Methods, devices, equipment, and media for orthodontic analysis based on multi-scale convolution.

Abstract

This application relates to the field of dental point cloud data processing technology, and discloses a method, apparatus, device, and storage medium for orthodontic analysis based on multi-scale convolution. The method includes: extracting tooth pose features and tooth geometric shape features from pre-acquired dental point cloud data using dynamic point cloud image convolution; performing multi-channel image convolution on the tooth pose features and the tooth geometric shape features to obtain multi-scale features of the dental point cloud data; and generating the target pose of the multi-scale features using a pre-defined multilayer perceptron. Through the implementation of this application, the accuracy of target pose analysis in orthodontics can be effectively improved by using dynamic point cloud image convolution and multilayer perceptron techniques.

Description

Technical Field

[0001] This application relates to the field of dental point cloud data processing technology, and in particular to a method, apparatus, device and storage medium for orthodontic analysis of teeth based on multi-scale convolution. Background Technology

[0002] Existing methods for predicting the target position of teeth after orthodontic treatment can be mainly divided into the following three categories:

[0003] 1) Orthodontic target location prediction based on expert rules: This method usually requires a lot of manual intervention from doctors and usually requires some prior knowledge such as key points of teeth. However, this method requires a lot of pre-marked key point information of teeth and relies too much on prior knowledge, resulting in certain accuracy problems in target location analysis.

[0004] 2) Orthodontic target position prediction based on generative models: This method is mainly for 2D images. Given a 2D face image before orthodontic treatment, it generates a 2D face image to predict after orthodontic treatment. Some methods are also used for 3D models and do not require pre-segmentation of teeth. The desired target position is achieved by editing the reconstructed 3D model. However, the accuracy of target position analysis is uncontrollable when this method is used for 3D processing.

[0005] 3) Orthodontic target position prediction based on the rotation and translation transformation matrix of each tooth. However, this method requires training with paired tooth models before and after orthodontics, and it is difficult to handle some cases of severe dental malocclusion, which will ultimately result in inaccurate orthodontic target position analysis. Summary of the Invention

[0006] This application provides a method, apparatus, device, and medium for orthodontic analysis based on multi-scale convolution, the main purpose of which is to solve the problem of insufficient accuracy in target position analysis in orthodontics.

[0007] To achieve the above objectives, this application provides a method for orthodontic analysis based on multi-scale convolution, comprising:

[0008] We use point cloud dynamic graph convolution to extract tooth pose features and tooth geometric features from pre-acquired tooth point cloud data.

[0009] Multi-channel image convolution is performed on tooth pose features and tooth geometric features to obtain multi-scale features of tooth point cloud data.

[0010] The target pose with multi-scale features is generated using a pre-defined multilayer perceptron.

[0011] To address the aforementioned problems, this application also provides a dental orthodontic analysis device based on multi-scale convolution, comprising:

[0012] The feature extraction module is used to extract tooth pose features and tooth geometric shape features from pre-acquired tooth point cloud data using point cloud dynamic graph convolution.

[0013] The multi-channel analysis module is used to perform multi-channel image convolution on tooth pose features and tooth geometric features to obtain multi-scale features of tooth point cloud data.

[0014] The multilayer sensing module is used to generate target poses with multi-scale features using a pre-set multilayer perceptron.

[0015] To address the aforementioned problems, this application also provides an apparatus comprising:

[0016] At least one processor; and,

[0017] A memory that is communicatively connected to at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor can perform the above-described orthodontic analysis method based on multi-scale convolution.

[0019] To address the aforementioned issues, this application also provides a storage medium storing at least one computer program, which is executed by a processor in a device to implement the aforementioned multi-scale convolution-based orthodontic analysis method.

[0020] This application's embodiments extract tooth pose and geometric features from pre-acquired tooth point cloud data through point cloud dynamic graph convolution, avoiding data denoising and other operations in common methods, thus improving data analysis efficiency. Simultaneously, multi-channel graph convolution is used to analyze and process the tooth pose and geometric features to obtain multi-scale features of each tooth in the tooth point cloud data, providing more accurate data support for subsequent analysis. Furthermore, multi-layer perceptron processing is used to further enhance the accuracy of the final analyzed target pose. Therefore, the orthodontic analysis method, apparatus, device, and medium based on multi-scale convolution proposed in this application can solve the problem of insufficient accuracy in target position analysis during orthodontic treatment. Attached Figure Description

[0021] Figure 1 A flowchart illustrating a multi-scale convolution-based orthodontic analysis method provided in an embodiment of this application;

[0022] Figure 2 This is a schematic diagram of the structure of a three-head DGCNN architecture provided in an embodiment of this application;

[0023] Figure 3-1A schematic diagram of a global adjacency matrix provided in an embodiment of this application;

[0024] Figure 3-2 A schematic diagram of a local adjacency matrix provided in an embodiment of this application;

[0025] Figure 3-3 A schematic diagram of an intersecting adjacency matrix provided in an embodiment of this application;

[0026] Figure 4 This is a schematic diagram of the network structure of a multilayer perceptron provided in an embodiment of this application;

[0027] Figure 5 A functional block diagram of a multi-scale convolution-based orthodontic analysis device provided in an embodiment of this application;

[0028] Figure 6 This is a schematic diagram of the structure of an apparatus for implementing a multi-scale convolution-based orthodontic analysis method for teeth, provided in an embodiment of this application.

[0029] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0030] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0031] Existing methods for predicting the target position of teeth after orthodontic treatment mainly include expert rule-based prediction, generative model-based prediction, and prediction based on regression of the rotation and translation transformation matrix of each tooth. However, all of these methods suffer from insufficient accuracy in the target position analysis. An embodiment of the present invention provides a multi-scale convolution-based orthodontic analysis method, which is an effective way to improve the accuracy of target position analysis in orthodontics by using point cloud dynamic map convolution and multilayer perceptron.

[0032] Reference Figure 1 The diagram shown is a flowchart illustrating a multi-scale convolution-based orthodontic analysis method according to an embodiment of this application. In this embodiment, the multi-scale convolution-based orthodontic analysis method includes:

[0033] S1. Extract tooth pose features and tooth geometric features from pre-acquired tooth point cloud data using point cloud dynamic graph convolution.

[0034] In this embodiment of the application, the tooth point cloud data includes point cloud data of the teeth in the local coordinate system, i.e., the world coordinate system, collected by a preset sensor in the user's oral cavity.

[0035] In detail, because the data directly collected by the sensor contains noise, the data is somewhat unstable. Therefore, before processing such point cloud data, commonly used convolutional networks need to perform operations such as denoising the data or reconstructing the network itself. This method will result in low data processing efficiency and high processing cost.

[0036] Therefore, this application embodiment uses point cloud dynamic convolution to directly extract tooth pose features and tooth geometric features from pre-acquired tooth point cloud data, avoiding data denoising operations and improving data analysis efficiency.

[0037] In this embodiment, tooth pose features and tooth geometric features are extracted from pre-acquired tooth point cloud data using point cloud dynamic graph convolution, including: extracting the local coordinate system and world coordinate system of each tooth from the tooth point cloud data; calculating the tooth pose features of each tooth based on the local coordinate system, world coordinate system and preset rotation matrix; and performing edge convolution on the tooth point cloud data using a pre-trained deep autoencoder to obtain the geometric features of each tooth.

[0038] In detail, such as Figure 2 As shown, this application can utilize a pre-acquired three-head DGCNN architecture to extract the local coordinates of each tooth in three-dimensional space from tooth point cloud data. Figure 2 In the process, the three-head DGCNN architecture identifies the local coordinates of each tooth in three-dimensional space from the x, y, and z dimensions respectively.

[0039] In this embodiment, the tooth pose is a representation of the tooth's position and orientation in its local coordinate system relative to the world coordinate system, described in three-dimensional space by rotation and translation. The step of calculating the tooth pose features of each tooth based on the local coordinate system, the world coordinate system, and a preset rotation matrix includes:

[0040] The transformation of the coordinate system is represented by the following formula, which is then used to calculate the tooth pose characteristics of each tooth:

[0041]

[0042] in, Let R be the tooth pose feature of the i-th tooth in the world coordinate system. -1 Represents the rotation matrix. Let C be the local coordinates of the i-th tooth in the local coordinate system, and let C be the centroid of the i-th tooth.

[0043] Therefore, the pose of a tooth can be represented by a quaternion four-dimensional vector of the rotation matrix and a three-dimensional vector of the tooth centroid.

[0044] Furthermore, in the process of analyzing the target position of teeth for orthodontic treatment, in order to improve the accuracy of the final analysis results, it is necessary to consider not only the tooth position characteristics, but also the geometric data of teeth such as shape and size, surface texture and occlusal groove.

[0045] Therefore, in order to effectively extract the geometric features of the tooth surface, this application utilizes a pre-trained deep autoencoder to analyze tooth point cloud data. The deep autoencoder is used to compress and encode the geometric information of each tooth in the tooth point cloud data.

[0046] In this embodiment of the application, a pre-trained deep autoencoder is used to perform edge convolution on tooth point cloud data to obtain the geometric shape features of each tooth. This includes: using a pre-trained deep autoencoder to perform convolutional encoding on tooth point cloud data to obtain encoded data; using a mapping matrix within the deep autoencoder to perform latent vector mapping on the encoded data to obtain a mapping matrix; performing max pooling on the mapping matrix; and using the attention mechanism within the deep autoencoder to calculate the max pooled mapping matrix to obtain the geometric shape features of each tooth.

[0047] In detail, using a pre-trained deep autoencoder to convert tooth point cloud data into bit-coded data can improve the efficiency of subsequent extraction of geometric features of each tooth.

[0048] Specifically, the mapping matrix is ​​a pre-defined weight matrix or bias matrix within the deep autoencoder, which can be used to map the encoded data of the current layer within the deep autoencoder into latent vectors, and then use them as inputs for subsequent layers. Through the mapping of multiple latent vectors and max pooling, the geometric features of each tooth in the tooth point cloud data can be roughly extracted.

[0049] Furthermore, by performing attention calculations on the max-pooled mapping matrix through the attention mechanism within the deep autoencoder, refined geometric features of each tooth within the tooth point cloud data can be generated.

[0050] S2. Perform multi-channel image convolution on the tooth pose features and the tooth geometric shape features to obtain multi-scale features of the tooth point cloud data.

[0051] In this embodiment of the application, in order to achieve the final analysis of the target pose of orthodontics, it is necessary to further extract features from the tooth pose features and tooth geometric features to obtain multi-scale features of each tooth in the tooth point cloud data, thereby providing more accurate data support for subsequent analysis.

[0052] In this embodiment, multi-channel graph convolution is performed on tooth pose features and tooth geometric shape features to obtain multi-scale features of tooth point cloud data. This includes: constructing a global adjacency matrix, a local adjacency matrix, and an intersecting adjacency matrix for each tooth based on the tooth pose features and tooth geometric shape features; performing graph convolution on the global adjacency matrix, local adjacency matrix, and intersecting adjacency matrix of each tooth using a preset graph convolutional neural network to obtain a multi-channel convolutional graph; and concatenating the multi-channel convolutional graphs corresponding to the global adjacency matrix, local adjacency matrix, and intersecting adjacency matrix of each tooth to obtain the multi-scale features of tooth point cloud data.

[0053] In detail, refer to Figure 3-1 The diagram shown is a schematic of the global adjacency matrix in this application. The schematic of the global adjacency matrix is ​​used to represent a fully connected graph composed of teeth in the tooth point cloud data. The nodes represented by each tooth in the graph are connected to each other, so that the overall shape information of the teeth can be extracted from the global adjacency matrix.

[0054] In detail, refer to Figure 3-2 The diagram shown is a schematic diagram of the local adjacency matrix in this application. The schematic diagram of the local adjacency matrix is ​​used to show that each tooth not only has a connection relationship with the opposite tooth, but also has a connection relationship with its own adjacent teeth, as well as with the teeth adjacent to the opposite tooth. This allows information on the local crowding or sparseness of the tooth row to be extracted from the local adjacency matrix.

[0055] In detail, refer to Figure 3-3 The diagram shown is a schematic diagram of the intersecting adjacency matrix in this application. The schematic diagram of the intersecting adjacency matrix is ​​used to show that each tooth is connected to the opposite tooth and the adjacent tooth, which can reflect the collision relationship between the teeth.

[0056] In this embodiment, a preset Graph Convolutional Network (GCN) is used to perform graph convolution on the global adjacency matrix, local adjacency matrix, and intersecting adjacency matrix of each tooth to obtain the multi-channel convolutional graphs corresponding to the global adjacency matrix, local adjacency matrix, and intersecting adjacency matrix respectively.

[0057] Specifically, the multi-channel convolutional maps corresponding to the obtained global adjacency matrix, local adjacency matrix, and intersecting adjacency matrix are stitched together to obtain multi-scale features of the tooth point cloud data. When stitching the multi-channel convolutional maps, a feature pyramid structure can be used to stitch the multi-channel convolutional maps to achieve fast and accurate stitching of different convolutional images.

[0058] S3. Generate the target pose of the multi-scale features using a preset multilayer perceptron.

[0059] In the embodiments of the present application, in order to more accurately analyze the target pose of each tooth in the tooth point cloud data for orthodontics, a preset multi-layer perceptron is used to process multi-scale features, so as to improve the accuracy of the finally analyzed target pose.

[0060] In the embodiments of the present application, generating the target pose of multi-scale features by using a preset multi-layer perceptron includes: performing convolution on the multi-scale features for a preset number of rounds by using a pre-acquired shared convolution kernel to obtain convolution features; performing a global information embedding operation on the convolution features to obtain embedding features; performing an adaptive recalibration on the embedding features to obtain calibrated features; and generating the target pose of the multi-scale features according to the calibrated features.

[0061] Refer Figure 4 as shown Figure 4 is a schematic diagram of the network structure of the multi-layer perceptron. After multi-channel graph convolution obtains multi-scale features of tooth point cloud data, the multi-scale features pass through the multi-level network of the multi-layer perceptron and are calculated by using the same shared convolution kernel as a parameter, and finally the target pose of the multi-scale features is output.

[0062] Specifically, the shared convolution kernel is a pre-set convolution kernel matrix used for performing dot product operations with all data in the multi-scale features. By performing convolution operations on the multi-scale features for a preset number of rounds by using the shared convolution kernel, the convolution features corresponding to the multi-scale features can be obtained.

[0063] In the embodiments of the present application, using the shared convolution kernel (that is, adopting SharedMLP) to perform convolution on multi-scale features is different from the traditional MLP (Multi-Layer Perceptron) processing method. In the traditional MLP processing method, each data has an independent MLP network, and each data has an independent weight and bias matrix, which will lead to the bloatedness of the network structure and the low computational efficiency.

[0064] The shared convolution kernel adopted in the present application (that is, adopting SharedMLP) realizes the representation of features by sharing the same convolution kernel, that is, all data in the multi-scale features perform dot product operations with the shared convolution kernel to achieve parameter sharing. This sharing greatly reduces the number of parameters, enables the network of the multi-layer perceptron to have better generalization ability, and improves the efficiency of analyzing multi-scale features.

[0065] Furthermore, since the method of multi-channel graph convolution is used when analyzing the above convolution features, it may cause the problem of channel dependence. In order to solve the problem of channel dependence existing when analyzing and processing the convolution features subsequently, refer Figure 4As shown, the multilayer perceptron of this application uses an SE block (Squenze-Excitation block). This SE block can compress global information in tooth point cloud data into a preset channel descriptor. That is, through global average pooling of the channels, the feature data containing global information is compressed into a feature vector, so that the generated vector is embedded into the convolutional feature to obtain the embedded feature, thereby alleviating the channel dependency.

[0066] In this embodiment of the application, the reference Figure 4 As shown, after the SE block performs a global information embedding operation on the convolutional features to generate embedded features, it still needs to perform an excitation (adaptive recalibration) operation to fully capture channel dependencies.

[0067] In this embodiment of the application, adaptive recalibration of the embedded features to obtain calibrated features includes: performing fully connected processing on the embedded features using a preset first activation function to obtain an intermediate dataset; performing fully connected processing on the intermediate dataset using a preset second activation function to obtain adaptive weights; and performing weighted calculation on the embedded features using the adaptive weights to obtain calibrated features.

[0068] In detail, the first activation function includes the ReLU activation function, and the second activation function includes the Sigmoid activation function.

[0069] Specifically, this embodiment uses two fully connected layers to process the embedded features. The first fully connected layer uses the ReLU activation function to compress the embedded features of multiple channels, thereby reducing the amount of data for analysis and computation. The second fully connected layer uses the Sigmoid activation function to activate the output data of the first fully connected layer, obtaining weights for characterizing the features of different channels. Then, through weighted processing of the weights, adaptive correction of the embedded features is achieved.

[0070] In this embodiment of the application, a residual network for predicting and analyzing the target pose of each tooth can be added after the SE block structure to achieve the prediction of the target pose of each tooth.

[0071] In detail, the residual network can be trained using the following loss function:

[0072] Loss = Loss rotation +Loss translation

[0073] Among them, the Loss value is the training loss of the residual network. rotation Loss is the cosine similarity between the predicted orthodontic tooth pose in the pre-acquired training data and the corresponding real orthodontic tooth pose in the training data. translationThis is the distance between the predicted centroid of the teeth after orthodontic treatment in the training data and the corresponding actual centroid of the teeth after orthodontic treatment in the training data.

[0074] In this embodiment, the trained residual network can be used to calculate the correction features to obtain the target pose of each tooth in the multi-scale features.

[0075] like Figure 5 The diagram shown is a functional block diagram of a multi-scale convolution-based orthodontic analysis device provided in an embodiment of this application.

[0076] The multi-scale convolution-based orthodontic analysis device 500 of this application can be installed in a device. Depending on the functions implemented, the multi-scale convolution-based orthodontic analysis device 500 may include a feature extraction module 501, a multi-channel analysis module 502, and a multi-layer sensing module 503. In this application, a module can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the device processor and perform a fixed function, and are stored in the device's memory.

[0077] In this embodiment, the functions of each module / unit are as follows:

[0078] The feature extraction module 501 is used to extract tooth pose features and tooth geometric shape features from pre-acquired tooth point cloud data using point cloud dynamic graph convolution.

[0079] The multi-channel analysis module 502 is used to perform multi-channel image convolution on tooth pose features and tooth geometric shape features to obtain multi-scale features of tooth point cloud data.

[0080] The multilayer sensing module 503 is used to generate target poses with multi-scale features using a preset multilayer perceptron.

[0081] In detail, each module in the multi-scale convolution-based orthodontic analysis device 500 in this application embodiment adopts the same approach as described above during use. Figures 1 to 4 The technique is the same as that used in the multi-scale convolution-based orthodontic analysis method, and it can produce the same technical effect, so it will not be elaborated here.

[0082] like Figure 6 The diagram shown is a structural schematic of a device for implementing a multi-scale convolution-based orthodontic analysis method for teeth, according to an embodiment of this application.

[0083] The device may include a processor 601, a memory 602, a communication bus 603, and a communication interface 604. It may also include a computer program stored in the memory 602 and capable of running on the processor 601, such as a dental orthodontic analysis program based on multi-scale convolution.

[0084] In some embodiments, the processor 601 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 601 is the control unit of the device, connecting various components of the device through various interfaces and lines. It executes programs or modules stored in the memory 602 (e.g., executing a multi-scale convolution-based orthodontic analysis program) and calls data stored in the memory 602 to perform various functions of the device and process data.

[0085] The memory 602 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 602 can be an internal storage unit of the device, such as the device's portable hard drive. In other embodiments, the memory 602 can also be an external storage device of the device, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 602 can include both internal and external storage units of the device. The memory 602 can be used not only to store application software and various types of data installed on the device, such as code for a multi-scale convolution-based orthodontic analysis program, but also to temporarily store data that has been output or will be output.

[0086] The communication bus 603 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to implement communication between the memory 602 and at least one processor 601.

[0087] Communication interface 604 is used for communication between the aforementioned device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the device and other devices. The user interface may be a display, an input unit (such as a keyboard), or optionally, a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed within the device and to display a visual user interface.

[0088] The figure only shows the device with components. Those skilled in the art will understand that the structure shown in the figure does not constitute a limitation on the device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0089] For example, although not shown, the device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to at least one processor 601 via a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power sources, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0090] It should be understood that the embodiments are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0091] Specifically, the processor 601's specific implementation method of the above instructions can be found in the description of the relevant steps in the corresponding embodiments of the accompanying drawings, and will not be repeated here.

[0092] Furthermore, if the modules / units integrated into the device are implemented as software functional units and sold or used as independent products, they can be stored in a storage medium. The storage medium can be volatile or non-volatile. For example, computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0093] This application also provides a storage medium storing a computer program. When executed by a device's processor, the computer program can implement the multi-scale convolution-based orthodontic analysis method of any of the above embodiments. It should be noted that the storage medium can be volatile or non-volatile. For example, the storage medium may include: any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0094] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0095] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0096] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0097] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application.

[0098] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects. The scope of the invention is not limited to the foregoing description, and all variations within the meaning and scope of equivalents falling within the protection scope are intended to be included in the invention.

[0099] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0100] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A method for orthodontic analysis based on multi-scale convolution, characterized in that, The method includes: We use point cloud dynamic graph convolution to extract tooth pose features and tooth geometric features from pre-acquired tooth point cloud data. Multi-channel image convolution is performed on the tooth pose features and the tooth geometric shape features to obtain multi-scale features of the tooth point cloud data. The target pose of the multi-scale features is generated using a pre-defined multilayer perceptron. The multi-channel image convolution of the tooth pose features and the tooth geometric shape features yields multi-scale features of the tooth point cloud data, including: Based on the tooth pose features and tooth geometry, a global adjacency matrix, a local adjacency matrix, and an intersecting adjacency matrix are constructed for each tooth. The global adjacency matrix represents a fully connected graph composed of teeth from the tooth point cloud data, and the overall shape information of the teeth can be extracted based on this matrix. The local adjacency matrix indicates the connections between each tooth and its opposite tooth, its own adjacent teeth, and its adjacent teeth, and the local crowding or sparseness information of the tooth row can be extracted based on this matrix. The intersecting adjacency matrix indicates that each tooth is connected to both its opposite and adjacent teeth, indicating the collision relationships between teeth. A pre-defined graph convolutional neural network is used to perform graph convolution on the global adjacency matrix, the local adjacency matrix, and the intersecting adjacency matrix of each tooth to obtain a multi-channel convolutional graph. The multi-channel convolutional maps corresponding to the global adjacency matrix, the local adjacency matrix, and the intersecting adjacency matrix of each tooth are concatenated to obtain the multi-scale features of the tooth point cloud data.

2. The orthodontic analysis method based on multi-scale convolution as described in claim 1, characterized in that, The method of extracting tooth pose and geometric features from pre-acquired tooth point cloud data using point cloud dynamic graph convolution includes: The local coordinate system and world coordinate system of each tooth are extracted from the tooth point cloud data; The tooth pose features of each tooth are calculated based on the local coordinate system, the world coordinate system, and the preset rotation matrix. The tooth point cloud data is subjected to edge convolution using a pre-trained deep autoencoder to obtain the geometric shape features of each tooth.

3. The orthodontic analysis method based on multi-scale convolution as described in claim 2, characterized in that, The step of calculating the tooth pose features of each tooth based on the local coordinate system, the world coordinate system, and a preset rotation matrix includes: The tooth pose characteristics of each tooth are calculated using the following formula: in, For the first The tooth pose characteristics of a tooth in the world coordinate system, R -1 For the preset rotation matrix, For the first The local coordinates of a tooth in a local coordinate system. For the first The center of mass of a tooth.

4. The orthodontic analysis method based on multi-scale convolution as described in claim 2, characterized in that, The step of performing edge convolution on the tooth point cloud data using a pre-trained deep autoencoder to obtain the geometric shape features of each tooth includes: The tooth point cloud data is convolutionally encoded using a pre-trained deep autoencoder to obtain encoded data; The encoded data is mapped using the mapping matrix within the deep autoencoder to obtain the mapping matrix; Max pooling is performed on the mapping matrix, and the attention mechanism in the deep autoencoder is used to calculate the max pooled mapping matrix to obtain the geometric shape features of each tooth.

5. The orthodontic analysis method based on multi-scale convolution as described in claim 1, characterized in that, The generation of the target pose using a preset multilayer perceptron with multiscale features includes: The multi-scale features are convolved for a preset number of rounds using pre-acquired shared convolution kernels to obtain convolutional features; Perform a global information embedding operation on the convolutional features to obtain embedded features; The embedded features are adaptively recalibrated to obtain the corrected features; The target pose of the multi-scale features is generated based on the correction features.

6. The orthodontic analysis method based on multi-scale convolution as described in claim 5, characterized in that, The adaptive recalibration of the embedded features to obtain the calibrated features includes: The embedded features are processed using a preset first activation function to obtain an intermediate dataset; The intermediate dataset is fully connected using a preset second activation function to obtain adaptive weights. The embedded features are weighted using the adaptive weights to obtain the corrected features.

7. A dental orthodontic analysis device based on multi-scale convolution, characterized in that, The device includes: The feature extraction module is used to extract tooth pose features and tooth geometric shape features from pre-acquired tooth point cloud data using point cloud dynamic graph convolution. The multi-channel analysis module is used to perform multi-channel image convolution on the tooth pose features and the tooth geometric shape features to obtain multi-scale features of the tooth point cloud data. A multi-layer sensing module is used to generate the target pose of the multi-scale features using a preset multi-layer perceptron; The multi-channel analysis module is specifically used for: constructing a global adjacency matrix, a local adjacency matrix, and an intersecting adjacency matrix for each tooth based on the tooth pose features and tooth geometry features; wherein, the global adjacency matrix is ​​used to represent a fully connected graph composed of each tooth in the tooth point cloud data, and the overall shape information of the tooth can be extracted based on the global adjacency matrix; the local adjacency matrix is ​​used to represent the connection relationship between each tooth and its opposite tooth, the connection relationship between each tooth and its adjacent tooth, and the connection relationship between each tooth and its adjacent tooth, and the local crowding or sparseness information of the tooth row composed of the teeth can be extracted based on the local adjacency matrix; the intersecting adjacency matrix is ​​used to represent that each tooth is connected to its opposite tooth and its adjacent tooth, and can indicate the collision relationship between the teeth; using a preset graph convolutional neural network, graph convolution is performed on the global adjacency matrix, the local adjacency matrix, and the intersecting adjacency matrix of each tooth to obtain a multi-channel convolutional graph; the multi-channel convolutional graphs corresponding to the global adjacency matrix, the local adjacency matrix, and the intersecting adjacency matrix of each tooth are concatenated to obtain the multi-scale features of the tooth point cloud data.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the multi-scale convolution-based orthodontic analysis method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the orthodontic analysis method based on multi-scale convolution as described in any one of claims 1 to 6.

Images