Method for segmenting a computed tomography image of a tooth

Abstract

An aspect of the present application provides a computer-implemented method for segmenting a computed tomography image of a tooth, comprising: obtaining a first three-dimensional digital model representing a first dental arch of erupted tooth crowns and a sequence of two-dimensional tomographic images of the first dental arch; using a local image classification model, selecting a starting segmentation two-dimensional tomographic image for each erupted tooth from the sequence of two-dimensional tomographic images of the first dental arch based on the sequence of two-dimensional tomographic images of the first dental arch, the local image classification model being a trained deep neural network for classifying a local two-dimensional tomographic image into a class comprising a tooth crown and a tooth root; obtaining position and extent information for each erupted tooth based on the first three-dimensional digital model; and for each erupted tooth, using the position and extent information, using a local image segmentation model, segmenting a local image of the erupted tooth in a direction of the tooth crown and the tooth root from the corresponding starting segmentation two-dimensional tomographic image to obtain a sequence of binary mask images.

Description

TECHNICAL FIELD

[0001] The present application relates generally to a method for segmenting a computed tomography image of a tooth. BACKGROUND

[0002] With the continuous development of computer science, dental professionals increasingly rely on computer technology to improve the efficiency of dental diagnosis and treatment. In computer-aided dental diagnosis and treatment, three-dimensional digital models of teeth are often used.

[0003] At present, a three-dimensional digital model of a tooth crown with high accuracy can be obtained by intraoral scanning or by scanning a dental impression or a physical model, but this method cannot obtain information about the tooth root.

[0004] For some diagnosis and treatment projects that require information about the tooth root (for example, orthodontic treatment), a two-dimensional tomographic image sequence of the entire tooth (including the tooth crown and the tooth root) can be obtained using computed tomography (CT) technology, and a three-dimensional digital model of the entire tooth can be generated based on this. Since the two-dimensional tomographic images obtained using computed tomography technology not only include teeth but also include jaw bones, it is necessary to segment them before establishing a three-dimensional digital model of the entire tooth to remove the jaw bone part and retain the tooth part.

[0005] In view of the above, it is necessary to provide a method for segmenting a computed tomography image of a tooth. SUMMARY

[0006] One aspect of the present application provides a computer-implemented method for segmenting a computed tomography image of a tooth, which comprises: obtaining a first three-dimensional digital model representing a tooth crown of a first dental arch and a two-dimensional tomographic image sequence of the first dental arch; using a local image classification model, selecting a starting segmented two-dimensional tomographic image for each erupted tooth from the two-dimensional tomographic image sequence of the first dental arch based on the two-dimensional tomographic image sequence of the first dental arch, the local image classification model being a trained deep neural network for classifying a local two-dimensional tomographic image into a category including a tooth crown and a tooth root; obtaining position and range information of each erupted tooth based on the first three-dimensional digital model; and for each erupted tooth, using the position and range information, using a local image segmentation model to segment a local image of the erupted tooth from the corresponding starting segmented two-dimensional tomographic image in the direction of the tooth crown and the tooth root to obtain a binary mask image sequence thereof.

[0007] In some embodiments, the first three-dimensional digital model is obtained by one of the following methods: intraoral scanning or scanning a dental impression or a physical model.

[0008] In some embodiments, the two-dimensional tomographic images of the first dentition are obtained by cone-beam computed tomography.

[0009] In some embodiments, the computer-implemented method of segmenting a computed tomography image of a tooth further comprises: for each erupted tooth, using the location and range information, using the local image segmentation model, segmenting a local image of the erupted tooth in the initially segmented two-dimensional tomographic image of the erupted tooth to obtain a mask binary image of the erupted tooth corresponding to the initially segmented two-dimensional tomographic image; and for each erupted tooth, using the mask binary image corresponding to the previous two-dimensional tomographic image as range information, and using the range information, using the local image segmentation model to segment a local image of the erupted tooth in the next two-dimensional tomographic image.

[0010] In some embodiments, the computer-implemented method of segmenting a computed tomography image of a tooth further comprises: using a global image segmentation model and the local image classification model, extracting a sequence of mask images of the crown portion of the erupted teeth of the first dentition based on the sequence of two-dimensional tomographic images of the first dentition, wherein the global image segmentation model is a trained deep neural network for segmenting a two-dimensional tomographic image to extract a global tooth mask image; and projecting the first three-dimensional digital model and the sequence of mask images of the crown portion of the erupted teeth of the first dentition on a first plane and registering the projections of the two to obtain the location and range information.

[0011] In some embodiments, for each erupted tooth, the local image of the erupted tooth in the initially segmented two-dimensional tomographic image is located in the middle section of the tooth.

[0012] In some embodiments, for each erupted tooth, the local image of the erupted tooth in the initially segmented two-dimensional tomographic image is located in the neck of the tooth.

[0013] In some embodiments, the computer-implemented method for segmenting a dental computed tomography image further comprises: extracting a global dental mask image sequence of the first dental arch based on the sequence of two-dimensional tomographic images of the first dental arch using a global image segmentation model, wherein the global image segmentation model is a trained deep neural network for segmenting a two-dimensional tomographic image to extract a global dental mask image; removing the mask of the root portion of all erupted teeth in the global dental mask image sequence of the first dental arch using the local image classification model to obtain a second mask image sequence; generating a mask intensity curve based on the second mask image sequence and determining the range of the sequence of two-dimensional tomographic images of the first dental arch where the impacted teeth are located based on the intensity curve; for each impacted tooth, determining a starting segmented two-dimensional tomographic image within the range where the impacted tooth is located; and for each impacted tooth, performing a final segmentation of the local image of the impacted tooth in the direction of the crown and root from the corresponding starting segmented two-dimensional tomographic image using the local image segmentation model to obtain a sequence of binary mask images thereof.

[0014] In some embodiments, the computer-implemented method for segmenting a dental computed tomography image further comprises: for each impacted tooth, performing a pre-segmentation of the local image thereof within the range of the two-dimensional tomographic images using the local image segmentation model, and determining the starting segmented two-dimensional tomographic image thereof based on the area of the mask obtained by the segmentation.

[0015] In some embodiments, for each impacted tooth, the two-dimensional tomographic image corresponding to the largest area mask obtained by the pre-segmentation is taken as the starting segmented two-dimensional tomographic image.

[0016] In some embodiments, the computer-implemented method for segmenting a dental computed tomography image further comprises: if the distance between the center points of the largest area masks of two impacted teeth is less than a first threshold value, then the final segmentation of the impacted tooth corresponding to the smaller mask is not performed using the local image segmentation model.

[0017] In some embodiments, the computer-implemented method for segmenting a dental computed tomography image further comprises: if the distance between the center point of the largest area mask of an impacted tooth and the center point of the mask of the starting two-dimensional tomographic image corresponding to the closest erupted tooth is greater than a second distance threshold value, then the final segmentation of the impacted tooth is not performed using the local image segmentation model. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and other features of the present application will be further understood and appreciated, along with one or more various advantages thereof, upon review of the following detailed description in conjunction with the accompanying drawings, in which:

[0019] Figure 1 a schematic flowchart of a computer-implemented method of fusing three-dimensional digital models of teeth according to an embodiment of the present application;

[0020] Figure 2A a two-dimensional tomographic image classified as a mandible according to an example of the present application is shown;

[0021] Figure 2B a two-dimensional tomographic image classified as a maxilla according to an example of the present application is shown;

[0022] Figure 2C a two-dimensional tomographic image classified as a tooth according to an example of the present application is shown;

[0023] Figure 2D a two-dimensional tomographic image classified as an occlusal tooth according to an example of the present application is shown;

[0024] Figure 2E a two-dimensional tomographic image classified as an open-mouth tooth according to an example of the present application is shown; and

[0025] Figure 3 a global tooth mask extracted using a global image segmentation model according to an example of the present application is shown. DETAILED DESCRIPTION

[0026] The following detailed description references the drawings, wherein like numerals indicate like elements throughout the several views. The illustrative embodiments described herein are not meant to be limiting but merely exemplary. Many variations, as well as many alternatives, can be contemplated depending on the specific application. In addition, many aspects of the disclosure have been presented for purposes of illustration and description, and are not intended to be exhaustive or to limit the disclosure to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. It is intended that the scope of the disclosure be limited not with the specifically described embodiments herein, but rather by the appended claims.

[0027] One aspect of the present application provides a method of segmenting a computed tomography image of a tooth for extracting a tooth portion in a sequence of two-dimensional tomographic images.

[0028] In one embodiment, a sequence of two-dimensional tomographic images of teeth can be obtained using cone-beam computed tomography (CBCT). It is understood that other computed tomography techniques can also be used to obtain the same sequence. The two-dimensional tomographic images of the teeth are grayscale images.

[0029] Please refer to Figure 1 This is a schematic flowchart of a method 100 for segmenting a computed tomography image of teeth according to one embodiment of this application.

[0030] For ease of explanation, the following describes the segmentation method 100 of a computed tomography image of a tooth, using a single jaw (i.e., the maxilla or mandible) as an example.

[0031] In 101, a first three-dimensional digital model representing the crown of the first jaw and a sequence of two-dimensional tomographic images of the first jaw are obtained.

[0032] In one embodiment, a first three-dimensional digital model can be obtained by intraoral scanning, or by scanning a dental impression or physical model (e.g., a plaster model of a tooth), wherein the first three-dimensional digital model may be segmented, i.e., each crown is independent. The technical means of obtaining the first three-dimensional digital model are well known in the industry and will not be described in detail here.

[0033] Typically, computed tomography (CT) scans yield a sequence of two-dimensional tomographic images of the entire upper and lower jaw. To obtain a sequence of two-dimensional tomographic images of the upper or lower jaw, a global image classification model can be used to divide it into two parts: the upper jaw and the lower jaw. This global image classification model is based on a deep convolutional neural network, such as VGG, Inception, or ResNet. In one embodiment, the global image classification model can classify two-dimensional tomographic images into five categories: mandible, teeth, occlusal teeth, open bite teeth, and maxilla, assigning a category to each two-dimensional tomographic image. Specifically, two-dimensional tomographic images containing only the mandible or maxilla will be classified as mandible or maxilla; two-dimensional tomographic images containing both the mandible and teeth will be classified as teeth; two-dimensional tomographic images containing both maxillary and mandibular teeth will be classified as occlusal teeth; and two-dimensional tomographic images containing only a single molar (e.g., maxillary or mandibular) will be classified as open bite teeth (this type of two-dimensional tomographic image sequence is typically included in scans of patients in an open bite state).

[0034] Please refer to Figure 2A This example shows a two-dimensional tomographic image classified as the mandible.

[0035] Please refer to Figure 2BThis example shows a two-dimensional tomographic image classified as the maxilla.

[0036] Please refer to Figure 2C This example shows a two-dimensional tomographic image of a tooth.

[0037] Please refer to Figure 2D This example shows a two-dimensional tomographic image of teeth classified as occlusal teeth.

[0038] Please refer to Figure 2E This example shows a two-dimensional tomographic image of a tooth classified as an open bite.

[0039] In step 103, a mask for the crown portion is extracted based on a two-dimensional tomographic image sequence of the first jaw.

[0040] In one embodiment, a global image segmentation model can be used to extract tooth masks based on a sequence of two-dimensional tomographic images of the first jaw, resulting in a global tooth mask image sequence for the first jaw. Each global tooth mask image contains masks of all teeth in the corresponding two-dimensional tomographic image, and these tooth masks are treated as a whole, i.e., they are not segmented into individual teeth. The mask images are binary images. The global tooth segmentation model can be a trained deep convolutional neural network, such as an FCN network, a UNet network, or a VNet network. In one embodiment, when segmenting a two-dimensional tomographic image, the global tooth segmentation model classifies each pixel of the two-dimensional tomographic image to extract pixels belonging to teeth.

[0041] Please refer to Figure 3 This example demonstrates a global tooth mask extracted using a global image segmentation model.

[0042] Next, a local tooth classification model can be used to classify the local tooth images corresponding to each connected component of the tooth mask in the binary image sequence of the first jaw, removing the root mask and retaining the crown mask to obtain the crown mask image sequence of the first jaw. The local tooth classification model can be a trained deep convolutional neural network, such as a Vgg network, Inception network, or ResNet network. In one embodiment, the local tooth classification model can classify the local tooth image corresponding to each connected component of the tooth mask into three categories: crown, root, and background.

[0043] Since there is generally no adhesion between adjacent teeth in the root portion of the tooth mask, the root portion of each tooth can be removed from the global tooth mask image sequence of the first jaw based on the results of the local classification.

[0044] In step 105, the first three-dimensional digital model and the sequence of crown mask images of the first jaw are projected onto the first plane and registered.

[0045] In one embodiment, the first plane may be parallel to a two-dimensional tomographic image of the first jaw. In another embodiment, the first plane may be parallel to the xy plane of the coordinate system of the first three-dimensional digital model.

[0046] Hereinafter, the two-dimensional projection image of the first three-dimensional digital model on the first plane is denoted as T1, and the two-dimensional projection image of the crown masking image sequence of the first jaw on the first plane is denoted as I1.

[0047] In one embodiment, T1 and I1 can be registered by translating along the first plane and rotating about a first axis perpendicular to the first plane.

[0048] First, rotate T1 by R degrees to obtain TR1. Use TR1 as the template image and I1 as the source image for template matching. Calculate the matching coefficient matrix using the normalized sum of variance method. The pixel position with the smallest value in the matching coefficient matrix is ​​the optimal matching center point. R can be selected from some discrete values ​​within a set range, for example, from -10 degrees to 10 degrees, with a step size of 2 degrees for 11 rotation angles.

[0049] Next, for each of the obtained series of matching coefficient matrices, a matching coefficient value (i.e., the minimum value of the corresponding coefficient matrix) is calculated. The smallest matching coefficient value is selected, and the rotation angle corresponding to its matching coefficient matrix is ​​the optimal matching rotation angle, denoted as R. tm1 The offset of the template image is the coordinate values ​​min_loc_x1 and min_loc_y1 corresponding to the minimum matching coefficient value. R tm1 min_loc_x1 and min_loc_y1 are the optimal matching transformation parameters, and the registration is now complete.

[0050] At the outset of this application, it is understood that, in addition to the above registration methods, any other applicable registration methods may be used, which will not be listed here.

[0051] In step 107, the position and range information of the erupting tooth are determined based on the registration result and the first three-dimensional digital model.

[0052] In one embodiment, the center point of each crown in the first three-dimensional digital model can be calculated (for example, the center point can be obtained by calculating the average coordinates of the vertices of each crown), and then these center points can be projected onto the first plane to obtain a projected image C1.

[0053] Rotate C1 according to the transformation parameters obtained from the registration. tm1 By offsetting along the x-axis by min_loc_x1 and along the y-axis by min_loc_y1, the transformed image C is obtained. t1 This information is used as the positional information of each erupting tooth in the first jaw.

[0054] The projections of each tooth crown in the first three-dimensional digital model onto the first plane are rotated by R according to the transformation parameters obtained from the registration. tm1 The transformed image is obtained by offsetting along the x-axis by min_loc_x1 and along the y-axis by min_loc_y1, and is used as the range information of each erupting tooth in the first jaw.

[0055] In step 109, based on the position and extent information of the teeth, the two-dimensional tomographic image sequence of the first jaw is re-segmented using a local tooth classification model and a local tooth segmentation model to obtain the masking sequence of the erupting teeth.

[0056] In one embodiment, a local image of the tooth is cropped from the two-dimensional tomographic image of the first jaw, centered on the center point of the erupting tooth, within a preset range (e.g., a square with a side length of 15 mm; this preset range can be adjusted according to different pixel spacing and physical dimensions). Then, these local tooth images are classified as crown, root, or background using the local tooth classification model, thus obtaining the tomographic image positions where the root and crown are adjacent, i.e., the tomographic image position of the tooth neck.

[0057] In one embodiment, for each erupting tooth, segmentation can begin from the tomographic image containing the tooth neck and proceed in both directions towards the crown and root. That is, the tomographic image containing the tooth neck is segmented first to extract the mask of the erupting tooth. Then, the current mask of the erupting tooth is used as a reference to segment the next tomographic image. Following the guidance of this application, it can be understood that segmentation can also begin from a tomographic image near the tooth neck (i.e., a tomographic image of the mid-tooth segment) (since the shape of the tooth mask does not change much in adjacent tomographic images, it will not reduce the accuracy of segmentation), and proceed in both directions towards the crown and root. For each erupting tooth, a tomographic image is selected as the starting point for segmentation; this selected tomographic image is referred to below as the initial segmentation tomographic image of the erupting tooth.

[0058] In one embodiment, for each erupting tooth, based on the location and extent information (i.e., the projection of the crown of the erupting tooth onto the first plane in the first 3D digital model after transformation by the transformation parameters), a local tooth segmentation model can be used to segment the local tomographic image of the tooth in the initial tomographic image to extract its tooth mask in the current tomographic image (i.e., the initial tomographic image). The local tooth segmentation model can be a trained deep convolutional neural network, such as an FCN network, a UNet network, or a VNet network, used to extract the tooth mask from the local tomographic image of the tooth. When segmenting a local tomographic image, the local tooth segmentation model classifies each pixel of the local tomographic image to extract pixels belonging to the tooth.

[0059] In one embodiment, for segmentation of the next tomographic image of the same erupting tooth, the local tomographic image of the erupting tooth in the next tomographic image (obtained by the local tooth classification model) can be segmented based on the tooth mask of the erupting tooth in the current tomographic image, thereby extracting its tooth mask in the next tomographic image. This process is repeated until the mask of the erupting tooth in all tomographic images is extracted.

[0060] In simple terms, the accurate segmentation of the tomographic image sequence of the first jaw includes two steps: First, based on the location, the local tomographic images are classified using the local tooth classification model to locate the tomographic images containing the necks of each erupting tooth; second, based on the location and extent information, the local tomographic images are segmented using the local tooth segmentation model. By referencing the location and extent information, the accuracy and efficiency of the local tooth classification model and the local tooth segmentation model are effectively improved.

[0061] In some cases, in addition to erupted teeth, the first jaw may also include impacted teeth. In this case, it is also necessary to segment these impacted teeth. In the local tomographic image segmentation, in addition to segmenting to obtain the mask sequence of erupted teeth, the mask sequence of impacted teeth is also obtained. However, the accuracy of this segmentation result for impacted teeth may not be high. Therefore, the following operations can be performed to obtain a more accurate segmentation result for impacted teeth.

[0062] In 111, impacted teeth were detected using a tooth mask brightness curve.

[0063] In one embodiment, the horizontal axis of the brightness curve can be the location of the tomographic image, and the vertical axis can be the sum of the brightness of the tooth mask in each tomographic image, that is, the sum of the number of pixels or the sum of the pixel values ​​of the tooth mask.

[0064] Typically, in the z-axis direction, mandibular impacted teeth are located below the neck of mandibular erupting teeth, while maxillary impacted teeth are located above them. The previously obtained global tooth masking sequence is referred to as the first masking sequence. Since highly accurate local segmentation results of erupting teeth have already been obtained, the masks of all erupting tooth roots can be removed from the first masking sequence, retaining only the masks of the erupting tooth crowns, resulting in the second masking sequence. The tooth masking brightness curve generated based on this second masking sequence will include a main peak and a small peak. The main peak corresponds to the tomographic image index range where the erupting tooth crown is located, while the small peak corresponds to the tomographic image index range where the impacted tooth is located. Then, impacted teeth can be detected within the tomographic image index range corresponding to the small peak. Since the masks of the erupting tooth roots have been removed in the second tooth masking sequence, only the impacted tooth mask is present within the range of the second tooth masking sequence.

[0065] Then, for each impacted tooth, the local tomographic image is pre-segmented using a local tooth segmentation model to obtain its mask sequence. The mask with the largest area is then selected, and the center point of the mask is used as the seed point for subsequent local segmentation.

[0066] At this point, there may be false seed points among the seed points obtained. In one embodiment, these false seed points can be filtered out based on the distance discrimination criterion using the following method.

[0067] In one embodiment, a first distance threshold can be set to 3mm. If the distance (three-dimensional spatial distance) between the seed points of two impacted teeth is less than the first distance threshold, then the two seed points are likely to belong to the same impacted tooth. Therefore, the seed point with the smaller masking area is removed.

[0068] In one embodiment, a second distance threshold can be set to 15mm. If the distance between the seed point of the impacted tooth and the seed point of the nearest erupting tooth is greater than the second distance threshold, then the seed point is likely to belong to the jawbone, and therefore the seed point of the impacted tooth is removed.

[0069] Finally, for each remaining seed point, starting from the layer it belongs to, local segmentation is performed in both the crown and root directions using the local tooth segmentation model to obtain the complete mask sequence of the corresponding impacted tooth.

[0070] After obtaining the mask sequence of all teeth (including erupting and impacted teeth), it can be optimized using methods such as adjacent layer constraints, Gaussian smoothing, and watershed segmentation.

[0071] The tooth masking neighbor-layer constraint algorithm can remove over-segmented regions in tooth segmentation results. Its specific operation is as follows:

[0072] Let S be the slice index of the seed point. Segment the local tooth image Is of slice S to obtain the masked image M of the tooth in slice S. S ;

[0073] Using Is as the template image and S+1 tomographic image as the source image, template matching is performed to obtain the center point of the tooth in S+1 tomographic image.

[0074] Local images of teeth on S+1 section I S+1 Segmentation is performed to obtain the masked image M of the teeth in the S+1 section. S+1 ;

[0075] Perform a displacement transformation on Ms, so that M S The center point and M S+1 Their center points coincide;

[0076] Perform morphological dilation on Ms to obtain the dilated mask image M. SD The size of the structural element can be 3;

[0077] Use M SD and M S+1 Perform an AND operation to obtain the masking result image M after the adjacent crown masking constraint. Rs+1 ;

[0078] Use M RS+1 As the masking image of the S+1 section, the above operation is repeated for the S+2 section to calculate the masking image of the S+2 section, and so on, until all tooth masks have been processed.

[0079] In one embodiment, Gaussian smoothing can be used to smooth the tooth mask image sequence processed by the neighboring layer constraint algorithm.

[0080] The watershed segmentation algorithm can detect the boundaries of connected tooth crowns and remove adjacent tooth regions from the segmentation results. The specific operation is as follows:

[0081] The local image of the i-th crown tomographic layer is segmented using a local tooth segmentation model to obtain the crown mask image M. io ;

[0082] Determine whether the crown is connected to the adjacent tooth. If M io If the connected regions of the crown in the image contain the image boundary location, then it is determined that the crown is connected to the adjacent tooth, i.e., the crown masking image M. io It includes the adjacent tooth region;

[0083] The tooth masking image M of layer i-1 i-1 and M io The marker image (Marker) is generated according to the following equation (1).

[0084] Marker=Erode(((Mio-Dilate(Mi-1))|Mio)) Equation (1)

[0085] Where Erode and Dilate are morphological erosion and dilation operations, respectively, and "|" is the OR operation.

[0086] Using the marker image, the watershed segmentation algorithm is applied to the local image of the i-th layer of the crown tomography to obtain the tooth boundary image B;

[0087] Use M io Subtracting the boundary image B yields the separated crown mask image. Extracting the connected region containing the image center point from the crown mask image yields the crown mask image M of tomography i. i , of which M i This is a crown mask that does not include the area of ​​adjacent teeth.

[0088] After obtaining the mask sequence of all teeth in the first jaw, a three-dimensional digital model of these teeth (including crowns and roots) can be generated based on it. For some dental treatment projects, the overall three-dimensional digital model of the teeth is very useful because it can obtain not only the relationship between the crowns, but also the relationship between the roots.

[0089] Although various aspects and embodiments of this application have been disclosed herein, other aspects and embodiments of this application will be apparent to those skilled in the art upon inspiration from this application. The various aspects and embodiments disclosed herein are for illustrative purposes only and not for limiting purposes. The scope and spirit of this application are determined solely by the appended claims.

[0090] Similarly, the diagrams may illustrate exemplary architectures or other configurations of the disclosed methods and systems, which aid in understanding the features and functions that may be included in the disclosed methods and systems. The claims are not limited to the exemplary architectures or configurations shown, and the desired features may be implemented with various alternative architectures and configurations. Furthermore, the order of the blocks given herein with respect to flowcharts, functional descriptions, and method claims should not be limited to various embodiments implemented in the same order to perform the said functions, unless explicitly indicated in the context.

[0091] Unless otherwise expressly stated, the terms and phrases used herein, and their variations thereof, should be interpreted as open-ended rather than restrictive. In some instances, the appearance of extended words and phrases such as “one or more,” “at least,” “but not limited to,” or other similar expressions should not be construed as an intention or necessity to indicate a narrower scope in examples where such extended expressions might not exist.

Claims

1. A computer-implemented method of segmenting a computer tomography image of a tooth, comprising: obtaining a first three-dimensional digital model representing a first dental arch of a first dentition and a sequence of two-dimensional tomographic images of the first dental arch; extracting a sequence of mask images of the first dental arch of the first dentition based on the sequence of two-dimensional tomographic images of the first dental arch using a global image segmentation model, wherein the global image segmentation model is a trained deep neural network for segmenting a two-dimensional tomographic image to extract a global tooth mask image; registering the first three-dimensional digital model and the sequence of mask images of the first dental arch of the first dentition to obtain location and extent information for each tooth; selecting a starting segmented two-dimensional tomographic image for each tooth from the sequence of two-dimensional tomographic images of the first dental arch based on the sequence of two-dimensional tomographic images of the first dental arch using a local image classification model, wherein the local image classification model is a trained deep neural network for classifying a local two-dimensional tomographic image into a class comprising a crown and a root; and for each tooth, segmenting a local image of the tooth from the corresponding starting segmented two-dimensional tomographic image in a direction of a crown and a root using a local image segmentation model to obtain a sequence of binary mask images of the tooth. The first three-dimensional digital model is obtained by one of intraoral scanning or scanning a dental impression or a physical model.

2. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 1, wherein, The sequence of two-dimensional tomographic images of the first dental arch is obtained by cone beam computed tomography.

3. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 1, wherein, It further comprises:

4. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 1, wherein, for each tooth, segmenting a local image of the tooth in a starting segmented two-dimensional tomographic image of the tooth using the local image segmentation model to obtain a mask binary image of the tooth corresponding to the starting segmented two-dimensional tomographic image; and for each tooth, using a mask binary image corresponding to a previous two-dimensional tomographic image as extent information and segmenting a local image of the tooth in a next two-dimensional tomographic image using the local image segmentation model with the extent information. It further comprises: selecting a sequence of mask images of a crown portion from the sequence of mask images of the first dental arch of the first dentition using the local image classification model; and 5. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 1, wherein, projecting the first three-dimensional digital model and the sequence of mask images of the crown portion of the first dental arch of the first dentition onto a first plane and registering the projections to obtain the location and extent information. For each tooth, the local image of the tooth in the starting segmented two-dimensional tomographic image is located in a middle section of the tooth. For each tooth, the local image of the tooth in the starting segmented two-dimensional tomographic image is located in a cervical section of the tooth.

6. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 1, wherein, It further comprises:

7. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 6, wherein, extracting a sequence of global tooth mask images of the first dental arch based on the sequence of two-dimensional tomographic images of the first dental arch using a global image segmentation model, wherein the global image segmentation model is a trained deep neural network for segmenting a two-dimensional tomographic image to extract a global tooth mask image; 8. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 1, wherein, ​ ​ The local image classification model is used to delete the mask of the root part of all erupted teeth in the global tooth mask image sequence of the first dental arch, to obtain a second mask image sequence; A mask brightness curve is generated based on the second mask image sequence, and a range of the impacted tooth in the two-dimensional tomographic image sequence of the first dental arch is determined based on the brightness curve; For each impacted tooth, a starting segmented two-dimensional tomographic image is determined within the range of the impacted tooth; and For each impacted tooth, a local image of the impacted tooth is finally segmented in the direction of the crown and the root from the corresponding starting segmented two-dimensional tomographic image using the local image segmentation model, to obtain a binary mask image sequence thereof.

9. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 8, wherein, It further comprises, for each impacted tooth, pre-segmenting the local image thereof within the range of the two-dimensional tomographic image using the local image segmentation model, and determining the starting segmented two-dimensional tomographic image thereof based on the mask area obtained by segmentation.

10. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 9, wherein, For each impacted tooth, the two-dimensional tomographic image corresponding to the largest area mask obtained by pre-segmentation is taken as the starting segmented two-dimensional tomographic image.

11. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 10, wherein, It further comprises, if the distance between the center points of the largest area masks of two impacted teeth is less than a first threshold value, then the local image segmentation model is not used for final segmentation of the impacted tooth corresponding to the mask with smaller area.

12. The computer-implemented method of segmenting a computed tomography image of a tooth according to claim 10, wherein, It further comprises, if the distance between the center point of the largest area mask of an impacted tooth and the center point of the mask corresponding to the segmentation starting two-dimensional tomographic image of the nearest erupted tooth is greater than a second distance threshold value, then the local image segmentation model is not used for final segmentation of the impacted tooth. It further comprises, if the distance between the center point of the largest area mask of an impacted tooth and the center point of the mask corresponding to the segmentation starting two-dimensional tomographic image of the nearest erupted tooth is greater than a second distance threshold value, then the local image segmentation model is not used for final segmentation of the impacted tooth.

Images