A dental CT image three-dimensional segmentation method, system, device and medium
Patent Information
- Application Number
- CN202310973116.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-03
- Publication Date
- 2025-11-18
- Estimated Expiration
- 2043-08-03
AI Technical Summary
Metal artifacts and blurred boundary information exist in dental CT images, affecting the accuracy of tooth segmentation.
We employ an edge-based distance regularization level set evolution method and a labeled watershed transform method, combined with a Sobel filter and a boundary information module, to perform binarization, projection, edge segmentation, and feature enhancement on dental CT images. Through 3D-2D-3D projection transformation, we utilize the labeled watershed transform to perform foreground and background label segmentation.
It effectively eliminates metal artifacts, reduces noise effects, improves the accuracy of tooth image segmentation, reduces overfitting problems, and achieves more accurate individual tooth segmentation.
Smart Images
Figure CN117011318B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of medical image processing, in particular to a tooth CT image three-dimensional segmentation method, system, device and medium. BACKGROUND
[0002] With the rapid development of deep learning technology in recent years, deep learning has a great influence on computer vision. At the same time, the wide application of deep learning technology promotes the development of medical image segmentation technology. Teeth are important organs of the human body and play an irreplaceable role in daily life. Ensuring the health of teeth helps people improve diet, absorb nutrients and maintain physical health. For dentists, how to quickly and accurately measure tooth images has become a problem, and the importance of deep learning technology in tooth image segmentation has gradually emerged. However, in the direction of oral surgery tooth image segmentation, there are problems of metal artifacts and blurred boundary information in the collected tooth images, which greatly affect the accuracy of tooth segmentation. SUMMARY
[0003] The purpose of the present application is to provide a tooth CT image three-dimensional segmentation method, system, device and medium, which can improve the segmentation accuracy of tooth images.
[0004] To achieve the above purpose, the present application provides the following scheme:
[0005] A tooth CT image three-dimensional segmentation method, comprising:
[0006] obtaining a three-dimensional tooth CT image;
[0007] performing binaryzation processing on the three-dimensional tooth CT image, and projecting the image after binaryzation processing in a set direction to obtain a dental arch two-dimensional image; the dental arch two-dimensional image comprises a gray scale image corresponding to the upper jaw and the lower jaw;
[0008] using an edge-based distance regularization level set evolution method to perform edge segmentation on the dental arch two-dimensional image to obtain a dental arch two-dimensional feature image;
[0009] inputting the dental arch two-dimensional feature image into a Sobel filter and a boundary information module in sequence to perform feature enhancement and extraction on the tooth boundary information in the image to obtain a two-dimensional tooth segmentation prediction map; the boundary information module is constructed according to a residual block and a corresponding block connected with the residual block;
[0010] performing three-dimensional projection on the two-dimensional tooth segmentation prediction map in the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image;
[0011] The three-dimensional tooth segmentation image is marked and segmented by using a marked watershed transform method, so that a final three-dimensional segmentation image of the tooth CT image is obtained.
[0012] Optionally, the three-dimensional tooth CT image is binarized, and the binarized image is projected in a set direction to obtain a dental arch two-dimensional image, specifically including:
[0013] The three-dimensional tooth CT image is binarized by using a set threshold to obtain a two-dimensional tooth CT image in a space rectangular coordinate system.
[0014] The maxima intensity projection method is used in the z direction to project the two-dimensional tooth CT image to obtain a gray scale image corresponding to the upper jaw and the lower jaw, and the dental arch two-dimensional image is determined according to the gray scale image corresponding to the upper jaw and the lower jaw.
[0015] Optionally, an edge-based distance regularization level set evolution method is used to perform edge segmentation on the dental arch two-dimensional image to obtain a dental arch two-dimensional feature image, specifically including:
[0016] The edge-based distance regularization level set is used to segment the dental arch region in the dental arch two-dimensional image by providing a smooth closed contour to evolve the curve based on the image gradient to obtain a dental arch two-dimensional feature image.
[0017] Optionally, the training method of the boundary information module is:
[0018] Training data is obtained; the training data includes a tooth training image and a corresponding segmentation image.
[0019] A training model for boundary information extraction is constructed.
[0020] The training data is input into the training model, and the training is performed by using an edge loss function, and the trained training model is determined as the boundary information module.
[0021] Optionally, the edge loss function is:
[0022]
[0023] Wherein, Ipre(i) is a network prediction map, and Igt(i) is a real boundary map.
[0024] Optionally, the two-dimensional tooth segmentation prediction map is projected in three dimensions in the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image, specifically including:
[0025] project the two-dimensional tooth segmentation prediction graph in the three-dimensional tooth CT image in three dimensions, and generate a three-dimensional tooth segmentation image by sequentially performing two-dimensional projection and three-dimensional projection on the generated three-dimensional projection.
[0026] Optionally, the three-dimensional tooth segmentation image is foreground and background marked segmented by the marked watershed transformation method to obtain a final tooth CT image three-dimensional segmentation image, specifically including:
[0027] The marked watershed transformation method is used to take the foreground as a seed point, take the background as an irrelevant region, and perform segmentation near the seed point to obtain the final tooth CT image three-dimensional segmentation image.
[0028] The application further provides a tooth CT image three-dimensional segmentation system, including:
[0029] An image acquisition unit is configured to acquire a three-dimensional tooth CT image.
[0030] An image preprocessing unit is configured to perform binaryzation processing on the three-dimensional tooth CT image, and project the image after the binaryzation processing in a set direction to obtain a dental arch two-dimensional image; the dental arch two-dimensional image includes corresponding gray-scale images of the upper jaw and the lower jaw.
[0031] An evolution segmentation unit is configured to perform edge segmentation on the dental arch two-dimensional image by using an edge-based distance regularization level set evolution method to obtain a dental arch two-dimensional feature image.
[0032] A tooth region prediction unit is configured to input the dental arch two-dimensional feature image into a Sobel filter and a boundary information module in sequence, perform feature enhancement and extraction on tooth boundary information in the image to obtain a two-dimensional tooth segmentation prediction graph; the boundary information module is constructed according to a residual block and a corresponding block connected with the residual block.
[0033] A projection unit is configured to project the two-dimensional tooth segmentation prediction graph in the three-dimensional tooth CT image in three dimensions to generate a three-dimensional tooth segmentation image.
[0034] A final image generation unit is configured to perform foreground and background marked segmentation on the three-dimensional tooth segmentation image by using a marked watershed transformation method to obtain a final tooth CT image three-dimensional segmentation image.
[0035] The application further provides an electronic device including a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the tooth CT image three-dimensional segmentation method according to the above.
[0036] The application further provides a computer readable storage medium storing a computer program, which, when executed by a processor, implements the tooth CT image three-dimensional segmentation method.
[0037] According to the specific embodiments of the application, the following technical effects are disclosed.
[0038] The application discloses a tooth CT image three-dimensional segmentation method, system, device and medium, and the method comprises the following steps: acquiring a three-dimensional tooth CT image; performing binaryzation processing and projection on the three-dimensional tooth CT image to obtain a dental arch two-dimensional image; performing segmentation on the dental arch two-dimensional image by using an edge-based distance regularization level set evolution method to obtain a dental arch two-dimensional feature image; inputting the dental arch two-dimensional feature image into a Sobel filter and a boundary information module in sequence, performing feature enhancement and extraction on tooth boundary information in the image to obtain a two-dimensional tooth segmentation prediction image; projecting the two-dimensional tooth segmentation prediction image in the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image; and performing foreground and background marked segmentation on the three-dimensional tooth segmentation image by using a marked watershed transformation method to obtain a final tooth CT image three-dimensional segmentation image. BRIEF DESCRIPTION OF DRAWINGS
[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings described in the following are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.
[0040] Figure 1 The figure is a flowchart of the tooth CT image three-dimensional segmentation method. DETAILED DESCRIPTION
[0041] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.
[0042] The purpose of this invention is to provide a method, system, device, and medium for three-dimensional segmentation of dental CT images, which can improve the segmentation accuracy of dental images.
[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0044] like Figure 1 As shown, the present invention provides a three-dimensional segmentation method for dental CT images, comprising:
[0045] Step 100: Acquire 3D dental CT images;
[0046] Step 200: The three-dimensional dental CT image is binarized, and the binarized image is projected in a set direction to obtain a two-dimensional image of the dental arch; the two-dimensional image of the dental arch includes grayscale images corresponding to the maxilla and mandible;
[0047] Step 300: Use the edge-based distance regularization level set evolution method to perform edge segmentation on the two-dimensional dental arch image to obtain a two-dimensional feature image of the dental arch;
[0048] Step 400: The two-dimensional feature image of the dental arch is sequentially input into the Sobel filter and the boundary information module to enhance and extract the tooth boundary information in the image, thereby obtaining a two-dimensional tooth segmentation prediction map; the boundary information module is constructed based on the residual block and the corresponding block connected to the residual block;
[0049] Step 500: Project the two-dimensional tooth segmentation prediction image into the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image;
[0050] Step 600: Using the labeled watershed transform method, the three-dimensional tooth segmentation image is segmented by foreground and background labels to obtain the final three-dimensional segmentation image of the tooth CT image.
[0051] The training method for the boundary information module is as follows:
[0052] Acquire training data; the training data includes tooth training images and corresponding segmented images; construct a training model for boundary information extraction; input the training data into the training model and train it using an edge loss function, then determine the trained model as the boundary information module. Furthermore, the edge loss function is: Where Ipre(i) is the network's predicted image, and Igt(i) is the ground truth boundary image. As one implementation of step 200, it specifically includes: binarizing the three-dimensional dental CT image using a set threshold to obtain a two-dimensional dental CT image in a spatial rectangular coordinate system; extracting the maxillary and mandibular regions from the two-dimensional dental CT image using a connected region labeling method, projecting them in the z-direction using the maximum intensity projection method to obtain grayscale images corresponding to the maxilla and mandible, and determining a two-dimensional dental arch image based on the corresponding grayscale images of the maxilla and mandible. As one implementation of step 300, it specifically includes: segmenting the dental arch region in the two-dimensional dental arch image by providing a smooth, closed contour based on the edge distance regularization level set and allowing the curve to evolve based on the image gradient, thereby obtaining a two-dimensional dental arch feature image. As one implementation of step 500, it specifically includes: projecting the two-dimensional tooth segmentation prediction image into the three-dimensional dental CT image in three dimensions, and then sequentially performing two-dimensional and three-dimensional projections on the generated three-dimensional projection to generate a three-dimensional tooth segmentation image. As one implementation of step 600, it specifically includes: using the labeled watershed transform method, taking the foreground as seed points and the background as irrelevant regions, and performing segmentation near the seed points to obtain the final three-dimensional segmented image of the dental CT image. Based on the above technical solution, the following embodiment is provided. The following steps are performed on the acquired three-dimensional dental CT image: Step 1: Set a threshold for the input three-dimensional dental CT image and use the connected component labeling method to segment the maxilla and mandible; the acquired three-dimensional dental CT image includes maxilla, mandible, soft tissue, etc., and for the convenience of subsequent tooth segmentation, the maxilla and mandible need to be reconstructed separately.
[0053] To obtain binary images, we segmented the 3D dental CT images into three categories: air, soft tissue, and bone, and set thresholds Th0 and Th1 as follows:
[0054]
[0055] The obtained Th1 is the interface between soft tissue and bone. Therefore, we can obtain the binarized image of the three-dimensional image X(x,y,z) as follows:
[0056]
[0057] In the formula, Th0 and Th1 are the thresholds between air and bone, and between soft tissue and bone, respectively. q(k) is the average histogram corresponding to the CT image. This is the final binary image obtained.
[0058] Step 2: Perform two-dimensional slicing on the segmented maxilla and mandible; due to factors such as equipment, metal artifact noise often exists when actually acquiring three-dimensional dental CT images. Therefore, we reconstruct the acquired three-dimensional CT images to eliminate metal artifacts.
[0059] To better extract the maxilla and mandible from the binarized 3D dental CT images obtained in Step 1 using Connected Region Labeling (CCL), the mandible is the largest connected region, and the maxilla is the second largest connected region. To obtain the 2D image of the dental arch from CBCT, Maximum Intensity Projection (MIP) is applied in the z-direction to obtain the corresponding grayscale images of the maxilla and mandible.
[0060]
[0061] In the formula, X u For the reconstructed maxillary image, X l Image of the reconstructed mandible. Xu with I Xl These are two-dimensional images of the maxilla and mandible obtained after MIP operation.
[0062] Step 3: Construct a two-dimensional panoramic image by using edge-based distance regularization level sets; for the reconstructed image after eliminating metal artifact noise, it is first necessary to obtain the rough image curves of the maxillary and mandibular teeth for overall segmentation.
[0063] A coarse tooth contour image is obtained using an edge-based distance-regularized level set evolution method. Edge-based level sets provide a smooth, closed contour, allowing the curve to evolve based on the image gradient, effectively addressing the uneven intensity distribution characteristic of CBCT tooth images. Regularized level sets eliminate the need for re-initialization during evolution, effectively overcoming the shortcomings of traditional level sets, such as oscillations and deviations from the objective function. The edge-based distance-regularized level set evolution method initially segments the dental arch region, resulting in the following curve:
[0064]
[0065] In the formula, N s The number of curve points, Cubic spline curves obtained from maxillary images
[0066] The final two-dimensional panoramic image IF is obtained based on this curve:
[0067]
[0068] In the formula, s is the same as in formula (4), r(s) is a part of the cubic spline curve obtained by formula (4), and n(s) is the unit normal vector of r(s).
[0069] Step 4: Design a boundary information module to extract tooth boundary information and perform segmentation; blurred tooth boundary information has always been a challenge in tooth CT image segmentation. Therefore, how to effectively extract boundary information to reduce the impact of boundary blur on segmentation accuracy is the key problem to be solved in this step and the focus of this embodiment.
[0070] A boundary information extraction (BIE) module is designed to extract boundary information from images. A key challenge in CBCT tooth image segmentation lies in the blurred boundary information; adjacent teeth in the crown region, as well as the roots and alveolar bone, are often difficult to distinguish. We aim to reduce the impact of boundary information on segmentation results by extracting it from tooth images. A Sobel operator is designed, using the weights of the boundary information from the feature map obtained in step three to distinguish edge voxels from other voxels.
[0071]
[0072] In the formula, This represents the element-wise multiplication operation.
[0073] After the feature map is processed by a Sobel filter to enhance boundary information, the resulting feature map is input into the boundary information module to extract boundary information. For the specific implementation of the BIE module, a residual block is designed and summed with the corresponding block at position i+1. Simultaneously, to reduce the impact of accumulated errors, the combined feature map is also processed through a BIE module before the input of the next round, ultimately obtaining a two-dimensional tooth segmentation prediction map.
[0074] The designed edge loss function is as follows:
[0075]
[0076] In the formula, Ipre(i) is the prediction graph of the network, and Igt(i) is the ground truth boundary graph.
[0077] Step 5: Reproject back to the 3D image; After segmenting the panoramic image, it needs to be reprojected back to the 3D image to segment individual teeth.
[0078] Based on formulas (4) and (5), the two-dimensional image of a single tooth segmented in step four is reprojected back into the three-dimensional tooth segmentation image. First, the three-dimensional tooth image is projected to generate a two-dimensional image, and then reprojected. This effectively removes the influence of metal artifact noise during the acquisition of the three-dimensional CBCT image of the tooth. The specific formulas are as follows:
[0079] D={(r(s)+xn(s),z):-α≤x≤α, (s,z)∈Ipre} (8)
[0080] In the formula, α is a hyperparameter.
[0081] Step 6: Segment individual teeth using the labeled watershed transform to obtain the final tooth segmentation image.
[0082] The ultimate goal of this embodiment is to segment and recognize a single tooth image, which requires segmentation in the reconstructed 3D CT image.
[0083] Using the 3D CT image reprojected in step five as input for the label watershed transform, and guided by the segmentation results in the 2D image from step four, as well as the foreground and background labels in the image, single tooth segmentation of the CBCT image is achieved, resulting in the final segmented image.
[0084] Specifically, in this embodiment, a thresholding operation is performed on the probability map output by the network, and then foreground and background labels are generated using the processed image. The label watershed transform method is used, with the foreground as seed points and the background as irrelevant regions. The algorithm performs segmentation near the seed points, effectively reducing overfitting and mitigating the impact of noise, resulting in more effective and accurate segmentation of individual tooth images.
[0085] Furthermore, the present invention also provides a three-dimensional segmentation system for dental CT images, comprising:
[0086] Image acquisition unit, used to acquire three-dimensional dental CT images;
[0087] An image preprocessing unit is used to binarize the three-dimensional dental CT image and project the binarized image in a set direction to obtain a two-dimensional image of the dental arch; the two-dimensional image of the dental arch includes grayscale images corresponding to the maxilla and mandible.
[0088] An evolutionary segmentation unit is used to perform edge segmentation on the two-dimensional dental arch image using an edge-based distance regularized level set evolution method to obtain a two-dimensional dental arch feature image.
[0089] The tooth region prediction unit is used to sequentially input the two-dimensional feature image of the dental arch into the Sobel filter and the boundary information module, perform feature enhancement and extraction on the tooth boundary information in the image, and obtain a two-dimensional tooth segmentation prediction map; the boundary information module is constructed based on the residual block and the corresponding block connected to the residual block;
[0090] The projection unit is used to project the two-dimensional tooth segmentation prediction map into the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image.
[0091] The final image generation unit is used to perform foreground and background labeling and segmentation on the three-dimensional tooth segmentation image using the labeled watershed transform method to obtain the final three-dimensional segmentation image of the tooth CT image.
[0092] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the three-dimensional segmentation method for dental CT images described above.
[0093] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the three-dimensional segmentation method for dental CT images as described above.
[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0095] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for three-dimensional segmentation of dental CT images, characterized in that, include: Acquire 3D dental CT images; The three-dimensional dental CT image is binarized, and the binarized image is projected in a set direction to obtain a two-dimensional image of the dental arch; the two-dimensional image of the dental arch includes grayscale images corresponding to the maxilla and mandible. The two-dimensional dental arch image is segmented using an edge-based distance regularization level set evolution method to obtain a two-dimensional dental arch feature image. The two-dimensional feature image of the dental arch is sequentially input into the Sobel filter and the boundary information module to enhance and extract the tooth boundary information in the image, thereby obtaining a two-dimensional tooth segmentation prediction map; the boundary information module is constructed based on the residual block and the corresponding block connected to the residual block; The two-dimensional tooth segmentation prediction map is projected into the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image. Using the labeled watershed transform method, the three-dimensional tooth segmentation image is segmented by foreground and background labels to obtain the final three-dimensional segmentation image of the tooth CT image; The training method for the boundary information module is as follows: Acquire training data; the training data includes tooth training images and corresponding segmented images; Construct a training model for boundary information extraction; The training data is input into the training model, and the edge loss function is used for training. The trained model is then identified as the boundary information module. The edge loss function is: Where Ipre(i) is the network's prediction graph, and Igt(i) is the ground truth boundary graph.
2. The three-dimensional segmentation method for dental CT images according to claim 1, characterized in that, The three-dimensional dental CT image is binarized, and the binarized image is projected in a set direction to obtain a two-dimensional image of the dental arch, specifically including: By using a set threshold, the three-dimensional dental CT image is binarized to obtain a two-dimensional dental CT image in a spatial rectangular coordinate system. The maxillary and mandibular regions in the two-dimensional dental CT image are extracted using the connected region labeling method. The maximum intensity projection method is used to project the images in the z-direction to obtain grayscale images corresponding to the maxilla and mandible. The two-dimensional image of the dental arch is then determined based on the grayscale images corresponding to the maxilla and mandible.
3. The three-dimensional segmentation method for dental CT images according to claim 1, characterized in that, The dental arch 2D image is segmented using an edge-based distance regularization level set evolution method to obtain a dental arch 2D feature image, specifically including: Based on the edge distance regularization level set, the dental arch region in the two-dimensional dental arch image is segmented by providing a smooth closed contour to allow the curve to evolve based on the image gradient, thus obtaining a two-dimensional feature image of the dental arch.
4. The three-dimensional segmentation method for dental CT images according to claim 1, characterized in that, The two-dimensional tooth segmentation prediction map is projected into the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image, specifically including: The two-dimensional tooth segmentation prediction map is projected into the three-dimensional tooth CT image in three dimensions. The generated three-dimensional projection is then projected into two dimensions and then into three dimensions in sequence to generate a three-dimensional tooth segmentation image.
5. The three-dimensional segmentation method for dental CT images according to claim 1, characterized in that, The three-dimensional tooth segmentation image is segmented using the labeled watershed transform method to obtain the final three-dimensional segmented tooth CT image, specifically including: Using the labeled watershed transform method, the foreground is used as a seed point and the background is used as an unrelated region. Segmentation is performed near the seed point to obtain the final three-dimensional segmentation image of the tooth CT image.
6. A three-dimensional segmentation system for dental CT images, using the method as described in any one of claims 1-5, characterized in that, include: Image acquisition unit, used to acquire three-dimensional dental CT images; An image preprocessing unit is used to binarize the three-dimensional dental CT image and project the binarized image in a set direction to obtain a two-dimensional image of the dental arch; the two-dimensional image of the dental arch includes grayscale images corresponding to the maxilla and mandible. An evolutionary segmentation unit is used to perform edge segmentation on the two-dimensional dental arch image using an edge-based distance regularized level set evolution method to obtain a two-dimensional dental arch feature image. The tooth region prediction unit is used to sequentially input the two-dimensional feature image of the dental arch into the Sobel filter and the boundary information module, perform feature enhancement and extraction on the tooth boundary information in the image, and obtain a two-dimensional tooth segmentation prediction map; the boundary information module is constructed based on the residual block and the corresponding block connected to the residual block; The projection unit is used to project the two-dimensional tooth segmentation prediction map into the three-dimensional tooth CT image to generate a three-dimensional tooth segmentation image. The final image generation unit is used to perform foreground and background labeling and segmentation on the three-dimensional tooth segmentation image using the labeled watershed transform method to obtain the final three-dimensional segmentation image of the tooth CT image.
7. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the three-dimensional segmentation method for dental CT images according to claims 1-5.
8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the three-dimensional segmentation method for dental CT images as described in claims 1-5.