking, the occlusal surface of a tooth or tooth representation can be regarded as the occlusal surface of the tooth for chewing or grinding. The second axis of the initial tooth posture of the first segmented tooth representation can be calculated as the average patch normal of all patches forming the tooth representation in the virtual 3D representation. A patch is a surface on a 3D mesh defined by at least three points or vertices, which are interconnected by straight lines or edges. In the case of a point cloud representation of the first segmented tooth representation, the second axis of the initial tooth posture of the first segmented tooth representation can be calculated as the average vertex normal of all vertices forming the tooth representation in the virtual 3D representation.

[0038] According to the method of the embodiment, the third axis of the initial tooth posture represented by the first segmented tooth can be obtained as a cross product between the first axis of the initial tooth posture represented by the first segmented tooth and the second axis of the initial tooth posture represented by the first segmented tooth. The third axis of the initial tooth posture represented by the first segmented tooth can be orthogonal to the first and second axes of the initial tooth posture represented by the first segmented tooth, and can face outward along the vestibular or facial direction.

[0039] The method according to an embodiment may also include a step of adjusting the first segmented tooth representation by reducing a number of vertices and / or patches representing the first segmented tooth representation. The reduction in the number of vertices may also be referred to as mesh extraction. One way in which vertex reduction may be performed is through a mesh-specific operation in which edges with low curvature are collapsed so that no useful information is lost. For example, edge collapse includes removing an edge by merging two vertices connected by the edge into a new vertex at a new position (e.g., at the center of the collapsed edge). The mesh extraction step is advantageous because it facilitates faster estimation of the overall tooth posture because the trained neural network processes less input information.

[0040] In an embodiment where the first segmented tooth representation is represented by a point cloud, adjusting the first segmented tooth representation may be achieved by reducing the number of points in the point cloud. A new point cloud may be constructed in an iterative manner by adding points from the point cloud representing the first segmented tooth representation to the new point cloud if a predetermined criterion (e.g., a minimum distance condition) is met. For example, a minimum point-to-point distance within a point cloud may be defined. Then, if a point from the point cloud representing the first segmented tooth representation is at least the minimum point-to-point distance away from all other points in the point cloud representing the first segmented tooth representation, the point is added to the new point cloud. In this way, a new point cloud is created in which a uniform distribution of points is ensured.

[0041] In some embodiments, a method for tooth pose estimation may include the use of a trained neural network. Advantageously, the tooth pose may be correctly estimated more quickly with the aid of a trained neural network than using an algorithmic approach. In addition, the trained neural network is configured to handle a wider range of so-called edge cases where teeth may be extremely misaligned. A neural network that is particularly suitable for this purpose is a PointNet neural network or a similar type of neural network that is capable of making predictions on a point cloud as input. A point cloud is represented as a set of three-dimensional points, where each point is a vector of its coordinates. Three main properties of point clouds are that they are disordered, there are interactions between the points, and the point set is invariant to transformations such as translation and rotation. These properties may define the necessary requirements for point cloud processing by a trained neural network.

[0042] The output of the trained neural network may include a three-dimensional rotation vector in an axis-angle representation. The three-dimensional rotation vector may describe how the first segmented tooth representation should be rotated so as to be placed in its correct local coordinate system, and may be described by two quantities, a unit vector indicating the direction of the rotation axis and an angle describing the magnitude of the rotation about the rotation axis.

[0043] The output of the trained neural network may also include a three-dimensional translation vector including the amount of translation of the first segmented tooth representation. The three-dimensional translation vector may describe how the first segmented tooth representation should be translated in order to be placed in its correct local coordinate system.

[0044] According to an embodiment, a computer-implemented method for training a neural network for dental pose estimation is disclosed, the method comprising accessing an input training data set, the input training data set comprising training dental representations and a reference truth dental pose for the training dental representations, wherein the reference truth dental pose for the training dental representations is a dental pose manually assigned to the training dental representations, inputting at least one training dental representation into a neural network for dental pose estimation, obtaining an output of the neural network for dental pose estimation, the output comprising a predicted dental pose for the training dental representation, comparing the predicted dental pose for the training dental representation with the reference truth dental pose for the training dental representation to determine a loss and minimize the loss. In an example, the loss can be a function of a geodesic distance between the predicted dental pose for the training dental representation and the reference truth dental pose for the training dental representation.

[0045] In one embodiment, the loss can be the sum of the geodesic distance between the predicted tooth pose of the training tooth representation and the reference truth tooth pose of the training tooth representation, and the squared value of the Euclidean distance between the origin of the predicted tooth pose of the training tooth representation and the origin of the reference truth tooth pose of the training tooth representation. The sum can then be minimized using a stochastic gradient descent algorithm.

[0046] The training tooth representation can be normalized to bring it into the origin of the global coordinate system, and then its vertices are extracted and fed into a neural network for tooth pose estimation for training purposes.

[0047] The described training process can be performed on the remaining training tooth representations contained in the input training dataset.

[0048] The training of a neural network is an iterative process that is repeated until the neural network can no longer improve its output. This means that after a number of iterations, the output of the neural network continues to be equal to the ground truth data. The number of iterations required to train a neural network can be between 200,000 and 400,000 iterations, but this can depend on the complexity of the neural network.

[0049] A dental scanning system may include a data processing device configured to perform a method according to one or more embodiments of the present disclosure.

[0050] A non-transitory computer-readable storage medium may be included in the dental scanning system. The non-transitory computer-readable medium may carry instructions that, when executed by a computer, cause the computer to perform a method according to one or more embodiments of the present disclosure.

[0051] A computer program product may be embodied in a non-transitory computer-readable storage medium. The computer program product may include instructions, which, when executed by a computer, cause the computer to perform a method according to any embodiment presented herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Various aspects of the present disclosure may be best understood from the following detailed description in conjunction with the accompanying drawings. These drawings are schematic and simplified, and for clarity, they only show details to improve understanding of the claims, while other details are omitted. The individual features of each aspect may be combined with any or all features of the other aspects, respectively. These and other aspects, features and / or technical effects will be apparent from and explained with reference to the drawings described below, in which:

[0053] Figure 1 A virtual 3D representation of the mandible is shown, wherein the posture of the teeth is displayed;

[0054] Figure 2 is a flow chart illustrating a method according to an embodiment of the present disclosure;

[0055] Figure 3A An initial tooth pose of a first segmented tooth representation is shown;

[0056] Figure 3B showing a normalized tooth representation of a first segmented tooth representation;

[0057] Figure 3C An initial tooth pose of a first segmented tooth representation is shown;

[0058] Figure 3D The correct tooth posture of the first segmented tooth representation is shown;

[0059] Figure 4 The architecture of a trained neural network for determining tooth posture is shown;

[0060] Figure 5 is a flow chart illustrating a method of training a neural network for tooth posture estimation;

[0061] Figure 6 A dental scanning system is shown. DETAILED DESCRIPTION

[0062] In the following description, reference is made to the accompanying drawings, which show by way of illustration how the invention may be practiced.

[0063] Figure 1A virtual 3D representation 100 of a patient's dentition is shown. In this case, the virtual 3D representation includes a lower jaw with multiple tooth representations as shown, but in general, the virtual 3D representation may include an upper jaw or both an upper jaw and a lower jaw. The virtual 3D representation 100 represents the patient's dentition and may be constructed based on scan data collected during a scanning process, such as scanning the inside of the patient's mouth (including teeth and gums) using an intraoral scanner. The captured scan data may be used to construct the virtual 3D representation 100 and present it on a display screen. The scan data or virtual 3D representation 100 may be stored in and accessed from a memory device of a computer system.

[0064] The virtual 3D representation 100 may initially be in the form of a 3D mesh representing geometric information. The virtual 3D representation may also include texture information, such as color. The geometric information represents the lower jaw (as shown), but may also represent only the upper jaw or both the upper jaw and the lower jaw of the patient. Alternatively, the virtual 3D representation 100 may be in the form of a point cloud or voxel representation.

[0065] In order to identify individual teeth within the virtual 3D representation 100, a segmentation process of the virtual 3D representation 100 may be applied. By segmenting the virtual 3D representation 100, at least a first segmented tooth representation 101 and a second segmented tooth representation 105 may be obtained. At least the first and second segmented tooth representations 101, 105 may represent adjacent teeth in the same jaw of a patient's dentition. The segmentation process may be performed in different ways, for example based on the identification of individual patches or groups of patches belonging to the tooth representation. Thus, it allows identifying objects such as individual teeth and / or surrounding gums in the virtual 3D representation 100 representing the patient's dentition, which may include multiple teeth. Thus, the output of the segmentation process may be individual tooth representations, which are typically displayed in the form of a solid dental object, a tooth mesh or a tooth point cloud.

[0066] for Figure 1Each segmented tooth representation in the diagram shows a tooth posture. The tooth posture can also be referred to as a local coordinate system specific to each segmented tooth representation. Each tooth posture consists of an origin where a first axis, a second axis, and a third axis intersect. The first axis 102 of the tooth posture of the first segmented tooth representation 101 is along the mesiodistal direction of the first segmented tooth representation 101, and points in the direction of the second segmented tooth representation 105. The second axis 103 of the tooth posture of the first segmented tooth representation 101 points in the crown direction of the crown of the first segmented tooth representation 101, and is orthogonal to the occlusal surface or occlusal surface of the first segmented tooth representation 101. The third axis 104 of the tooth posture of the first segmented tooth representation 101 can be orthogonal to the first axis 102 and the second axis 103, and can face outward along the vestibular or facial direction. The tooth posture of the first segmented tooth representation 101 defined in this way can also be referred to as the correct tooth posture of the first segmented tooth representation 101. Other tooth postures of the remaining tooth representations are defined in a similar manner.

[0067] Figure 2 The steps of a computer-implemented method 200 for tooth posture estimation according to an embodiment of the present disclosure are shown. At 201, a virtual 3D representation 100 representing a patient's dentition can be obtained. The virtual 3D representation 100 can be obtained by scanning the patient's intraoral situation (e.g., using an intraoral scanner) to obtain scan data. The scan data may include 2D images, color information, fluorescence and / or infrared information. Instead of an intraoral scanner, a desktop scanner or a laboratory scanner may be used to scan a physical model of the teeth. The scan data can then be directly converted into a virtual 3D representation 100. The scan data can also be stored in a memory device of a computer system, for example. Then, by accessing the scan data from the memory device of the computer system, the virtual 3D representation 100 can be obtained. Typically, the virtual 3D representation 100 is displayed on a display screen in the form of a three-dimensional mesh (e.g., a triangular mesh). Alternatively, the virtual 3D representation 100 can be displayed on a display screen in the form of a point cloud, a volume representation, or any other suitable 3D representation.

[0068] In step 202, the virtual 3D representation 100 can be segmented to obtain at least a first segmented tooth representation 101 and a second segmented tooth representation 105. The virtual 3D representation 100 can be segmented so as to determine the patches of the virtual 3D representation belonging to the individual teeth according to the universal numbering system / symbol (UNN) and to determine the patches belonging to the gingiva. The individual tooth representations can then be extracted and used, for example, to create an orthodontic treatment. Once the individual tooth representations are obtained, their movement can be simulated by applying rotation and translation operations to their respective tooth poses. The segmentation of the virtual 3D representation 100 can be performed in a variety of ways. According to an example, the segmentation can include using surface curvature to identify the boundaries of the tooth representation. The minimum principal curvature and the average principal curvature can be used to quantitatively measure surface properties. Then, based on a principal component analysis algorithm (PCA), a cutting plane for separating the gingival and tooth parts can be generated. A curvature threshold (e.g., a constant value) can be selected to distinguish between the tooth boundary area and the rest of the surface.

[0069] In another example, segmenting the virtual 3D representation 100 may include identifying tooth boundaries using a harmonic field. On the virtual 3D representation 100, the harmonic field is a scalar attached to each mesh vertex that satisfies the following condition: ΔΦ=0, where Δ is the Laplace operator, subject to the Dirichlet boundary constraint. The above equation can be solved, for example, using a least squares method to calculate the harmonic field. The segmented tooth representations 101, 105 can then be extracted by selecting the best contour line connecting data points with the same value as the tooth representation boundary.

[0070] In yet another example, segmenting the virtual 3D representation 100 can include using a segmentation machine learning model. In particular, the virtual 3D representation can be converted into a series of 2D virtual images acquired from different perspectives. The segmentation machine learning model can be applied to the series of 2D virtual images. For each 2D virtual image, classification can be performed to distinguish different tooth categories and gums. After classifying each 2D virtual image, back-projection can be performed on the virtual 3D representation 100. This method for segmenting a virtual 3D representation can be advantageous because the segmentation machine learning model utilizes the series of 2D virtual images, which can generally achieve fast and accurate classification.

[0071] The first and second segmented tooth representations 101, 105 represent adjacent teeth in the patient's dentition. The term "adjacent teeth" should be understood as two teeth that are placed adjacent to each other in the upper or lower jaw. Most commonly, adjacent teeth will have consecutive numbers in the universal numbering symbol or system (UNN). In the UNN system, numbers 1-32 are assigned to human teeth. In some cases, adjacent teeth are not marked with consecutive numbers in the UNN symbol, for example if the teeth between the current adjacent teeth have been removed before. The first and second segmented tooth representations 101, 105 are representations of adjacent teeth in the patient's dentition. This is advantageous in the tooth pose estimation method because the initial tooth pose of the first segmented tooth representation 101 can be determined or predicted based on the spatial relationship between the first segmented tooth representation 101 and the second segmented tooth representation 105 representing the adjacent teeth.

[0072] The method 200 also includes, in step 203, determining an initial tooth posture 302 of the first segmented tooth representation 101 using geometric parameters of the first segmented tooth representation 101 and geometric parameters of the second segmented tooth representation 105. Determining the initial tooth posture 302 of the first segmented tooth representation 101 in this way can quickly estimate the correct tooth posture of the first segmented tooth representation 101. For example, the geometric parameters of the first segmented tooth representation 101 and the geometric parameters of the second segmented tooth representation 105 can be used to set a first axis, a second axis, or a third axis of the initial tooth posture of the first segmented tooth representation 101. The initial tooth posture 302 of the first segmented tooth representation 101 may be different from the desired tooth posture, that is, it may be different from the correct tooth posture of the first segmented tooth representation 101 (see Figure 3C and Figure 3D ). However, it may be a good estimation of the correct tooth posture of the first segmented tooth representation 101, which means that the initial tooth posture 302 of the first segmented tooth representation 101 may closely match the correct tooth posture of the first segmented tooth representation 101. The accuracy of the initial tooth posture 302 of the first segmented tooth representation 101 depends on the actual alignment of the first and second tooth representations 101, 105. The more aligned the two adjacent teeth represented by the two segmented tooth representations are, the closer the initial tooth posture is to the correct tooth posture.

[0073] In one embodiment, the geometric parameters of the first segmented tooth representation 101 and the second segmented tooth representation 105 may be the center of mass of the first segmented tooth representation 101 and the center of mass of the second segmented tooth representation 105. The center of mass of the segmented tooth representation is a point in 3D space located at the geometric center of the segmented tooth representation.

[0074] In an alternative embodiment, the geometric parameters of the first segmented tooth representation 101 and the geometric parameters of the second segmented tooth representation 105 may be the center of gravity of the first segmented tooth representation 101 and the center of gravity of the second segmented tooth representation 105, respectively. The center of gravity of a segmented tooth representation is a point in 3D space that is located at the center of gravity of the segmented tooth representation. If the segmented tooth representation represents a molar and / or a premolar, the cusp located at the top of the tooth will cause the center of gravity to be offset compared to the center of mass of the tooth representation, i.e., located closer to the top of the tooth or the crown portion of the tooth. The advantage of using the center of mass over the center of gravity may be reflected in that the center of mass is more robust to different tooth morphologies.

[0075] In a preferred additional step 204, a standardized tooth representation 301 of the first segmented tooth representation 101 (eg, Figure 3B The transformation may include rotating the first segmented tooth representation 101 according to the initial local coordinate system or initial tooth pose 302, and translating the first segmented tooth representation 101 according to the negative value of the coordinates of the origin of the initial tooth pose 302 of the first segmented tooth representation 101 (e.g., its center of mass). The result is reflected in the global coordinate system placement of the first segmented tooth representation 101 around the origin with coordinates (0, 0, 0) (e.g., Figure 3B shown).

[0076] also, Figure 2 205 in the figure shows a standardized tooth representation 301 (eg, Figure 3B The first segmented tooth representation 101 may be a standardized tooth representation 301 of a first segmented tooth representation 101, and the trained neural network 400 may be a deep learning network that can directly process three-dimensional (3D) point clouds and is therefore compatible with the input format. If the standardized tooth representation 301 of the first segmented tooth representation 101 is in the form of a 3D mesh, it may first be converted into a point cloud representation by eliminating the edges of the 3D mesh, leaving vertices to represent the standardized tooth representation 301 of the first segmented tooth representation 101.

[0077] The method 200 shows in step 206 the correct tooth posture 413 (eg, Figure 4 The output of the trained neural network 400 is obtained in the form of . In particular, the output of the trained neural network 400 provides translation and rotation parameters, which, when applied to the first segmented tooth representation 101, provide the correct tooth posture 413 of the first segmented tooth representation 101.

[0078] The method 200 according to an embodiment may optionally include a display step 207, wherein the virtual 3D representation 100 is displayed, for example on a display screen, with a correct tooth pose 413 (eg, Figure 4 Additionally or alternatively, the display may include displaying only the first segmented tooth representation 101 with the correct tooth posture 413. In one embodiment, the method may include displaying the first segmented tooth representation 101 with a second axis of the correct tooth posture 413 of the first segmented tooth representation 101. The second axis of the correct tooth posture 413 of the first segmented tooth representation 101 may be along the crown direction of the first segmented tooth representation 101 and may be orthogonal to the occlusal plane or occlusal surface of the first segmented tooth representation 101.

[0079] Some patients may suffer from a malocclusion in which the relationship of the maxillary teeth to the mandibular teeth is not correct. For example, the teeth represented by the first segmented tooth representation 101 may be misaligned with their adjacent teeth and cause a malocclusion. Such misalignment may be visualized by displaying a virtual 3D representation 100 including the first segmented tooth representation 101 and displaying a correct tooth pose 413 of the first segmented tooth representation 101. The difference between the correct tooth pose 413 of the first segmented tooth representation 101 and the ideal tooth pose of the first segmented tooth representation 101 may then be automatically quantified, wherein the ideal tooth pose represents the teeth as if they were moved into a position to correct the malocclusion. Such a difference may be reflected in the translation and rotation parameters required to overlap the correct tooth pose 413 of the first segmented tooth representation 101 with the ideal tooth pose of the first segmented tooth representation 101. The ideal tooth pose of the segmented tooth representation may be defined as the tooth pose of the segmented tooth representation, wherein the segmented tooth representation represents teeth that are substantially aligned with the ideal dental arch line, for example after orthodontic treatment is applied to move the teeth into a position aligned with the ideal dental arch line.

[0080] In order to correct the malocclusion, orthodontic treatment may be required. The result of the orthodontic treatment can be simulated by applying rotation and translation parameters to the first segmented tooth representation 101, wherein the rotation and translation parameters are determined as the difference between the correct tooth posture 413 of the first segmented tooth representation 101 and the ideal tooth posture of the first segmented tooth representation 101. Such simulation of the result of orthodontic treatment is beneficial because it can provide dental professionals with information about the final tooth arrangement after the orthodontic treatment is applied. For example, in some cases, applying the rotation and translation parameters to the correct tooth posture 413 of the first segmented tooth representation 101 results in the first segmented tooth representation 101 being aligned and placed in the virtual 3D representation without colliding with adjacent tooth representations. In other cases, applying the rotation and translation parameters to the correct tooth posture 413 of the first segmented tooth representation 101 results in the first segmented tooth representation 101 being aligned and placed in the virtual 3D representation 100, where collisions with adjacent tooth representations may occur. It may then be necessary to remove the teeth represented by one of the adjacent tooth representations in order to achieve a satisfactory placement of the first segmented tooth representation 101 in the virtual 3D representation 100. It is also possible to determine the point on the tooth represented by the first segmented tooth representation 101 at which a force should be applied to correct the malocclusion, as well as the direction of the force.

[0081] The tooth pose may be used to align at least two tooth representations of the same tooth scanned at different times.Aligning the at least two tooth representations of the same tooth may include applying translation and rotation transformations so that the tooth poses of the at least two tooth representations of the same tooth coincide.

[0082] Figure 3A An initial tooth pose 302 of the first segmented tooth representation 101 and a global coordinate system 300 are shown, wherein the origin of the global coordinate system 300 is located at a point O with coordinates (0, 0, 0). By aligning the initial tooth pose 302 of the first segmented tooth representation 101 with the global coordinate system 300, a standardized tooth representation 301 of the first segmented tooth representation 101 can be obtained. Figure 3B A normalized tooth representation 301 of the first segmented tooth representation 101 is shown. After normalization, the origin of the initial tooth pose 302 of the first segmented tooth representation 101 coincides with the point O with coordinates (0, 0, 0).

[0083] Figure 3C and Figure 3D A portion of a virtual 3D representation 100 is shown, wherein the teeth represented by the first segmented teeth representation 101 and the second segmented teeth representation 105 are not aligned with each other. Figure 3C FIG. 3 shows an initial tooth pose 302 of the first segmented tooth representation 101 determined according to the method discussed. It can be seen that the initial tooth pose 302 is Figure 3DThe correct tooth posture of the first segmented tooth representation 101 shown in FIG. 1 is different. More specifically, the first axis ( Figure 3C )and Figure 3D 105. However, the initial tooth pose 302 may be a good estimate of the correct tooth pose of the first segmented tooth representation 101. The accuracy of the initial tooth pose 302 of the first segmented tooth representation 101 depends on the actual alignment of the first and second tooth representations 101, 105.

[0084] Figure 4 An example of a neural network architecture suitable for the method for tooth pose estimation according to an embodiment of the present disclosure is shown. The trained neural network 400 may be a PointNet neural network capable of processing the standardized tooth representation 301 of the first segmented tooth representation 101 in the form of a point cloud. Figure 4 The model shown in can take an unordered point set as input. A general model of this network architecture was developed at Stanford University (“PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation”, Charles R. Qi, Hao Su, Kaichun Mo, Leonidas J. Guibas). In addition to the PointNet neural network, the trained neural network 400 can also be PointNet++ or a neural network similar to PointNet.

[0085] The standardized tooth representation 301 of the first segmented tooth representation 101 is used as an input 401 of the trained neural network 400. The input 401 can be in n x 3 format, where n is the number of 3D points representing the standardized tooth representation 301 of the first segmented tooth representation 101, and the dimension 3 represents the three coordinates representing each point of the standardized tooth representation 301. The input 401 is processed by an input transformation block 402, which can be a 3 x 3 rotation transformation. The input transformation block 402 has the effect of achieving invariance under transformation. Block 403 can be an input to a transformation with an n x 3 format. Alternatively, the input 401 can be in n x 6 format, where n is the number of 3D points representing the standardized tooth representation 301 of the first segmented tooth representation 101, and the dimension 6 represents the three coordinates representing each point of the standardized tooth representation 101 and the three coordinates representing the normal of each point.

[0086] Block 404 represents a first multilayer perceptron with shared weights, which can process all 3D points independently for feature extraction. Its function is to achieve invariance under the sequential arrangement of 3D points. The architecture of the first multilayer perceptron 404 (MLP) can include a 3-dimensional input layer, a hidden layer of 64 neurons, and an output layer can include 64 features. The size of block 405 can be n x 64.

[0087] The feature transformation 406 block may include a transformation matrix of size 64x 64, which may be a regularization matrix, such as an L2 type. The size of block 407 may be nx 64. Block 408 represents a second multilayer perceptron with shared weights. The architecture of the second multilayer perceptron 408 (MLP) may include an input layer of 64 features, two hidden layers of 64 neurons and 128 neurons, and the output layer of the second multilayer perceptron 408 may include 1024 features. The size of block 409 may be nx1024. In the maximum pooling block 410, the feature vectors of the 3D points are aggregated to create a 1024-dimensional global feature representation 411 of the standardized tooth representation 301. Block 412 represents a third multilayer perceptron. The architecture of the third multilayer perceptron 412 (MLP) may include an input layer of 1024 features, two hidden layers of 512 neurons and 256 neurons, and the output layer of the third multilayer perceptron 412 (also referred to as the output of the trained neural network 400) may include a classification score.

[0088] The output of the trained neural network 400 may include a three-dimensional rotation vector in an axis-angle representation. The three-dimensional rotation vector may describe how the first segmented tooth representation 101 should be rotated in order to place it in its correct tooth pose 413. It may be described by two quantities, a unit vector indicating the direction of the rotation axis and an angle having a value between 0 and 2π radians, describing the magnitude of the rotation about the rotation axis.

[0089] The output of the trained neural network 400 may also include a three-dimensional translation vector including the translation amount of the first segmented teeth representation 101. The three-dimensional translation vector may describe how the first segmented teeth representation 101 should be translated in order to place it in the correct teeth pose 413.

[0090] The trained neural network 400 can be used in Figure 5 The computer-implemented method 500 for training a neural network for tooth pose estimation is shown. An input training data set including training tooth representations and reference truth tooth poses of the training tooth representations can be used during the training process. The input training data set is accessed in 501. The reference truth tooth poses of the training tooth representations can be tooth poses manually assigned to the training tooth representations. The input training data set can also include hundreds or thousands of training tooth representations or segmented tooth representations.

[0091] The training tooth representation may preferably be normalized to place it at the origin of the global coordinate system. Afterwards, the vertices of the training tooth representation may be extracted and fed into a neural network for tooth pose estimation in 502 for training purposes. In 503, the output of the neural network for tooth pose estimation may include the predicted tooth pose of the training tooth representation. The predicted tooth pose of the training tooth representation may then be compared with the reference truth tooth pose of the training tooth representation in 504. Then, the difference between the predicted tooth pose of the training tooth representation and the reference truth tooth pose of the training tooth representation may be minimized in 505, for example by using a stochastic gradient descent algorithm (SGD). The difference may also be referred to as a loss, a loss function, or an error. In particular, the loss may be a function of the geodesic distance between the predicted tooth pose of the training tooth representation and the reference truth tooth pose of the training tooth representation and the Euclidean distance between the origin of the predicted tooth pose of the training tooth representation and the origin of the reference truth tooth pose of the training tooth representation. When the loss reaches a stopping criterion, the training of the neural network for tooth pose estimation is stopped in 506. For example, the stopping criterion may be a threshold value of the loss.

[0092] The training process may optionally include an augmentation step, where a small amount of rotation or jitter is added to the training tooth representations. The effect of the augmentation step is that the neural network for tooth pose estimation "sees" slightly altered inputs, resulting in a more robust behavior of the neural network for tooth pose estimation, able to generalize better to new unseen data.

[0093] The geodesic distance between the predicted dental pose of the training dental representation and the reference truth dental pose of the training dental representation represents the amount of radians that the predicted dental pose of the training dental representation needs to be rotated to be equal to the reference truth dental pose of the training dental representation. The Euclidean distance between the origin of the predicted dental pose of the training dental representation and the origin of the reference truth dental pose of the training dental representation represents the distance that the predicted dental pose of the training dental representation needs to be translated so that the origin of the predicted dental pose of the training dental representation overlaps with the origin of the reference truth dental pose of the training dental representation.

[0094] In one embodiment, the loss can be the sum of the geodesic distance between the predicted tooth pose of the training tooth representation and the reference truth tooth pose of the training tooth representation, and the squared value of the Euclidean distance between the origin of the predicted tooth pose of the training tooth representation and the origin of the reference truth tooth pose of the training tooth representation. The sum can then be minimized using a stochastic gradient descent algorithm.

[0095] The training of the neural network for tooth pose estimation is an iterative process and can be repeated until the neural network for tooth pose estimation can no longer improve its output. This means that after a number of iterations, the output of the neural network for tooth pose estimation continues to be equal to the reference truth data. The number of iterations required for the training of the neural network for tooth pose estimation can be between 200,000 and 400,000 iterations, preferably 300,000 iterations.

[0096] Figure 6 A dental scanning system 600 is shown, which may include a computer 610 capable of executing method 200 or method 500. The computer may include a wired or wireless interface to a server 615, a cloud server 620, and an intraoral scanner 625. The intraoral scanner 625 may be capable of recording scan data including geometric and textural data of a patient's dentition. The intraoral scanner 625 may be equipped with various modules, such as a fluorescent module or an infrared module, so as to be able to capture information related to diagnosing a dental condition (e.g., caries, tooth cracks, gingivitis, or dental plaque).

[0097] The dental scanning system 600 may include a data processing device configured to perform the method according to one or more embodiments of the present disclosure. The data processing device may be part of the computer 610 , the server 615 , or the cloud server 620 .

[0098] A non-transitory computer-readable storage medium may be included in the dental scanning system 600. The non-transitory computer-readable medium may carry instructions, which, when executed by a computer, cause the computer to perform a method according to one or more embodiments of the present disclosure.

[0099] A non-transitory computer-readable medium may include instructions that, when executed by a computer, cause the computer to obtain a virtual 3D representation 100 representing a patient's dentition; segment the virtual 3D representation 100 to obtain at least a first segmented tooth representation 101 and a second segmented tooth representation 105, wherein at least the first segmented tooth representation 101 and the second segmented tooth representation 105 represent adjacent teeth in the patient's dentition; determine an initial tooth posture 302 of the first segmented tooth representation 101 using geometric parameters of the first segmented tooth representation 101 and geometric parameters of the second segmented tooth representation 105; obtain a standardized tooth representation 301 of the first segmented tooth representation 101 by transforming the first segmented tooth representation 101 using the initial tooth posture 302 of the first segmented tooth representation 101; input the standardized tooth representation 301 of the first segmented tooth representation 101 into a trained neural network 400 and generate an output from the trained neural network 400, wherein the output includes a correct tooth posture 413 of the first segmented tooth representation 101.

[0100] The computer program product may be embodied in a non-transitory computer-readable storage medium. The computer program product may include instructions that, when executed by a computer, cause the computer to perform a method according to any embodiment presented herein.

[0101] The computer program product may include instructions that, when executed by a computer, cause the computer to obtain a virtual 3D representation 100 representing a patient's dentition, segment the virtual 3D representation 100 to obtain at least a first segmented tooth representation 101 and a second segmented tooth representation 105, wherein at least the first segmented tooth representation 101 and the second segmented tooth representation 105 represent adjacent teeth in the patient's dentition, determine an initial tooth pose 302 of the first segmented tooth representation 101 using geometric parameters of the first segmented tooth representation 101 and geometric parameters of the second segmented tooth representation 105, obtain a standardized tooth representation 301 of the first segmented tooth representation 101 by transforming the first segmented tooth representation 101 using the initial tooth pose 302 of the first segmented tooth representation 101, input the standardized tooth representation 301 of the first segmented tooth representation 101 into a trained neural network 400 and generate an output from the trained neural network 400, wherein the output includes a correct tooth pose 413 of the first segmented tooth representation 101.

[0102] Although some embodiments have been described and shown in detail, the present disclosure is not limited to these details, but can also be embodied in other ways within the scope of the subject matter defined in the appended claims. In particular, it should be understood that other embodiments can be utilized and structural and functional modifications can be made without departing from the scope of the present invention.

[0103] It should be understood that references throughout the specification to "one embodiment" or "embodiment" or "aspect" or features that "may" be included mean that the particular features, structures, or characteristics described in connection with the embodiment are included in at least one embodiment of the present disclosure. In addition, the particular features, structures, or characteristics may be combined in one or more embodiments of the present disclosure as desired. The previous description is provided to enable those skilled in the art to practice the various aspects described herein. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects.

Claims

1. A computer-implemented method for tooth posture estimation, the method comprising: - obtaining a virtual 3D representation (100) representing the patient's dentition; - segmenting the virtual 3D representation (100) to obtain at least a first segmented tooth representation (101) and a second segmented tooth representation (105), wherein the at least first segmented tooth representation and the second segmented tooth representation (101, 105) represent adjacent teeth in the dentition of the patient; - using the geometric parameters of the first segmented tooth representation (101) and the geometric parameters of the second segmented tooth representation (105) to determine the initial tooth posture (302) of the first segmented tooth representation (101), wherein the geometric parameters of the first segmented tooth representation (101) are the center of mass of the first segmented tooth representation (101), and the geometric parameters of the second segmented tooth representation (105) are the center of mass of the second segmented tooth representation (105), wherein each tooth posture is composed of an origin where a first axis (102), a second axis (103) and a third axis (104) intersect, wherein the first segmented tooth representation (101) is used ... first segmented tooth representation (101) is used to determine the initial tooth posture (302) of the first segmented tooth representation (101), wherein the first segmented tooth representation (101) is used to determine the initial tooth posture (302) of the first segmented tooth representation (101), and the geometric parameters of the second segmented tooth representation (105) are the center of mass of the second segmented tooth representation (105), wherein the first segmented tooth representation (101) is used to determine the initial tooth posture (302) of the first segmented tooth representation (101), wherein the first segmented tooth representation (101) is used to determine the initial tooth posture (302) of the Determining an initial tooth pose (302) of the first segmented tooth representation (101) based on geometric parameters of the segmented tooth representation (101) and geometric parameters of the second segmented tooth representation (105) includes determining the first axis, the second axis and the third axis (102, 103, 104) of the initial tooth pose (302) of the first segmented tooth representation (101), wherein the first axis (102) of the initial tooth pose (302) of the first segmented tooth representation (101) is calculated as a difference between a center of mass of the first segmented tooth representation (101) and a center of mass of the second segmented tooth representation (105); - transforming the first segmented tooth representation (101) by using the initial tooth pose (302) of the first segmented tooth representation (101) to obtain a standardized tooth representation (301) of the first segmented tooth representation (101); - inputting the standardized tooth representation (301) of the first segmented tooth representation (101) into a trained neural network (400); - generating an output from the trained neural network (400), wherein the output comprises a correct tooth pose (413) of the first segmented tooth representation (101).

2. The method according to claim 1, wherein: The second axis (103) of the initial tooth posture (302) of the first segmented tooth representation (101) is orthogonal to the occlusal plane of the first segmented tooth representation (101).

3. A method according to any one of the preceding claims, wherein: The third axis (104) of the initial tooth posture (302) of the first segmented tooth representation (101) is obtained as the cross product between the first axis (102) of the initial tooth posture (302) of the first segmented tooth representation (101) and the second axis (103) of the initial tooth posture (302) of the first segmented tooth representation (101).

4. The method according to any preceding claim, further comprising adjusting the first segmented tooth representation (101) by reducing the number of vertices representing the first segmented tooth representation (101).

5. A method according to any preceding claim, wherein: The trained neural network (400) is a PointNet neural network.

6. A method according to any preceding claim, wherein: The output of the trained neural network (400) comprises a three-dimensional rotation vector in axis-angle representation.

7. A method according to any preceding claim, wherein: The output of the trained neural network (400) comprises a three-dimensional translation vector comprising an amount of translation of the first segmented tooth representation (101).

8. The method according to any preceding claim, further comprising converting the first segmented tooth representation (101) into a point cloud representation.

9. A method according to any preceding claim, wherein: The output of the trained neural network (400) provides translation and rotation parameters which, when applied to the first segmented teeth representation (101), provide a correct teeth pose (413) of the first segmented teeth representation (101).

10. The method according to any preceding claim, further comprising displaying a virtual 3D representation (100) of a correct tooth pose (413) with the first segmented tooth representation (101) on a display screen.

11. A method according to any preceding claim, wherein: Transforming the first segmented tooth representation (101) using the initial tooth posture (302) of the first segmented tooth representation (101) includes rotating the first segmented tooth representation (101) according to the initial tooth posture (302) and translating the first segmented tooth representation (101) according to the negative value of the coordinates of the origin of the initial tooth posture (302) of the first segmented tooth representation (101).

12. A data processing device comprising means for executing the method according to any one of claims 1 to 11.

13. A computer program product comprising instructions which, when the program is executed by a computer (610), cause the computer (610) to perform the method according to any one of claims 1 to 11.

14. A non-transitory computer readable medium comprising instructions which, when executed by a computer (610), cause the computer (610) to perform the method according to any one of claims 1-11.

15. A dental scanning system (600), comprising a computer (610), a server (615), a cloud server (620), an intraoral scanner (625), and a data processing device configured to perform the method according to any one of claims 1-11.

Images