Method and device for detecting three-dimensional cephalometric landmark points, medium and terminal

Abstract

This application discloses a method, device, medium, and terminal for detecting landmark points in three-dimensional cephalometric radiographs, relating to the field of orthodontic technology. Its main objective is to improve the low localization accuracy of existing landmark point detection models trained using two-dimensional lateral cephalometric radiographs as training samples due to overlapping anatomical structures, image distortion, and loss of depth information. The method primarily utilizes a transformation matrix from a two-dimensional view coordinate system to a three-dimensional anatomical coordinate system, enabling the conversion of 2D view coordinates to 3D anatomical coordinates. This ensures that landmark points are no longer limited to their fixed points on the two-dimensional structure of the lateral cephalometric radiograph, avoiding problems such as overlapping anatomical structures and image distortion, thus improving the accuracy of landmark point detection. During the detection process, different detection methods are employed for different types of landmark points, solving the problem that overlapping anatomical structures cannot encompass all landmark point positions, further improving the accuracy of landmark point detection.

Description

Technical Field

[0001] This application relates to the field of orthodontic technology, and in particular to a method, device, medium, and terminal for detecting three-dimensional cephalometric landmarks. Background Technology

[0002] In orthodontic treatment, cephalometrics is a core analytical tool for diagnosing dentofacial deformities, developing treatment plans, and evaluating treatment effectiveness. Traditional cephalometric methods mainly involve manual annotation by doctors on two-dimensional lateral cephalometric radiographs, which is time-consuming, labor-intensive, and prone to inconsistencies between different doctors.

[0003] Currently, in order to solve the problems of time-consuming, labor-intensive and inconsistent traditional cephalometric methods, existing methods have introduced neural networks. Specifically, they acquire lateral cephalometric images and doctor-labeled landmarks, and use these as training samples to train a landmark detection model. The trained landmark detection model is then used to determine the landmark locations of new patients.

[0004] However, due to the overlap of bilateral anatomical structures and image distortion in two-dimensional lateral cephalometric images of the human skull in three-dimensional structures, some landmarks are difficult to identify due to occlusion and deformation, resulting in a decrease in the accuracy of training samples. This, in turn, leads to a decrease in the localization accuracy of the landmark detection model after model training. Furthermore, since two-dimensional lateral cephalometric images lose depth information, they cannot accurately reflect the true positional relationship of landmarks in three-dimensional space, further exacerbating the problem of low localization accuracy. Summary of the Invention

[0005] In view of this, this application provides a method, device, medium, and terminal for detecting three-dimensional cephalometric landmarks. The main purpose is to improve the problem that the positioning accuracy of the landmark detection model obtained by using two-dimensional cephalometric lateral radiographs as training samples is low due to the overlapping of bilateral anatomical structures, image distortion, and loss of depth information.

[0006] According to one aspect of this application, a method for detecting three-dimensional cephalometric landmarks is provided, comprising: A three-dimensional skull model of the target patient is obtained, and a multi-view two-dimensional view cropping operation is performed on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. The two-dimensional views from multiple perspectives are input into multiple predictable landmark detection models that have completed model training, respectively, to obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on a preset number of two-dimensional views from different perspectives. For each predictable landmark, coordinate transformation is performed based on the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. For each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. By integrating the three-dimensional anatomical coordinates of all predictable and all unpredictable landmarks, the three-dimensional cephalometric landmark detection results of the target patient are obtained.

[0007] Preferably, after integrating the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient, the method further includes: The detection results of the three-dimensional head shadow measurement markers are superimposed and rendered on the three-dimensional skull model in a preset form, and output in the 3D rendering area of ​​the user interface. The system calls a preset formula library and uses the cephalometric analysis formulas in the preset formula library to calculate multiple cephalometric clinical indicators based on the detection results of the three-dimensional cephalometric landmarks. These indicators are then output in the cephalometric clinical indicator output area of ​​the user interface.

[0008] Preferably, the step of inputting the two-dimensional views from the multiple perspectives into multiple pre-trained predictable landmark detection models to obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view includes: Input the front view and the right side view into the first predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the root of the nose in the front view, the two-dimensional view coordinates of the root of the nose in the right side view, the two-dimensional view coordinates of the right eye socket in the front view, and the two-dimensional view coordinates of the right eye socket in the right side view. Input the front view and the left side view into the second predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the left eye socket in the front view, the two-dimensional view coordinates of the left eye socket in the left side view, the two-dimensional view coordinates of the upper alveolar seat in the front view, and the two-dimensional view coordinates of the upper alveolar seat in the left side view. Input the bottom view and the right side view into the third predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the right side view, the two-dimensional view coordinates of the right mandibular angle point in the bottom view, and the two-dimensional view coordinates of the right mandibular angle point in the right side view. Input the bottom view and the left side view into the fourth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the left side view, the two-dimensional view coordinates of the left mandibular angle point in the bottom view, and the two-dimensional view coordinates of the left mandibular angle point in the left side view. The side view and the front view are input into the fifth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the mandibular alveolar seat in the front view, the two-dimensional view coordinates of the mandibular alveolar seat in the side view, the two-dimensional view coordinates of the anterior chin point in the front view, the two-dimensional view coordinates of the anterior chin point in the side view, the two-dimensional view coordinates of the chin vertex in the front view, the two-dimensional view coordinates of the chin vertex in the side view, the two-dimensional view coordinates of the submental point in the front view, and the two-dimensional view coordinates of the submental point in the side view. The side view is either a left side view or a right side view, determined by the mandibular deviation of the target patient. Input the cross-sectional view into the sixth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the pituitary fossa in the cross-sectional view; The front view, the left side view, and the right side view are input into the seventh predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the tooth cusp in the front view and the fused two-dimensional view coordinates of the tooth cusp in the side view. The fused two-dimensional view coordinates of the tooth cusp in the side view are obtained by fusing the two-dimensional view coordinates of the tooth cusp in the left side view and the two-dimensional view coordinates of the tooth cusp in the right side view.

[0009] Preferably, for each predictable landmark, coordinate transformation is performed based on the two-dimensional view coordinates of the predictable landmark in the corresponding two-dimensional view and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each view, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark, including: For each predictable marker point, obtain the two-dimensional view coordinates of the predictable marker point in the first-view two-dimensional view as the first two-dimensional view coordinates, and obtain the two-dimensional view coordinates of the predictable marker point in the second-view two-dimensional view as the second two-dimensional view coordinates. Multiply the first two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the first viewpoint to obtain the coordinates of the predictable landmark point on the first anatomical plane; and multiply the second two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the second viewpoint to obtain the coordinates of the predictable landmark point on the second anatomical plane. The coordinates of the predictable landmark on the first anatomical plane and the coordinates on the second anatomical plane are combined to obtain the three-dimensional anatomical coordinates of the predictable landmark.

[0010] Preferably, before inputting the multiple 2D views from various perspectives into multiple pre-trained predictable landmark detection models to obtain the 2D view coordinates of each predictable landmark in the corresponding 2D view, the method further includes: Obtain multiple sample 3D skull models; For each sample three-dimensional skull model, a predictable landmark three-dimensional anatomical coordinate annotation operation is performed on the sample three-dimensional skull model to obtain a sample three-dimensional skull model annotated with the predictable landmark standard three-dimensional anatomical coordinates. Then, a multi-view two-dimensional view cropping operation is performed on the sample three-dimensional skull model annotated with the predictable landmark standard three-dimensional anatomical coordinates to obtain sample two-dimensional views annotated with the predictable landmark standard three-dimensional anatomical coordinates from multiple perspectives. For each sample two-dimensional view labeled with the standard three-dimensional anatomical coordinates of the predictable landmarks, the standard two-dimensional view coordinates of the predictable landmarks are determined based on the standard three-dimensional anatomical coordinates of the predictable landmarks, thus obtaining a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable landmarks. All the predictable landmarks are divided into multiple groups according to their anatomical location. Each group of predictable landmarks corresponds to a preset number of sample two-dimensional views labeled with the standard two-dimensional view coordinates of the predictable landmarks from different perspectives. For each set of predictable marker points, an initial predictable marker point detection model is constructed, and the initial predictable marker point detection model is trained using a sample two-dimensional view group labeled with the standard two-dimensional view coordinates of the predictable marker points, to obtain the corresponding predictable marker point detection model that has completed model training.

[0011] Preferably, the predictable marker points include visible predictable marker points and invisible predictable marker points. The step of determining the standard two-dimensional view coordinates of the predictable marker points based on their standard three-dimensional anatomical coordinates, to obtain a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable marker points, includes: For each visible and predictable landmark, referencing the standard three-dimensional anatomical coordinates of the visible and predictable landmark, directly annotate the standard two-dimensional view coordinates of the landmark. For each invisible and predictable landmark, the standard two-dimensional view coordinates of the landmark are determined based on the standard three-dimensional anatomical coordinates of the landmark. The method of inversely inferring two-dimensional view coordinates from three-dimensional anatomical coordinates is used.

[0012] Preferably, before performing coordinate transformation processing on each predictable landmark point according to its two-dimensional view coordinates in the corresponding two-dimensional view and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each viewpoint, to obtain the coordinates of the predictable landmark point on the two anatomical planes, the method further includes: The spatial range of the three-dimensional skull model of the target patient along the first anatomical direction and the spatial range along the second anatomical direction in the three-dimensional anatomical coordinate system are determined, wherein the spatial range of the first anatomical direction is used to characterize the range between the minimum and maximum values ​​of the first anatomical direction, and the spatial range of the second anatomical direction is used to characterize the range between the minimum and maximum values ​​of the second anatomical direction. For each 2D view, the pixel width and pixel height of the 2D view are determined. Based on the spatial range of the first anatomical direction, the spatial range of the second anatomical direction, and the pixel width and pixel height of the 2D view, an affine transformation relationship between the 2D view coordinate system and the 3D anatomical coordinate system of the 2D view is established. Based on the affine transformation relationship, the transformation matrix from the 2D view coordinate system to the 3D anatomical coordinate system of the view is calculated.

[0013] According to another aspect of this application, a device for detecting three-dimensional cephalometric marker points is provided, comprising: The two-dimensional view cropping module is used to acquire a three-dimensional skull model of the target patient and perform multi-view two-dimensional view cropping operations on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. The predictable landmark detection module is used to input the two-dimensional views of the multiple perspectives into multiple predictable landmark detection models that have completed model training, and obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on the two-dimensional views of the corresponding preset number of perspectives. For each predictable landmark, coordinate transformation is performed according to the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. An unpredictable landmark detection module is used to calculate the three-dimensional anatomical coordinates of each unpredictable landmark based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. The detection result integration module integrates the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient.

[0014] Preferably, after the detection result integration module, the method further includes a detection result output module, used for: The detection results of the three-dimensional head shadow measurement markers are superimposed and rendered on the three-dimensional skull model in a preset form, and output in the 3D rendering area of ​​the user interface. The system calls a preset formula library and uses the cephalometric analysis formulas in the preset formula library to calculate multiple cephalometric clinical indicators based on the detection results of the three-dimensional cephalometric landmarks. These indicators are then output in the cephalometric clinical indicator output area of ​​the user interface.

[0015] Preferably, the predictable marker detection module is used for: Input the front view and the right side view into the first predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the root of the nose in the front view, the two-dimensional view coordinates of the root of the nose in the right side view, the two-dimensional view coordinates of the right eye socket in the front view, and the two-dimensional view coordinates of the right eye socket in the right side view. Input the front view and the left side view into the second predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the left eye socket in the front view, the two-dimensional view coordinates of the left eye socket in the left side view, the two-dimensional view coordinates of the upper alveolar seat in the front view, and the two-dimensional view coordinates of the upper alveolar seat in the left side view. Input the bottom view and the right side view into the third predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the right side view, the two-dimensional view coordinates of the right mandibular angle point in the bottom view, and the two-dimensional view coordinates of the right mandibular angle point in the right side view. Input the bottom view and the left side view into the fourth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the left side view, the two-dimensional view coordinates of the left mandibular angle point in the bottom view, and the two-dimensional view coordinates of the left mandibular angle point in the left side view. The side view and the front view are input into the fifth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the mandibular alveolar seat in the front view, the two-dimensional view coordinates of the mandibular alveolar seat in the side view, the two-dimensional view coordinates of the anterior chin point in the front view, the two-dimensional view coordinates of the anterior chin point in the side view, the two-dimensional view coordinates of the chin vertex in the front view, the two-dimensional view coordinates of the chin vertex in the side view, the two-dimensional view coordinates of the submental point in the front view, and the two-dimensional view coordinates of the submental point in the side view. The side view is either a left side view or a right side view, determined by the mandibular deviation of the target patient. Input the cross-sectional view into the sixth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the pituitary fossa in the cross-sectional view; The front view, the left side view, and the right side view are input into the seventh predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the tooth cusp in the front view and the fused two-dimensional view coordinates of the tooth cusp in the side view. The fused two-dimensional view coordinates of the tooth cusp in the side view are obtained by fusing the two-dimensional view coordinates of the tooth cusp in the left side view and the two-dimensional view coordinates of the tooth cusp in the right side view.

[0016] Preferably, the predictable marker detection module is further configured to: For each predictable marker point, obtain the two-dimensional view coordinates of the predictable marker point in the first-view two-dimensional view as the first two-dimensional view coordinates, and obtain the two-dimensional view coordinates of the predictable marker point in the second-view two-dimensional view as the second two-dimensional view coordinates. Multiply the first two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the first viewpoint to obtain the coordinates of the predictable landmark point on the first anatomical plane; and multiply the second two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the second viewpoint to obtain the coordinates of the predictable landmark point on the second anatomical plane. The coordinates of the predictable landmark on the first anatomical plane and the coordinates on the second anatomical plane are combined to obtain the three-dimensional anatomical coordinates of the predictable landmark.

[0017] Preferably, before the predictable marker detection module, the device further includes a predictable marker detection model training module, comprising: The sample 3D skull model acquisition unit is used to acquire multiple sample 3D skull models. A standard three-dimensional anatomical coordinate generation unit is used to perform predictable landmark three-dimensional anatomical coordinate annotation on each sample three-dimensional skull model to obtain a sample three-dimensional skull model annotated with predictable landmark standard three-dimensional anatomical coordinates, and to perform multi-view two-dimensional view cropping operation on the sample three-dimensional skull model annotated with predictable landmark standard three-dimensional anatomical coordinates to obtain sample two-dimensional views annotated with predictable landmark standard three-dimensional anatomical coordinates from multiple perspectives. A standard two-dimensional view coordinate generation unit is used to determine the standard two-dimensional view coordinates of the predictable marker points based on the standard three-dimensional anatomical coordinates of the predictable marker points for each sample two-dimensional view labeled with the standard three-dimensional anatomical coordinates of the predictable marker points, thereby obtaining a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable marker points. The sample two-dimensional view grouping unit is used to divide all the predictable landmarks into multiple groups according to the anatomical location. Each group of predictable landmarks corresponds to a preset number of sample two-dimensional views labeled with the standard two-dimensional view coordinates of the predictable landmarks. The training unit is used to construct an initial predictable landmark detection model for each group of predictable landmarks, and to train the initial predictable landmark detection model using a sample two-dimensional view group labeled with the standard two-dimensional view coordinates of the predictable landmarks, so as to obtain the corresponding predictable landmark detection model that has completed model training.

[0018] Preferably, the predictable marker points include visible predictable marker points and invisible predictable marker points, and the standard two-dimensional view coordinate generation unit is used for: For each visible and predictable landmark, referencing the standard three-dimensional anatomical coordinates of the visible and predictable landmark, directly annotate the standard two-dimensional view coordinates of the landmark. For each invisible and predictable landmark, the standard two-dimensional view coordinates of the landmark are determined based on the standard three-dimensional anatomical coordinates of the landmark. The method of inversely inferring two-dimensional view coordinates from three-dimensional anatomical coordinates is used.

[0019] Preferably, prior to the predictable marker detection module, the device further includes a transformation matrix construction module, used for: The spatial range of the three-dimensional skull model of the target patient along the first anatomical direction and the spatial range along the second anatomical direction in the three-dimensional anatomical coordinate system are determined, wherein the spatial range of the first anatomical direction is used to characterize the range between the minimum and maximum values ​​of the first anatomical direction, and the spatial range of the second anatomical direction is used to characterize the range between the minimum and maximum values ​​of the second anatomical direction. For each 2D view, the pixel width and pixel height of the 2D view are determined. Based on the spatial range of the first anatomical direction, the spatial range of the second anatomical direction, and the pixel width and pixel height of the 2D view, an affine transformation relationship between the 2D view coordinate system and the 3D anatomical coordinate system of the 2D view is established. Based on the affine transformation relationship, the transformation matrix from the 2D view coordinate system to the 3D anatomical coordinate system of the view is calculated.

[0020] According to another aspect of this application, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform an operation corresponding to the above-described method for detecting three-dimensional cephalometric marker points.

[0021] According to another aspect of this application, a terminal is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described method for detecting three-dimensional cephalometric marker points.

[0022] By employing the above technical solutions, the technical solutions provided in the embodiments of this application have at least the following advantages: This application provides a method, apparatus, medium, and terminal for detecting three-dimensional cephalometric landmarks. First, a three-dimensional skull model of the target patient is acquired, and the model is subjected to multi-view two-dimensional view cropping operations to obtain multiple two-dimensional views, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. Second, the multiple two-dimensional views are input into multiple pre-trained predictable landmark detection models to obtain the two-dimensional coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on a preset number of two-dimensional views, and for each predictable landmark, the coordinates are calculated according to the predicted... The coordinates of the landmark points in the two-dimensional view of the corresponding perspective, and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, are transformed separately to obtain the coordinates of the predictable landmark points on the two anatomical planes. The coordinates on the two anatomical planes are then combined to obtain the three-dimensional anatomical coordinates of the predictable landmark points. Further, for each unpredictable landmark point, the three-dimensional anatomical coordinates of the unpredictable landmark point are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark point. Finally, the three-dimensional anatomical coordinates of all predictable landmark points and all unpredictable landmark points are integrated to obtain the three-dimensional cephalometric landmark point detection results of the target patient. Compared with the prior art, the embodiments of this application realize the transformation from two-dimensional view coordinates to three-dimensional anatomical coordinates based on the transformation matrix from two-dimensional view coordinates to three-dimensional anatomical coordinates. This makes the landmark points no longer limited to the two-dimensional structure of the lateral cephalometric image, avoiding problems such as the overlap of bilateral anatomical structures and image distortion, and improving the accuracy of landmark point detection. Furthermore, in the detection process, different detection methods are used according to different types of landmark points, solving the problem that the overlap of bilateral anatomical structures cannot encompass all landmark point positions, further improving the accuracy of landmark point detection.

[0023] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0024] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1A flowchart of a method for detecting three-dimensional cephalometric marker points provided in an embodiment of this application is shown; Figure 2 This paper presents a flowchart of another method for detecting three-dimensional cephalometric marker points according to an embodiment of this application. Figure 3 This paper shows a block diagram of a detection device for three-dimensional cephalometric marker points provided in an embodiment of this application. Figure 4 A schematic diagram of the structure of a terminal provided in an embodiment of this application is shown. Detailed Implementation

[0025] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0026] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0027] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0028] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0029] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0030] The embodiments of this application can be applied to computer systems / servers that can operate with a wide range of other general-purpose or special-purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations suitable for use with computer systems / servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems, etc.

[0031] Computer systems / servers can be described in the general context of computer system executable instructions (such as program modules) executed by the computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in distributed cloud computing environments, where tasks are performed by remote processing devices linked through a communication network. In distributed cloud computing environments, program modules can reside on local or remote computing system storage media, including storage devices.

[0032] This application provides a method for detecting three-dimensional cephalometric landmarks, such as... Figure 1 As shown, the method includes: 101. Obtain a three-dimensional skull model of the target patient, and perform multi-view two-dimensional view cropping operations on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives.

[0033] The 3D skull model is derived from the target patient's CBCT or CT scan data, obtained through 3D reconstruction, and is in STL format. Multi-view 2D view capture operations can be performed automatically based on VTK automatic rendering screenshot technology, creating an image dataset containing multiple 2D views. These multiple 2D views include a front view, left side view, right side view, bottom view, and cross-sectional view. In this embodiment, the current execution end can be the background computing unit of the Cepha orthodontic cephalometric landmark automatic prediction, reconstruction, and evaluation software.

[0034] 102. Input the two-dimensional views from multiple perspectives into multiple pre-trained predictable landmark detection models to obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. For each predictable landmark, perform coordinate transformation processing based on the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective to obtain the coordinates of the predictable landmark on the two anatomical planes. Combine the coordinates on the two anatomical planes to obtain the three-dimensional anatomical coordinates of the predictable landmark.

[0035] Predictable landmarks are used to characterize landmarks whose 2D coordinates in the corresponding 2D view can be directly predicted using a deep learning model. These include the nasal root point, right orbit point, left orbit point, upper alveolar fossa point, uppermost point of the right external auditory canal, right mandibular angle point, uppermost point of the left external auditory canal, left mandibular angle point, lower alveolar fossa point, anterior chin point, apex of chin, submental point, pituitary fossa point, two upper incisor cusps, and two lower incisor cusps. These 17 predictable landmarks are divided into 7 groups based on their anatomical location, with each group corresponding to a predictable landmark detection model. The preset number of viewing angles is 1, 2, or 3. Specifically, the four incisor cusps are detected using 3 2D views; the pituitary fossa point is detected using 1 2D view; and the remaining predictable landmarks are detected using 2 2D views. The transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each viewpoint is calculated based on the spatial dimensions of the target patient's three-dimensional skull model and the pixel dimensions of the two-dimensional viewpoint at that viewpoint, and thus varies from person to person.

[0036] 103. For each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark.

[0037] Unpredictable markers are used to represent markers whose 2D coordinates in the corresponding 2D view cannot be directly predicted using a deep learning model. These include four root points. Anatomical prior knowledge includes anatomical structure knowledge and statistical laws, such as anatomical structure knowledge like the spatial relationship between root points and cusps, and statistical laws like root length.

[0038] In this embodiment, for the four root points, since they are all located inside the skull and are not visible, it is impossible to directly predict their two-dimensional view coordinates in the corresponding two-dimensional view using a deep learning model. In this case, the three-dimensional anatomical coordinates of the four incisor cusps can be calculated based on the three-dimensional anatomical coordinates of the four incisor cusps directly predicted by the deep learning model in step 102 of the embodiment, and based on the spatial relationship between the root points and the cusps and the statistical law of root length, etc., according to anatomical prior knowledge. Compared with the three-dimensional anatomical coordinates of the root points obtained by hard prediction using a deep learning model, the accuracy is effectively improved.

[0039] 104. Integrate the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient.

[0040] By integrating the three-dimensional anatomical coordinates of the 17 predictable landmarks in step 102 of the embodiment and the three-dimensional anatomical coordinates of the 4 unpredictable landmarks in step 103 of the embodiment, the three-dimensional anatomical coordinates of all 21 landmarks are obtained as the three-dimensional cephalometric landmark detection results of the target patient, and are ready for output.

[0041] Compared with the prior art, the embodiments of this application realize the transformation from two-dimensional view coordinates to three-dimensional anatomical coordinates based on the transformation matrix from two-dimensional view coordinates to three-dimensional anatomical coordinates. This makes the landmark points no longer limited to the two-dimensional structure of the lateral cephalometric image, avoiding problems such as the overlap of bilateral anatomical structures and image distortion, and improving the accuracy of landmark point detection. Furthermore, in the detection process, different detection methods are used according to different types of landmark points, solving the problem that the overlap of bilateral anatomical structures cannot encompass all landmark point positions, further improving the accuracy of landmark point detection.

[0042] This application provides another method for detecting three-dimensional cephalometric landmarks, such as... Figure 2 As shown, the method includes: 201. Construct a predictable landmark detection model that has been trained.

[0043] Accordingly, step 201 of the embodiment specifically includes: acquiring multiple sample three-dimensional skull models; for each sample three-dimensional skull model, performing a predictable landmark three-dimensional anatomical coordinate annotation operation on the sample three-dimensional skull model to obtain a sample three-dimensional skull model annotated with predictable landmark standard three-dimensional anatomical coordinates; and performing a multi-view two-dimensional view cropping operation on the sample three-dimensional skull model annotated with predictable landmark standard three-dimensional anatomical coordinates to obtain sample two-dimensional views annotated with predictable landmark standard three-dimensional anatomical coordinates from multiple perspectives; for each sample two-dimensional view annotated with predictable landmark standard three-dimensional anatomical coordinates, determining the standard three-dimensional anatomical coordinates based on the predictable landmark standard three-dimensional anatomical coordinates. Define the standard two-dimensional view coordinates of the predictable landmarks to obtain sample two-dimensional views labeled with the standard two-dimensional view coordinates of the predictable landmarks; divide all predictable landmarks into multiple groups according to their anatomical locations, with each group of predictable landmarks corresponding to a preset number of sample two-dimensional views labeled with the standard two-dimensional view coordinates of the predictable landmarks; for each group of predictable landmarks, construct the corresponding initial predictable landmark detection model, and use the corresponding sample two-dimensional view group labeled with the standard two-dimensional view coordinates of the predictable landmarks to train the initial predictable landmark detection model, thereby obtaining the corresponding predictable landmark detection model that has completed model training.

[0044] Furthermore, predictable landmarks include visible predictable landmarks and invisible predictable landmarks. Based on the standard three-dimensional anatomical coordinates of the predictable landmarks, the standard two-dimensional view coordinates of the predictable landmarks are determined, resulting in a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable landmarks. This includes: for each visible predictable landmark, referring to the standard three-dimensional anatomical coordinates of the visible predictable landmark, directly labeling the standard two-dimensional view coordinates of the visible predictable landmark; for each invisible predictable landmark, using the method of inverse two-dimensional view coordinate deduction from three-dimensional anatomical coordinates, determining the standard two-dimensional view coordinates of the invisible predictable landmark based on the standard three-dimensional anatomical coordinates of the invisible predictable landmark.

[0045] In this embodiment, firstly, multiple historical patients' three-dimensional skull models can be obtained as sample three-dimensional skull models. The number of samples can be set according to the model accuracy requirements, and all samples are proportionally divided into training samples, validation samples, and test samples. Further, using 3D Slicer software, 17 predictable landmark points are annotated with their three-dimensional anatomical coordinates on each sample three-dimensional skull model. Multi-view two-dimensional view cropping is then performed on the sample three-dimensional skull models annotated with the standard three-dimensional anatomical coordinates of the predictable landmark points to obtain sample two-dimensional views with annotated standard three-dimensional anatomical coordinates from multiple perspectives. These views serve as standard reference answers, reducing annotation errors. Furthermore, since the cusps of the two lower incisors are located internally and are invisible but predictable landmark points, their standard two-dimensional view coordinates cannot be directly annotated based on the aforementioned constructed sample two-dimensional views annotated with the standard three-dimensional anatomical coordinates of the predictable landmark points. Therefore, a method of inversely deducing two-dimensional view coordinates from three-dimensional anatomical coordinates can be used. The standard three-dimensional anatomical coordinates are transformed using a transformation matrix to obtain the standard two-dimensional view coordinates of the landmark points. Specifically, the transformation can be performed using the following formula.

[0046] Where (i, j) represents the standard two-dimensional view coordinates of the marker point, and (R, S, l) represents the standard three-dimensional anatomical coordinates of the marker point. The transformation matrix represents the conversion from a three-dimensional anatomical coordinate system to a two-dimensional view coordinate system. In other words, to convert the standard three-dimensional anatomical coordinates of a landmark point to the standard two-dimensional view coordinates of the landmark point, it is necessary to determine the three-dimensional coordinate values ​​of the origin (0,0), the boundary point (0,1) along the i-axis, and the boundary point (1,0) along the j-axis in the image in the 3D Slicer software. By mapping the two coordinate axes in the three-dimensional coordinate system to the horizontal and vertical coordinates in the two-dimensional view, the transformation matrix of the skull surface can be calculated.

[0047] Of the 17 predictable landmarks, apart from the two lower incisor cusps mentioned above, the remaining 15 landmarks are all visible and predictable. The standard two-dimensional view coordinates of the landmarks can be directly marked by referring to the sample two-dimensional view with standard three-dimensional anatomical coordinates of the predictable landmarks constructed above. Furthermore, the 17 predictable landmarks were divided into 7 groups based on their anatomical location. The first group consisted of the nasal root point and the right orbital point, which are visible in the frontal and right side views; the second group consisted of the left orbital point and the upper alveolar seat point, which are visible in the frontal and left side views; the third group consisted of the uppermost point of the right external auditory canal and the right mandibular angle point, which are visible in the bottom and right side views; the fourth group consisted of the uppermost point of the left external auditory canal and the left mandibular angle point, which are visible in the bottom and left side views; the fifth group consisted of the lower alveolar seat point, the premental point, the apex of the chin, and the submental point, which are visible in the lateral view (including the left or right side view, determined by the mandibular deviation of the target patient) and the frontal view; the sixth group consisted of the pituitary fossa point, which is visible in the sectional view; and the seventh group consisted of the cusps, which are visible in the frontal, left, and right side views. Furthermore, seven initial predictable landmark detection models are constructed, using the HRNet-W32 model structure. Each group of sample 2D views, labeled with the standard 2D coordinates of predictable landmarks, is used to train the corresponding initial predictable landmark detection model. Specifically, for each predictable landmark in each sample 2D view, an independent Gaussian heatmap is generated. The Gaussian heatmaps generated for all predictable landmarks in each sample 2D view are superimposed to form a multi-channel heatmap, which serves as the target for model training. This results in the target Gaussian heatmap for each sample 2D view. The sample 2D view is then input into the initial predictable landmark detection model. After each training cycle, the model outputs predicted Gaussian heatmaps for each sample 2D view at multiple scales. The predicted Gaussian heatmap for each sample 2D view is compared with the target Gaussian heatmap to calculate the loss for each sample 2D view in each cycle. The average of the losses from all sample 2D views is then used to obtain the total loss value for each cycle of model training. The loss function calculation formula is as follows:

[0048] in, This represents the total loss value. This indicates the number of predictable markers. This represents the predicted heatmap of the j-th predictable marker point. The target heatmap for the j-th predictable landmark. The model is trained for 210 epochs per set of images with a learning rate of 0.001 and a batch size of 64 per epoch. After training, the model with the best prediction performance from the 210 weight files is selected as the completed predictable landmark detection model. Finally, 7 completed predictable landmark detection models are fixed for detecting predictable landmarks.

[0049] 202. Obtain a three-dimensional skull model of the target patient, and perform multi-view two-dimensional view cropping operations on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives.

[0050] In this embodiment, CBCT or CT scan data of the target patient are first acquired, and then stacked using professional medical image processing software to reconstruct a three-dimensional volumetric data. Further, based on a preset density threshold, the skull tissue is separated from other tissues within this three-dimensional volumetric data. Further, based on the separated skull tissue, a surface mesh is calculated and generated to enclose the outer surface of the skull, and the generated surface mesh is exported as an STL three-dimensional file to obtain a three-dimensional skull model of the target patient. Finally, multi-view two-dimensional view cropping operations are performed on the three-dimensional skull model to obtain a front view, a left side view, a right side view, a bottom view, and a cross-sectional view.

[0051] 203. Input the two-dimensional views from multiple perspectives into multiple predictable landmark detection models that have completed model training, and obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view.

[0052] Accordingly, step 203 of the embodiment specifically includes: inputting the front view and the right side view into a first predictable landmark detection model that has completed model training, to obtain the two-dimensional view coordinates of the nasal root point in the front view, the two-dimensional view coordinates of the nasal root point in the right side view, the two-dimensional view coordinates of the right eye socket point in the front view, and the two-dimensional view coordinates of the right eye socket point in the right side view; inputting the front view and the left side view into a second predictable landmark detection model that has completed model training, to obtain the two-dimensional view coordinates of the left eye socket point in the front view, the two-dimensional view coordinates of the left eye socket point in the left side view, and the two-dimensional view coordinates of the upper alveolar seat point in the front view. The two-dimensional coordinates of the upper alveolar seat point in the left side view are obtained; the bottom view and right side view are input into the third predictable landmark detection model that has completed model training, to obtain the two-dimensional coordinates of the uppermost point of the right external auditory canal in the bottom view, the two-dimensional coordinates of the uppermost point of the right external auditory canal in the right side view, the two-dimensional coordinates of the right mandibular angle point in the bottom view, and the two-dimensional coordinates of the right mandibular angle point in the right side view; the bottom view and left side view are input into the fourth predictable landmark detection model that has completed model training, to obtain the two-dimensional coordinates of the uppermost point of the left external auditory canal in the bottom view, the two-dimensional coordinates of the uppermost point of the left external auditory canal in the left side view, and the two-dimensional coordinates of the left external auditory canal. The two-dimensional coordinates of the mandibular angle point in the bottom view and the two-dimensional coordinates of the left mandibular angle point in the left side view are obtained. The side view and front view are input into the fifth predictable landmark detection model that has completed model training, resulting in the two-dimensional coordinates of the mandibular alveolar seat point in the front view, the mandibular alveolar seat point in the side view, the anterior chin point in the front view, the anterior chin point in the side view, the chin vertex in the front view, the chin vertex in the side view, the submental point in the front view, and the submental point in the side view. The side view... The figure shows either a left or right lateral view, determined by the mandibular deviation of the target patient. The sectional view is input into the sixth predictive landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the pituitary fossa in the sectional view. The front, left, and right lateral views are input into the seventh predictive landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the cusp in the front view and the fused two-dimensional view coordinates of the cusp in the lateral view. The fused two-dimensional view coordinates of the cusp in the lateral view are obtained by fusing the two-dimensional view coordinates of the cusp in the left lateral view and the two-dimensional view coordinates of the cusp in the right lateral view.

[0053] It should be noted that due to congenital or acquired chewing habits, the human face is often asymmetrical, and facial asymmetry is mostly caused by mandibular asymmetry. Therefore, when the target patient's mandible deviates to the left, four landmarks on the mandible—the inferior alveolar point, premental point, apex of the chin, and submental point—are detected using the left and frontal views; when the target patient's mandible deviates to the right, the same four landmarks are detected using the right and frontal views. Furthermore, since the pituitary fossa is only visible in sectional views, the L-coordinate value of the pituitary fossa is set as the center of the pituitary fossa. The two upper incisor cusps are visible on the surface and are clearly displayed in both the frontal and side views. They are also visible in the left and right side views, with only slight differences in perspective. Therefore, the accuracy of detection is improved by fusing the two-dimensional coordinates of the upper incisor cusps in the left and right side views to obtain a fused two-dimensional coordinate system for the upper incisor cusps in the side view. Similarly, since the two lower incisor cusps are located inside the skull, they may be obscured or unclear in a single side view. Therefore, the accuracy of detection is also improved by fusing the two-dimensional coordinates of the lower incisor cusps in the left and right side views to obtain a fused two-dimensional coordinate system for the lower incisor cusps in the side view.

[0054] 204. Determine the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each viewpoint.

[0055] Accordingly, step 204 of the embodiment specifically includes: determining the spatial range of the target patient's three-dimensional skull model along the first anatomical direction and the spatial range along the second anatomical direction in the three-dimensional anatomical coordinate system, wherein the spatial range along the first anatomical direction is used to characterize the range between the minimum and maximum values ​​in the first anatomical direction, and the spatial range along the second anatomical direction is used to characterize the range between the minimum and maximum values ​​in the second anatomical direction; for each viewpoint's two-dimensional view, determining the pixel width and pixel height of the two-dimensional view, and establishing an affine transformation relationship between the two-dimensional view coordinate system and the three-dimensional anatomical coordinate system based on the spatial range along the first anatomical direction, the spatial range along the second anatomical direction, and the pixel width and pixel height of the two-dimensional view; and calculating the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system based on the affine transformation relationship.

[0056] For example, taking a frontal view, we iterate through all vertices of the 3D skull model to find the minimum and maximum values ​​of the L-coordinate and the S-coordinate, thereby determining the spatial extent of the skull in the L and S directions. Information in the A-direction is temporarily ignored because the frontal view does not provide A coordinates. Further, we binarize the 2D view to find the minimum bounding rectangle of the skull outline, determining the left, right, top, and bottom boundaries of the skull on the image, thus determining its pixel width and pixel height. Further, we establish an affine transformation relationship between the 2D view coordinate system and the 3D anatomical coordinate system. That is, assuming the top-left corner of the 2D view corresponds to the top-left corner of the skull, the top-right corner corresponds to the top-right corner, and the bottom-left corner corresponds to the bottom-left corner, we substitute this affine transformation relationship into the following formula.

[0057] Where L and S represent the sagittal and axial coordinates in the three-dimensional anatomical coordinate system, respectively, and (i, j) represent the two-dimensional view coordinates of the marker point. The transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system is shown below.

[0058]

[0059]

[0060] Substituting the two-dimensional view coordinates and three-dimensional anatomical coordinates of the three sets of corresponding points into the formula, the parameters in the transformation matrix A can be solved, and the image information is transformed onto the anatomical LS plane. The origin of the two-dimensional view and the boundary coordinates of the range in the i and j directions are respectively mapped to the origin of the anatomical coordinate system and the boundary coordinates of the range in the L and S directions, so as to obtain the transformation matrix A from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the front view.

[0061] 205. For each predictable landmark, perform coordinate transformation processing according to the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding viewpoint and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each viewpoint, to obtain the coordinates of the predictable landmark on the two anatomical planes. Combine the coordinates on the two anatomical planes to obtain the three-dimensional anatomical coordinates of the predictable landmark.

[0062] Accordingly, step 205 of the embodiment specifically includes: for each predictable landmark point, obtaining the two-dimensional view coordinates of the predictable landmark point in the first perspective two-dimensional view as the first two-dimensional view coordinates, and the two-dimensional view coordinates in the second perspective two-dimensional view as the second two-dimensional view coordinates; multiplying the first two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the first perspective to obtain the coordinates of the predictable landmark point on the first anatomical plane, and multiplying the second two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the second perspective to obtain the coordinates of the predictable landmark point on the second anatomical plane; merging the coordinates of the predictable landmark point on the first anatomical plane and the coordinates on the second anatomical plane to obtain the three-dimensional anatomical coordinates of the predictable landmark point.

[0063] For example, taking the nasal root point as an example, its two-dimensional coordinates (x1, y1) in the front view and its two-dimensional coordinates (x2, y2) in the right side view are obtained. (x1, y1) is multiplied by the transformation matrix from the two-dimensional coordinate system to the three-dimensional anatomical coordinate system corresponding to the front view to obtain the coordinates of the nasal root point on the first anatomical plane. Similarly, (x2, y2) is multiplied by the transformation matrix from the two-dimensional coordinate system to the three-dimensional anatomical coordinate system corresponding to the right side view to obtain the coordinates of the predictable landmark point on the second anatomical plane. Finally, the coordinates of the nasal root point on the first and second anatomical planes are combined to obtain the three-dimensional anatomical coordinates of the nasal root point.

[0064] It should be noted that for the pituitary fossa, the L coordinate can be preset as the average position of the center of the pituitary fossa. Therefore, its three-dimensional anatomical coordinates can be determined based solely on the cross-sectional view.

[0065] 206. For each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark.

[0066] Step 206 of the embodiment can be referred to in the detailed description of step 103 of the aforementioned embodiment, and will not be repeated here.

[0067] 207. Integrate the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient.

[0068] Step 207 of the embodiment can be referred to in the detailed description of step 104 of the aforementioned embodiment, and will not be repeated here.

[0069] 208. Output the detection results of the three-dimensional head shadow measurement markers.

[0070] Accordingly, step 208 of the embodiment specifically includes: superimposing and rendering the detection results of the three-dimensional cephalometric landmarks onto the three-dimensional skull model in a preset form, and outputting them in the 3D rendering area of ​​the user interface; calling the preset formula library, using the cephalometric analysis formulas in the preset formula library, calculating multiple cephalometric clinical indicators based on the detection results of the three-dimensional cephalometric landmarks, and outputting them in the cephalometric clinical indicator output area of ​​the user interface.

[0071] The preset shape can be a semi-transparent cube. The preset formula library can use the formula library of Peking University's cephalometric analysis method. The clinical indicators for cephalometric analysis include SNA (°), SNB (°), ANB (°), FH-NPo (°), NA-APo (°), FMA (°), SGn-FH (°), MP-SN (°), Po-NB (mm), U1-NA (mm), U1-NA (°), L1-NB (mm), L1-NB (°), U1-L1 (°), U1-SN (°), and IMPA (°).

[0072] This application provides a method for detecting landmark points in three-dimensional cephalometric measurements. First, a three-dimensional skull model of the target patient is acquired, and the model is then subjected to multi-view two-dimensional (2D) view cropping to obtain multiple 2D views, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. Second, these multiple 2D views are input into multiple pre-trained predictable landmark point detection models to obtain the 2D coordinates of each predictable landmark point in the corresponding 2D view. Each predictable landmark point is detected based on a preset number of 2D views, and for each predictable landmark point, the coordinates of the predictable landmark point in the corresponding 2D view are determined. The coordinates of the two-dimensional view in the corresponding two-dimensional view and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each view are transformed to obtain the coordinates of the predictable landmark on the two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. Further, for each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. Finally, the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks are integrated to obtain the three-dimensional cephalometric landmark detection results of the target patient. Compared with the prior art, the embodiments of this application realize the transformation from two-dimensional view coordinates to three-dimensional anatomical coordinates based on the transformation matrix from two-dimensional view coordinates to three-dimensional anatomical coordinates. This makes the landmark points no longer limited to the two-dimensional structure of the lateral cephalometric image, avoiding problems such as the overlap of bilateral anatomical structures and image distortion, and improving the accuracy of landmark point detection. Furthermore, in the detection process, different detection methods are used according to different types of landmark points, solving the problem that the overlap of bilateral anatomical structures cannot encompass all landmark point positions, further improving the accuracy of landmark point detection.

[0073] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this application provides a device for detecting three-dimensional cephalometric landmarks, such as... Figure 3 As shown, the device includes: Two-dimensional view capture module 31, predictable marker detection module 32, unpredictable marker detection module 33, detection result integration module 34; The two-dimensional view cropping module 31 is used to acquire a three-dimensional skull model of the target patient and perform multi-view two-dimensional view cropping operation on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives, wherein the two-dimensional views from multiple perspectives include a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. The predictable landmark detection module 32 is used to input the two-dimensional views of the multiple perspectives into multiple predictable landmark detection models that have completed model training, and obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on the two-dimensional views of the corresponding preset number of perspectives. For each predictable landmark, coordinate transformation is performed according to the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. The unpredictable landmark detection module 33 is used to calculate the three-dimensional anatomical coordinates of each unpredictable landmark based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. The detection result integration module 34 integrates the three-dimensional anatomical coordinates of all predictable landmarks and the three-dimensional anatomical coordinates of all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient.

[0074] In specific application scenarios, after the detection result integration module, the method further includes a detection result output module, used for: The detection results of the three-dimensional head shadow measurement markers are superimposed and rendered on the three-dimensional skull model in a preset form, and output in the 3D rendering area of ​​the user interface. The system calls a preset formula library and uses the cephalometric analysis formulas in the preset formula library to calculate multiple cephalometric clinical indicators based on the detection results of the three-dimensional cephalometric landmarks. These indicators are then output in the cephalometric clinical indicator output area of ​​the user interface.

[0075] In specific application scenarios, the predictable marker detection module is used for: Input the front view and the right side view into the first predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the root of the nose in the front view, the two-dimensional view coordinates of the root of the nose in the right side view, the two-dimensional view coordinates of the right eye socket in the front view, and the two-dimensional view coordinates of the right eye socket in the right side view. Input the front view and the left side view into the second predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the left eye socket in the front view, the two-dimensional view coordinates of the left eye socket in the left side view, the two-dimensional view coordinates of the upper alveolar seat in the front view, and the two-dimensional view coordinates of the upper alveolar seat in the left side view. Input the bottom view and the right side view into the third predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the right side view, the two-dimensional view coordinates of the right mandibular angle point in the bottom view, and the two-dimensional view coordinates of the right mandibular angle point in the right side view. Input the bottom view and the left side view into the fourth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the left side view, the two-dimensional view coordinates of the left mandibular angle point in the bottom view, and the two-dimensional view coordinates of the left mandibular angle point in the left side view. The side view and the front view are input into the fifth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the mandibular alveolar seat in the front view, the two-dimensional view coordinates of the mandibular alveolar seat in the side view, the two-dimensional view coordinates of the anterior chin point in the front view, the two-dimensional view coordinates of the anterior chin point in the side view, the two-dimensional view coordinates of the chin vertex in the front view, the two-dimensional view coordinates of the chin vertex in the side view, the two-dimensional view coordinates of the submental point in the front view, and the two-dimensional view coordinates of the submental point in the side view. The side view is either a left side view or a right side view, determined by the mandibular deviation of the target patient. Input the cross-sectional view into the sixth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the pituitary fossa in the cross-sectional view; The front view, the left side view, and the right side view are input into the seventh predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the tooth cusp in the front view and the fused two-dimensional view coordinates of the tooth cusp in the side view. The fused two-dimensional view coordinates of the tooth cusp in the side view are obtained by fusing the two-dimensional view coordinates of the tooth cusp in the left side view and the two-dimensional view coordinates of the tooth cusp in the right side view.

[0076] In specific application scenarios, the predictable marker detection module is also used for: For each predictable marker point, obtain the two-dimensional view coordinates of the predictable marker point in the first-view two-dimensional view as the first two-dimensional view coordinates, and obtain the two-dimensional view coordinates of the predictable marker point in the second-view two-dimensional view as the second two-dimensional view coordinates. Multiply the first two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the first viewpoint to obtain the coordinates of the predictable landmark point on the first anatomical plane; and multiply the second two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the second viewpoint to obtain the coordinates of the predictable landmark point on the second anatomical plane. The coordinates of the predictable landmark on the first anatomical plane and the coordinates on the second anatomical plane are combined to obtain the three-dimensional anatomical coordinates of the predictable landmark.

[0077] Before the predictable marker detection module in a specific application scenario, the device further includes a predictable marker detection model training module, including: The sample 3D skull model acquisition unit is used to acquire multiple sample 3D skull models. A standard three-dimensional anatomical coordinate generation unit is used to perform predictable landmark three-dimensional anatomical coordinate annotation on each sample three-dimensional skull model to obtain a sample three-dimensional skull model annotated with predictable landmark standard three-dimensional anatomical coordinates, and to perform multi-view two-dimensional view cropping operation on the sample three-dimensional skull model annotated with predictable landmark standard three-dimensional anatomical coordinates to obtain sample two-dimensional views annotated with predictable landmark standard three-dimensional anatomical coordinates from multiple perspectives. A standard two-dimensional view coordinate generation unit is used to determine the standard two-dimensional view coordinates of the predictable marker points based on the standard three-dimensional anatomical coordinates of the predictable marker points for each sample two-dimensional view labeled with the standard three-dimensional anatomical coordinates of the predictable marker points, thereby obtaining a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable marker points. The sample two-dimensional view grouping unit is used to divide all the predictable landmarks into multiple groups according to the anatomical location. Each group of predictable landmarks corresponds to a preset number of sample two-dimensional views labeled with the standard two-dimensional view coordinates of the predictable landmarks. The training unit is used to construct an initial predictable landmark detection model for each group of predictable landmarks, and to train the initial predictable landmark detection model using a sample two-dimensional view group labeled with the standard two-dimensional view coordinates of the predictable landmarks, so as to obtain the corresponding predictable landmark detection model that has completed model training.

[0078] In specific application scenarios, the predictable marker points include both visible and invisible predictable marker points. The standard two-dimensional view coordinate generation unit is used for: For each visible and predictable landmark, referencing the standard three-dimensional anatomical coordinates of the visible and predictable landmark, directly annotate the standard two-dimensional view coordinates of the landmark. For each invisible and predictable landmark, the standard two-dimensional view coordinates of the landmark are determined based on the standard three-dimensional anatomical coordinates of the landmark. The method of inversely inferring two-dimensional view coordinates from three-dimensional anatomical coordinates is used.

[0079] In specific application scenarios, prior to the predictable marker detection module, the device further includes a transformation matrix construction module, used for: The spatial range of the three-dimensional skull model of the target patient along the first anatomical direction and the spatial range along the second anatomical direction in the three-dimensional anatomical coordinate system are determined, wherein the spatial range of the first anatomical direction is used to characterize the range between the minimum and maximum values ​​of the first anatomical direction, and the spatial range of the second anatomical direction is used to characterize the range between the minimum and maximum values ​​of the second anatomical direction. For each 2D view, the pixel width and pixel height of the 2D view are determined. Based on the spatial range of the first anatomical direction, the spatial range of the second anatomical direction, and the pixel width and pixel height of the 2D view, an affine transformation relationship between the 2D view coordinate system and the 3D anatomical coordinate system of the 2D view is established. Based on the affine transformation relationship, the transformation matrix from the 2D view coordinate system to the 3D anatomical coordinate system of the view is calculated.

[0080] This application provides a device for detecting landmark points in three-dimensional cephalometric measurements. First, a three-dimensional skull model of the target patient is acquired, and the model is then subjected to multi-view two-dimensional (2D) view cropping to obtain multiple 2D views, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. Second, these multiple 2D views are input into multiple pre-trained predictable landmark point detection models to obtain the 2D coordinates of each predictable landmark point in the corresponding 2D view. Each predictable landmark point is detected based on a preset number of 2D views, and for each predictable landmark point, the coordinates of the predictable landmark point in the corresponding 2D view are determined. The coordinates of the two-dimensional view in the corresponding two-dimensional view and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each view are transformed to obtain the coordinates of the predictable landmark on the two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. Further, for each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. Finally, the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks are integrated to obtain the three-dimensional cephalometric landmark detection results of the target patient. Compared with the prior art, the embodiments of this application realize the transformation from two-dimensional view coordinates to three-dimensional anatomical coordinates based on the transformation matrix from two-dimensional view coordinates to three-dimensional anatomical coordinates. This makes the landmark points no longer limited to the two-dimensional structure of the lateral cephalometric image, avoiding problems such as the overlap of bilateral anatomical structures and image distortion, and improving the accuracy of landmark point detection. Furthermore, in the detection process, different detection methods are used according to different types of landmark points, solving the problem that the overlap of bilateral anatomical structures cannot encompass all landmark point positions, further improving the accuracy of landmark point detection.

[0081] According to one embodiment of this application, a storage medium is provided, the storage medium storing at least one executable instruction, the computer-executable instruction being able to execute the method for detecting three-dimensional cephalometric marker points in any of the above method embodiments.

[0082] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), and includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods described in the various implementation scenarios of this application.

[0083] Figure 4The diagram shows a structural schematic of a terminal according to one embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the terminal.

[0084] like Figure 4 As shown, the terminal may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0085] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408.

[0086] Communication interface 404 is used to communicate with other network elements such as clients or other servers.

[0087] The processor 402 is used to execute program 410, specifically to execute the relevant steps in the above embodiment of the method for detecting three-dimensional cephalometric marker points.

[0088] Specifically, program 410 may include program code that includes computer operation instructions.

[0089] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0090] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0091] Specifically, program 410 can be used to cause processor 402 to perform the following operations: A three-dimensional skull model of the target patient is obtained, and a multi-view two-dimensional view cropping operation is performed on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. The two-dimensional views from multiple perspectives are input into multiple predictable landmark detection models that have completed model training, respectively, to obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on a preset number of two-dimensional views from different perspectives. For each predictable landmark, coordinate transformation is performed based on the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. For each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. By integrating the three-dimensional anatomical coordinates of all predictable and all unpredictable landmarks, the three-dimensional cephalometric landmark detection results of the target patient are obtained.

[0092] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device for the aforementioned three-dimensional cephalometric marker detection method, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0094] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of this application are not limited to the order specifically described above, unless otherwise specifically stated. Furthermore, in some embodiments, this application may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the methods according to this application. Thus, this application also covers recording media storing programs for performing the methods according to this application.

[0095] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0096] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting three-dimensional cephalometric landmarks, characterized in that, include: A three-dimensional skull model of the target patient is obtained, and a multi-view two-dimensional view cropping operation is performed on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. The two-dimensional views from multiple perspectives are input into multiple predictable landmark detection models that have completed model training, respectively, to obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on a preset number of two-dimensional views from different perspectives. For each predictable landmark, coordinate transformation is performed based on the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. For each unpredictable landmark, the three-dimensional anatomical coordinates of the unpredictable landmark are calculated based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. By integrating the three-dimensional anatomical coordinates of all predictable and all unpredictable landmarks, the three-dimensional cephalometric landmark detection results of the target patient are obtained.

2. The method according to claim 1, characterized in that, After integrating the three-dimensional anatomical coordinates of all predictable and all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient, the method further includes: The detection results of the three-dimensional head shadow measurement markers are superimposed and rendered on the three-dimensional skull model in a preset form, and output in the 3D rendering area of ​​the user interface. The system calls a preset formula library and uses the cephalometric analysis formulas in the preset formula library to calculate multiple cephalometric clinical indicators based on the detection results of the three-dimensional cephalometric landmarks. These indicators are then output in the cephalometric clinical indicator output area of ​​the user interface.

3. The method according to claim 1, characterized in that, The step of inputting the multiple 2D views from various perspectives into multiple pre-trained predictable landmark detection models to obtain the 2D view coordinates of each predictable landmark in the corresponding 2D view includes: Input the front view and the right side view into the first predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the root of the nose in the front view, the two-dimensional view coordinates of the root of the nose in the right side view, the two-dimensional view coordinates of the right eye socket in the front view, and the two-dimensional view coordinates of the right eye socket in the right side view. Input the front view and the left side view into the second predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the left eye socket in the front view, the two-dimensional view coordinates of the left eye socket in the left side view, the two-dimensional view coordinates of the upper alveolar seat in the front view, and the two-dimensional view coordinates of the upper alveolar seat in the left side view. Input the bottom view and the right side view into the third predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the right external auditory canal in the right side view, the two-dimensional view coordinates of the right mandibular angle point in the bottom view, and the two-dimensional view coordinates of the right mandibular angle point in the right side view. Input the bottom view and the left side view into the fourth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the bottom view, the two-dimensional view coordinates of the uppermost point of the left external auditory canal in the left side view, the two-dimensional view coordinates of the left mandibular angle point in the bottom view, and the two-dimensional view coordinates of the left mandibular angle point in the left side view. The side view and the front view are input into the fifth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the mandibular alveolar seat in the front view, the two-dimensional view coordinates of the mandibular alveolar seat in the side view, the two-dimensional view coordinates of the anterior chin point in the front view, the two-dimensional view coordinates of the anterior chin point in the side view, the two-dimensional view coordinates of the chin vertex in the front view, the two-dimensional view coordinates of the chin vertex in the side view, the two-dimensional view coordinates of the submental point in the front view, and the two-dimensional view coordinates of the submental point in the side view. The side view is either a left side view or a right side view, determined by the mandibular deviation of the target patient. Input the cross-sectional view into the sixth predictable landmark detection model that has completed model training to obtain the two-dimensional view coordinates of the pituitary fossa in the cross-sectional view; The front view, the left side view, and the right side view are input into the seventh predictable marker detection model that has completed model training to obtain the two-dimensional view coordinates of the tooth cusp in the front view and the fused two-dimensional view coordinates of the tooth cusp in the side view. The fused two-dimensional view coordinates of the tooth cusp in the side view are obtained by fusing the two-dimensional view coordinates of the tooth cusp in the left side view and the two-dimensional view coordinates of the tooth cusp in the right side view.

4. The method according to claim 1, characterized in that, For each predictable landmark, coordinate transformation is performed based on its two-dimensional view coordinates in the corresponding viewpoint and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system for each viewpoint. This yields the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are then merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. This process includes: For each predictable marker point, obtain the two-dimensional view coordinates of the predictable marker point in the first-view two-dimensional view as the first two-dimensional view coordinates, and obtain the two-dimensional view coordinates of the predictable marker point in the second-view two-dimensional view as the second two-dimensional view coordinates. Multiply the first two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the first viewpoint to obtain the coordinates of the predictable landmark point on the first anatomical plane; and multiply the second two-dimensional view coordinates by the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to the second viewpoint to obtain the coordinates of the predictable landmark point on the second anatomical plane. The coordinates of the predictable landmark on the first anatomical plane and the coordinates on the second anatomical plane are combined to obtain the three-dimensional anatomical coordinates of the predictable landmark.

5. The method according to claim 1, characterized in that, Before inputting the multiple 2D views from various perspectives into multiple pre-trained predictable landmark detection models to obtain the 2D view coordinates of each predictable landmark in the corresponding 2D view, the method further includes: Obtain multiple sample 3D skull models; For each sample three-dimensional skull model, a predictable landmark three-dimensional anatomical coordinate annotation operation is performed on the sample three-dimensional skull model to obtain a sample three-dimensional skull model annotated with the predictable landmark standard three-dimensional anatomical coordinates. Then, a multi-view two-dimensional view cropping operation is performed on the sample three-dimensional skull model annotated with the predictable landmark standard three-dimensional anatomical coordinates to obtain sample two-dimensional views annotated with the predictable landmark standard three-dimensional anatomical coordinates from multiple perspectives. For each sample two-dimensional view labeled with the standard three-dimensional anatomical coordinates of the predictable landmarks, the standard two-dimensional view coordinates of the predictable landmarks are determined based on the standard three-dimensional anatomical coordinates of the predictable landmarks, thus obtaining a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable landmarks. All the predictable landmarks are divided into multiple groups according to their anatomical location. Each group of predictable landmarks corresponds to a preset number of sample two-dimensional views labeled with the standard two-dimensional view coordinates of the predictable landmarks from different perspectives. For each set of predictable marker points, an initial predictable marker point detection model is constructed, and the initial predictable marker point detection model is trained using a sample two-dimensional view group labeled with the standard two-dimensional view coordinates of the predictable marker points, to obtain the corresponding predictable marker point detection model that has completed model training.

6. The method according to claim 5, characterized in that, The predictable marker points include visible predictable marker points and invisible predictable marker points. The step of determining the standard two-dimensional view coordinates of the predictable marker points based on their standard three-dimensional anatomical coordinates, to obtain a sample two-dimensional view labeled with the standard two-dimensional view coordinates of the predictable marker points, includes: For each visible and predictable landmark, referencing the standard three-dimensional anatomical coordinates of the visible and predictable landmark, directly annotate the standard two-dimensional view coordinates of the landmark. For each invisible and predictable landmark, the standard two-dimensional view coordinates of the landmark are determined based on the standard three-dimensional anatomical coordinates of the landmark. The method of inversely inferring two-dimensional view coordinates from three-dimensional anatomical coordinates is used.

7. The method according to claim 1, characterized in that, Before performing coordinate transformation processing on each predictable landmark point according to its two-dimensional view coordinates in the corresponding viewpoint and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system for each viewpoint to obtain the coordinates of the predictable landmark point on the two anatomical planes, the method further includes: The spatial range of the three-dimensional skull model of the target patient along the first anatomical direction and the spatial range along the second anatomical direction in the three-dimensional anatomical coordinate system are determined, wherein the spatial range of the first anatomical direction is used to characterize the range between the minimum and maximum values ​​of the first anatomical direction, and the spatial range of the second anatomical direction is used to characterize the range between the minimum and maximum values ​​of the second anatomical direction. For each 2D view, the pixel width and pixel height of the 2D view are determined. Based on the spatial range of the first anatomical direction, the spatial range of the second anatomical direction, and the pixel width and pixel height of the 2D view, an affine transformation relationship between the 2D view coordinate system and the 3D anatomical coordinate system of the 2D view is established. Based on the affine transformation relationship, the transformation matrix from the 2D view coordinate system to the 3D anatomical coordinate system of the view is calculated.

8. A device for detecting three-dimensional cephalometric marker points, characterized in that, include: The two-dimensional view cropping module is used to acquire a three-dimensional skull model of the target patient and perform multi-view two-dimensional view cropping operations on the three-dimensional skull model to obtain two-dimensional views from multiple perspectives, including a front view, a left side view, a right side view, a bottom view, and a cross-sectional view. The predictable landmark detection module is used to input the two-dimensional views of the multiple perspectives into multiple predictable landmark detection models that have completed model training, and obtain the two-dimensional view coordinates of each predictable landmark in the corresponding two-dimensional view. Each predictable landmark is detected based on the two-dimensional views of the corresponding preset number of perspectives. For each predictable landmark, coordinate transformation is performed according to the two-dimensional view coordinates of the predictable landmark in the two-dimensional view of the corresponding perspective and the transformation matrix from the two-dimensional view coordinate system to the three-dimensional anatomical coordinate system corresponding to each perspective, to obtain the coordinates of the predictable landmark on two anatomical planes. The coordinates on the two anatomical planes are merged to obtain the three-dimensional anatomical coordinates of the predictable landmark. An unpredictable landmark detection module is used to calculate the three-dimensional anatomical coordinates of each unpredictable landmark based on prior anatomical knowledge and the three-dimensional anatomical coordinates of each predictable landmark. The detection result integration module integrates the three-dimensional anatomical coordinates of all predictable landmarks and all unpredictable landmarks to obtain the three-dimensional cephalometric landmark detection results of the target patient.

9. A storage medium storing at least one executable instruction, characterized in that, The executable instructions cause the processor to perform the operation corresponding to the method for detecting three-dimensional cephalometric marker points as described in any one of claims 1-7.

10. A terminal, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, characterized in that the executable instruction causes the processor to perform the operation corresponding to the three-dimensional cephalometric marker detection method as described in any one of claims 1-7.

Images