A tooth detection method, device, terminal equipment and readable storage medium

Abstract

The application relates to the technical field of image processing, and provides a tooth detection method and device, a terminal device and a readable storage medium, the method comprising the following steps: acquiring a CT image of a patient's oral cavity and preprocessing the CT image to obtain a three-dimensional tooth region image; performing post-processing by using a tooth detection model to obtain a tooth detection result of the patient; wherein the tooth detection model comprises a full-tooth segmenter, a tooth crown and root segmenter and a tooth crown classifier; when the post-processing is performed, the three-dimensional tooth region image is input into the full-tooth segmenter to obtain a full-tooth segmentation result; the full-tooth segmentation result and the CT image are input into the tooth crown and root segmenter to obtain a tooth crown and root segmentation result; and then the tooth crown and root segmentation result and the CT image are input into the tooth crown classifier to obtain a tooth crown classification result. The method can accurately obtain tooth segmentation information and missing tooth position information of the patient, and has high calculation speed.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a tooth detection method, apparatus, terminal device, and readable storage medium. Background Technology

[0002] In the field of oral medicine, precise tooth segmentation and the detection of missing tooth positions are crucial. This not only helps dentists gain a more comprehensive understanding of a patient's dental condition in three dimensions but also assists them in more accurately measuring tooth length and orientation, thereby enabling the development of treatment plans for oral diseases. However, current oral segmentation design methods are limited in their ability to precisely segment crowns and roots, and cannot accurately locate missing tooth positions from oral CT scans (3D computed tomography). Summary of the Invention

[0003] In view of this, embodiments of this application provide a tooth detection method, device, terminal equipment, and readable storage medium, which can effectively solve problems such as fine segmentation of tooth crowns and roots and accurate positioning of missing teeth.

[0004] In a first aspect, embodiments of this application provide a method for detecting teeth, including:

[0005] The patient's original oral cavity tomographic 3D images were acquired and preprocessed to obtain 3D images of the tooth region;

[0006] The original oral cavity tomographic three-dimensional image and the three-dimensional tooth region image are post-processed using a tooth detection model to obtain the patient's tooth detection results;

[0007] The tooth detection model includes a whole tooth segmenter, a crown and root segmenter, and a crown classifier; post-processing using the tooth detection model includes:

[0008] The three-dimensional tooth region image is input into the whole tooth segmenter for a first segmentation to obtain the whole tooth segmentation result; the whole tooth segmentation result and the original oral cavity tomographic three-dimensional image are input into the crown and root segmenter for a second segmentation to obtain the crown and root segmentation result; the crown and root segmentation result and the original oral cavity tomographic three-dimensional image are then input into the crown classifier for classification to obtain the crown classification result.

[0009] Based on the dental examination results, the patient's tooth segmentation information and missing tooth position information are obtained.

[0010] In some embodiments, the tooth detection model is obtained through pre-training, including:

[0011] Several preprocessed oral cavity tomographic 3D image samples were obtained to obtain a sample dataset, in which each oral cavity tomographic image sample contains annotation information for all tooth categories;

[0012] The tooth detection model is trained using the aforementioned sample dataset, including:

[0013] The first lightweight neural network is trained by using oral cavity tomographic 3D image samples containing the labeled information for tooth segmentation, thus obtaining the trained first lightweight neural network.

[0014] The whole tooth segmentation result output by the trained first lightweight neural network and the oral cavity tomographic three-dimensional image sample containing the annotation information are merged by channel layer to obtain the merged image sample. Then, the merged three-dimensional image sample is input into the second neural network for crown and root segmentation training to obtain the trained second neural network.

[0015] The whole tooth segmentation result output by the trained second lightweight neural network and the oral cavity tomographic three-dimensional image sample containing the annotation information are merged in the channel layer to obtain the merged image sample. Then, the merged three-dimensional image sample is input into the third neural network for crown category classification training to obtain the trained third neural network.

[0016] The trained first lightweight neural network, second neural network, and third neural network are used, in turn, as the whole tooth segmenter, the crown and root segmenter, and the crown classifier to construct the tooth detection model.

[0017] In some embodiments, the full-tooth segmenter employs a MobileNetV3 network;

[0018] The crown and root segmenter uses a first DeepLabV3 network, wherein the pooling layer in the first DeepLabV3 network uses a spatial pyramid pooling module.

[0019] The crown classifier uses a second DeepLabV3 network, wherein the last two layers of the encoder in the second DeepLabV3 network are respectively equipped with compressed excitation attention modules.

[0020] In some embodiments, the labeling information for all tooth categories includes: tooth roots labeled in sequence and crowns including 32 categories; wherein the crowns are located in the upper left, upper right, lower left and lower right regions of the oral cavity, and each region includes a central incisor, a lateral incisor, a canine, two premolars and three other adjacent molars near the two premolars.

[0021] In some embodiments, obtaining the patient's missing tooth position information based on the dental examination results includes:

[0022] Based on whether any of the 32 categories of crowns are present in the crown classification results, if at least one is missing, the position of the missing tooth is located according to the crown category of the missing tooth and the labeling order of the crowns.

[0023] In some embodiments, the preprocessing includes:

[0024] The original oral cavity 3D tomographic scan image is normalized and standardized to obtain a standardized 3D image; wherein, the standardization process includes: subtracting the normalization result after normalization from the average pixel value of all images to obtain the difference value; and taking the ratio of the obtained difference value to the standard deviation of pixels of all images as the result of the standardization process.

[0025] A rectangular region is extracted from the standardized 3D image using a defined area range to obtain a local 3D tooth region image.

[0026] In some embodiments, the tooth detection method further includes:

[0027] The three-dimensional image containing the tooth detection results is converted into a color point cloud to obtain the corresponding three-dimensional model, which is then displayed on a 3D display platform.

[0028] Secondly, embodiments of this application provide a tooth detection device, comprising:

[0029] The image acquisition module is used to acquire the patient's original oral cavity tomographic 3D images and preprocess them to obtain 3D dental region images;

[0030] The detection module is used to post-process the original oral cavity tomographic three-dimensional image and the three-dimensional tooth region image using a tooth detection model to obtain the patient's tooth detection results.

[0031] The tooth detection model includes a whole tooth segmenter, a crown and root segmenter, and a crown classifier; post-processing using the tooth detection model includes:

[0032] The three-dimensional tooth region image is input into the whole tooth segmenter for a first segmentation to obtain the whole tooth segmentation result; the whole tooth segmentation result and the original oral cavity tomographic three-dimensional image are input into the crown and root segmenter for a second segmentation to obtain the crown and root segmentation result; the crown and root segmentation result and the original oral cavity tomographic three-dimensional image are then input into the crown classifier for classification to obtain the crown classification result.

[0033] The result acquisition module is used to acquire the patient's tooth segmentation information and missing tooth position information based on the tooth detection results.

[0034] Thirdly, embodiments of this application provide a terminal device, the terminal device including a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the described tooth detection method.

[0035] Fourthly, embodiments of this application provide a readable storage medium storing a computer program that, when executed on a processor, implements the aforementioned tooth detection method.

[0036] The embodiments of this application have the following beneficial effects:

[0037] The tooth detection method of this application designs a deep learning-based tooth detection model, which includes three main parts: a whole tooth segmenter, a crown and root segmenter, and a crown classifier. When post-processing the pre-processed 3D tooth region image using the tooth detection model, the 3D tooth region image is input into the whole tooth segmenter for primary segmentation to obtain the whole tooth segmentation result. The whole tooth segmentation result and the original oral cavity 3D tomographic image are input into the crown and root segmenter for secondary segmentation to obtain the crown and root segmentation result. Then, the crown and root segmentation result and the original oral cavity 3D tomographic image are input into the crown classifier for classification to obtain the crown classification result. Thus, the patient's tooth detection result is obtained, and the patient's tooth segmentation information and missing tooth position information can be obtained. When performing tooth detection using this tooth detection model, not only can the tooth position segmentation be accurate and the interpretability be improved, but also the trained model has stronger robustness because general missing tooth samples are added during training. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A flowchart of a tooth detection method according to an embodiment of this application is shown;

[0040] Figure 2 A schematic diagram of an oral CT scan is shown;

[0041] Figure 3 A schematic diagram of a tooth detection model in an embodiment of this application is shown;

[0042] Figure 4 A flowchart of a tooth detection model training embodiment of this application is shown;

[0043] Figure 5 A schematic diagram of a crown and root segmenter according to an embodiment of this application is shown;

[0044] Figure 6 A schematic diagram of a crown classifier in an embodiment of this application is shown;

[0045] Figure 7 A three-dimensional model of a tooth detection result from an embodiment of this application is shown.

[0046] Figure 8 A schematic diagram of a tooth detection device according to an embodiment of this application is shown. Detailed Implementation

[0047] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0048] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0049] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0050] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0051] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0052] Since existing technologies offer limited solutions for crown and root segmentation and cannot accurately pinpoint the location of missing teeth, this application proposes a tooth detection method. This method employs a deep learning-based tooth detection model comprising three main components: a whole-tooth segmenter, a crown-root segmenter, and a crown classifier. These three components work synergistically to achieve post-processing of tooth images at a three-dimensional level. Specifically, the whole-tooth segmenter first segments the pre-processed three-dimensional tooth region image to obtain a whole-tooth segmentation result. Next, the crown-root segmenter performs a secondary segmentation based on the whole-tooth segmentation result combined with the original CT image to obtain a crown-root segmentation result. Finally, the crown classifier classifies the crown-root segmentation result combined with the original CT image to obtain a crown classification result. This yields the final tooth detection result, which includes accurate tooth segmentation information and information on the location of missing teeth.

[0053] The following examples illustrate this tooth detection method.

[0054] Figure 1 A flowchart of a tooth detection method according to an embodiment of this application is shown. Exemplarily, the tooth detection method includes the following steps:

[0055] S110: Acquire the patient's original oral cavity tomographic 3D image and preprocess it to obtain a 3D image of the tooth region.

[0056] The original three-dimensional oral cavity tomographic scan image refers to the original three-dimensional CT image obtained by scanning the patient's oral cavity with a tomographic scanning device, hereinafter referred to as CT image. To facilitate subsequent analysis and processing, this embodiment requires corresponding preprocessing of the CT image, such as normalization and standardization. In addition, to reduce the amount of computation, image region cropping can also be performed to obtain regional images that mainly include tooth information.

[0057] In one embodiment, the preprocessing includes: normalizing and standardizing the original oral cavity tomographic three-dimensional scan image to obtain a standardized three-dimensional image; then, using the set regional range coordinates, extracting a rectangular region from the standardized three-dimensional image to obtain a local three-dimensional tooth region image.

[0058] For example, the standardization process includes: subtracting the normalized result from the pixel mean of all images to obtain the difference; and taking the ratio of the difference to the pixel standard deviation of all images as the standardized result. Here, the pixel mean and pixel standard deviation of all images can be calculated from all image samples used during model training. If described by an expression, we have:

[0059] Normalization result = original image / 255;

[0060] Standardized result = (Normalized result – a) / b;

[0061] Where a is the average pixel value of all images, and b is the standard deviation of pixels for all images.

[0062] It is understandable that for the acquired raw CT images, by locating the three-dimensional matrix range of the area where the teeth are located, the image content without teeth can be removed, thereby reducing unnecessary calculations.

[0063] For example, in a preferred embodiment, the coordinate range of the three-dimensional matrix can be selected as [50:400, 0:512, 128:740] to ensure that all CT images can extract the rectangular region containing all teeth using this coordinate range. It is understood that other coordinates close to this range are also feasible, and those within a small error range should be considered as using this coordinate range or as a reference.

[0064] S120 uses a tooth detection model to post-process the original oral cavity tomographic 3D images and 3D tooth region images to obtain the patient's tooth detection results.

[0065] In this application, the tooth detection model includes a whole tooth segmenter, a crown and root segmenter, and a crown classifier. The whole tooth segmenter is used to segment the whole tooth, the crown and root segmenter is used to segment the crown and root separately, and the crown classifier is used to classify the segmented crowns to determine the category and location information of each tooth.

[0066] To obtain more accurate dental examination results, this method combines the original three-dimensional oral CT images with preprocessed three-dimensional images of the dental regions. For example... Figure 2 As shown, exemplarily, post-processing is performed using the aforementioned tooth detection model, including:

[0067] The three-dimensional tooth region image is input into the whole tooth segmenter for a first segmentation to obtain the whole tooth segmentation result; the whole tooth segmentation result and the original oral cavity tomographic three-dimensional image are input into the crown and root segmenter for a second segmentation to obtain the crown and root segmentation result; the crown and root segmentation result and the original oral cavity tomographic three-dimensional image are then input into the crown classifier for classification to obtain the crown classification result.

[0068] It is understood that the aforementioned tooth detection model can be obtained through pre-training of a constructed deep learning model. In one implementation, the training of this tooth detection model includes:

[0069] First, several preprocessed 3D oral cavity computed tomography (CT) image samples are acquired to obtain a sample dataset. Each CT image sample contains annotation information for all tooth categories.

[0070] like Figure 3 As shown, the labeling information for this complete tooth category includes: tooth roots labeled in sequence and crowns comprising 32 categories, totaling 33 types. Specifically, the crowns can be divided into four regions located in the oral cavity: upper left, upper right, lower left, and lower right. Each region includes one central incisor, one lateral incisor, one canine, two premolars, and three adjacent molars near the two premolars.

[0071] For example, for normal oral dentition, the categories of all teeth can be labeled as follows: 1. Tooth root; 2. Upper right central incisor; 3. Upper right lateral incisor; 4. Upper right canine; 5. Upper right first premolar; 6. Upper right second premolar; 7. Upper right first molar; 8. Upper right second molar; 9. Upper right third molar (wisdom tooth); 10. Upper left central incisor; 11. Upper left lateral incisor; 12. Upper left canine; 13. Upper left first premolar; 14. Upper left second premolar; 15. Upper left first molar; 16. Upper left second molar; 17. 18. Lower left central incisor; 19. Lower left lateral incisor; 20. Lower left canine; 21. Lower left first premolar; 22. Lower left second premolar; 23. Lower left first molar; 24. Lower left second molar; 25. Lower left third molar (wisdom tooth); 26. Lower right central incisor; 27. Lower right lateral incisor; 28. Lower right canine; 29. ​​Lower right first premolar; 30. Lower right second premolar; 31. Lower right first molar; 32. Lower right second molar; 33. Lower right third molar (wisdom tooth). It can be understood that the labels correspond to the tooth categories; in other words, by identifying the missing tooth in these categories, one can pinpoint which tooth is missing.

[0072] These samples can be derived from a certain number (e.g., 100 cases) of oral CT three-dimensional images of healthy individuals and a certain number (e.g., 100 cases) of oral CT three-dimensional images of patients with missing teeth. Through image data preprocessing, and with all CT images being labeled and confirmed by professional dentists, a sample dataset for training is obtained.

[0073] Furthermore, considering that the actual number of samples acquired may be limited, data augmentation techniques such as inversion, rotation, blurring, distortion, and Gaussian filtering can be randomly applied to the labeled masks and images to simulate oral CT images under certain special circumstances. It is understandable that augmentation processing can increase sample diversity, thereby making the trained model more robust.

[0074] Next, the constructed tooth detection model was trained using the aforementioned sample dataset.

[0075] As an alternative approach, the three main components of the tooth detection model can be trained separately: the whole tooth segmenter, the crown and root segmenter, and the crown classifier can be trained sequentially, and then the trained neural networks can be combined to obtain the tooth detection model. Alternatively, all three components can be trained simultaneously, i.e., the three neural network structures can be trained as a whole. In this case, the loss function needs to consider the loss and weights of each component to achieve a unified training objective.

[0076] If we take training each component individually as an example, such as Figure 4 As shown, the training process includes:

[0077] S210, the first lightweight neural network is trained on tooth segmentation using oral cavity tomographic scan three-dimensional image samples containing the above-mentioned annotation information to obtain the trained first lightweight neural network.

[0078] For example, the first neural network used for tooth segmentation can employ a lightweight neural network model. Considering that models such as Unet_3D have good overall segmentation capabilities for CT images and are relatively easy to segment teeth with significant density differences, in one implementation, the lightest MobileNetV3_3D network can be used to improve segmentation speed while having almost no loss of accuracy.

[0079] S220: The whole tooth segmentation results output by the trained first lightweight neural network and the oral cavity tomographic scan 3D image samples containing annotation information are merged through the channel layer to obtain merged image samples. Then, the merged 3D image samples are input into the second neural network for crown and root segmentation training to obtain the trained second neural network.

[0080] For crown and root segmentation, in this embodiment, the output whole tooth segmentation result is used as an indication of the segmentation region for CT image samples containing annotation information, so as to speed up and further improve the segmentation accuracy.

[0081] For example, the second neural network used for crown and root segmentation can employ a DeepLabV3 network. Furthermore, considering the need for more refined segmentation of the crown and root, and to achieve density partitioning of small regions, this embodiment incorporates a 3D image spatial pyramid pooling module into the DeepLabV3 network. Specifically, several parallel pyramid pooling structures are added to the last layer of the DeepLabV3 network, such as... Figure 5 As shown, this makes small-scale density differences more sensitive.

[0082] S230: The crown and root segmentation results output by the trained second lightweight neural network and the oral cavity tomographic 3D image samples containing annotation information are merged through the channel layer to obtain merged image samples. The merged 3D image samples are then input into the third neural network for crown category classification training to obtain the trained third neural network.

[0083] For crown classification, in this embodiment, the output crown and root segmentation results are used as crown classification indicators for CT image samples containing annotation information, which can also speed up and further improve the accuracy of classification.

[0084] For example, the third neural network used for crown classification can also employ the DeepLabV3 network. Furthermore, for the third neural network, to refine the segmentation of tooth boundaries, in this embodiment, in addition to maintaining the pyramid pooling structure in the crown and root segmenter, a compressed excitation attention module (i.e., SENet) is added to the last two layers of the encoder in the DeepLabV3 network (corresponding to Block 4 and Block 5). This module replicates channels and uses the sigmoid function to weight the features corresponding to the original CT image, such as... Figure 6 As shown. It can be understood that the output layer of this third neural network is set to 32 categories, where 32 categories correspond to the 32 categories of teeth mentioned above.

[0085] In this embodiment, when training the three neural networks, the initial learning rate is uniformly set to 0.0005, the training duration is 100 epochs, and the learning rate update method is WarmUp + Cosine Annealing. That is, in the first epoch, the learning rate is warmed up with 0.0001, and subsequent learning rates are calculated using Cosine Annealing, where T is set to 50 and eta is 0.0001. The optimizer used during training is the Adam optimizer, with momentum set to (0.9, 0.999). It can be understood that using Cosine Annealing better balances the model's accelerated learning process and refined convergence. Furthermore, when selecting the loss function, the DiceLoss loss function can be used, i.e., loss = (prediction ∩ label + 1) / (prediction ∪ label + 1). Actual training tests show that the three neural networks achieve their highest accuracy at 25 epochs, 44 epochs, and 72 epochs, respectively.

[0086] S240, the first lightweight neural network, the second neural network, and the third neural network trained above are used as the whole tooth segmenter, the crown and root segmenter, and the crown classifier respectively, thereby constructing a trained tooth detection model.

[0087] Therefore, by connecting the outputs and inputs of these trained neural networks, they can be combined to obtain, for example... Figure 2 The trained tooth detection model is shown.

[0088] S130 obtains the patient's tooth segmentation information and missing tooth position information based on the tooth examination results.

[0089] For example, the output of the dental examination results may include information such as the crown and root segmentation results and the crown classification results. It can be understood that the crown and root segmentation results are the patient's tooth segmentation information; for missing tooth information, it can be accurately located based on the crown classification results.

[0090] Based on the pre-labeled correspondence between each tooth and its category number, in one implementation, the missing tooth position can be located by checking if any of the 32 categories of crowns are present in the crown classification results. If at least one is missing, the missing tooth position can be determined according to its crown category and the labeling order. For example, if a certain tooth position is not included in the output classification results, such as a missing tooth in category 15, the upper left first molar (corresponding to label 15) can be identified as the missing tooth position based on the pre-labeled correspondence between tooth position and category.

[0091] As an alternative, to facilitate a more intuitive viewing of the patient's dental examination results, the method also includes displaying the results using a three-dimensional model.

[0092] As an example, the 3D image containing the above-mentioned dental examination results can be converted into a color (RGB) point cloud, where different colors can be used to distinguish different tooth categories, thus obtaining point cloud data. This point cloud data can then be used for 3D modeling to obtain the corresponding 3D model, which can then be displayed through a 3D display platform. For example, for the dental examination results of a particular patient, the following can be obtained: Figure 7 The three-dimensional model shown above can be understood to assist doctors in quickly designing procedures for filling missing teeth in real-world scenarios.

[0093] The tooth detection method proposed in this application employs a deep learning-based tooth detection model. This model comprises three main components: a whole-tooth segmenter, a crown and root segmenter, and a crown classifier. These three components work synergistically to achieve post-processing of tooth images at a three-dimensional level. It can be understood that by improving the network structure of each component in the tooth detection model and utilizing the output of the previous layer as the input for the next layer, not only can tooth segmentation and localization be accurately achieved, but the calculation speed is also faster, reducing waiting time for doctors and patients.

[0094] Figure 8 A schematic diagram of a dental detection device 100 according to an embodiment of this application is shown. Exemplarily, the dental detection device 100 includes:

[0095] The image acquisition module 110 is used to acquire the patient's original oral cavity tomographic three-dimensional image and perform preprocessing to obtain a three-dimensional image of the tooth region.

[0096] The detection module 120 is used to perform post-processing on the original oral cavity tomographic three-dimensional image and the three-dimensional tooth region image using a tooth detection model to obtain the patient's tooth detection results.

[0097] The tooth detection model includes a whole tooth segmenter, a crown and root segmenter, and a crown classifier; post-processing using the tooth detection model includes:

[0098] The three-dimensional tooth region image is input into the whole tooth segmenter for a first segmentation to obtain the whole tooth segmentation result; the whole tooth segmentation result and the original oral cavity tomographic three-dimensional image are input into the crown and root segmenter for a second segmentation to obtain the crown and root segmentation result; the crown and root segmentation result and the original oral cavity tomographic three-dimensional image are then input into the crown classifier for classification to obtain the crown classification result.

[0099] The result acquisition module 130 is used to acquire the patient's tooth segmentation information and missing tooth position information based on the tooth detection results.

[0100] It is understood that the device in this embodiment corresponds to the tooth detection method in the above embodiment, and the options in the above embodiment are also applicable to this embodiment, so they will not be described again here.

[0101] This application also provides a terminal device, such as a computer, dental testing instrument, etc., the specific form of which is not limited. Exemplarily, the terminal device includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to enable the terminal device to perform the functions of the various modules in the aforementioned dental testing method or dental testing device.

[0102] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0103] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0104] This application also provides a readable storage medium for storing the computer program used in the aforementioned terminal device. For example, the readable storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0105] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0106] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0107] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0108] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A tooth detection method, characterized by, The method comprises the following steps: obtaining and preprocessing original oral tomography three-dimensional images of a patient to obtain three-dimensional tooth region images; the preprocessing comprises locating the three-dimensional matrix range position where all teeth are located and cutting out the corresponding region image as the three-dimensional tooth region image; postprocessing the original oral tomography three-dimensional images and the three-dimensional tooth region images by using a tooth detection model to obtain tooth detection results of the patient; wherein the tooth detection model comprises a full tooth segmenter, a tooth crown and root segmenter and a tooth crown classifier; the tooth detection model is trained by using oral tomography image samples containing labeled information of a full tooth category, the full tooth category comprising tooth roots labeled in sequence and tooth crowns comprising 32 categories; wherein the full tooth segmenter adopts a MobileNetV3 network; the tooth crown and root segmenter adopts a first DeepLabV3 network, wherein the pooling layer of the first DeepLabV3 network uses a spatial gold tower pooling module; the tooth crown classifier adopts a second DeepLabV3 network, wherein the last two layers of the encoder of the second DeepLabV3 network are additionally provided with a compression excitation attention module; the postprocessing by using the tooth detection model comprises: inputting the three-dimensional tooth region image into the full tooth segmenter for one-time segmentation to obtain a full tooth segmentation result; inputting the full tooth segmentation result and the original oral tomography three-dimensional image into the tooth crown and root segmenter for two-time segmentation after channel layer merging processing to obtain a tooth crown and root segmentation result; and inputting the tooth crown and root segmentation result and the original oral tomography three-dimensional image into the tooth crown classifier for classification after channel layer merging processing to obtain a tooth crown classification result; obtaining tooth segmentation information and missing tooth position information of the patient based on the tooth detection results.

2. The tooth detection method of claim 1, wherein, The tooth detection model is pre-trained by using oral tomography image samples containing labeled information of a full tooth category, comprising the following steps: obtaining a plurality of preprocessed oral tomography three-dimensional image samples to obtain a sample data set, wherein each oral tomography three-dimensional image sample contains labeled information of a full tooth category; training a constructed tooth detection model by using the sample data set, comprising the following steps: training a first lightweight neural network by using the oral tomography three-dimensional image samples containing the labeled information to obtain a trained first lightweight neural network; performing channel layer merging processing on the full tooth segmentation result output by the trained first lightweight neural network and the oral tomography three-dimensional image samples containing the labeled information to obtain a merged image sample, and inputting the merged three-dimensional image sample into a second neural network for tooth crown and root segmentation training to obtain a trained second neural network; The full tooth segmentation result output by the trained second lightweight neural network and the oral tomography three-dimensional image sample containing the annotation information are subjected to channel layer merging processing to obtain a merged image sample, and the merged three-dimensional image sample is input into a third neural network for crown category classification training to obtain a trained third neural network; The trained first lightweight neural network, the second neural network and the third neural network are sequentially used as the full tooth segmenter, the crown and root segmenter and the crown classifier to construct the tooth detection model.

3. The tooth detection method of claim 1, wherein, The 32 categories of crowns are respectively located in the upper left, upper right, lower left and lower right regions of the oral cavity, and each of the regions includes a central incisor, a lateral incisor, a canine, two premolars and three adjacent molars close to the two premolars.

4. The tooth detection method of claim 3, wherein, The method further includes: According to whether the 32 categories of crowns exist in the crown classification result, if at least one is missing, the missing tooth position of the patient is located according to the category of the missing tooth and the annotation order of the crown.

5. The tooth detection method of claim 2, wherein, The preprocessing includes: The original oral tomography three-dimensional scanning image is subjected to normalization and standardization processing to obtain a standardized three-dimensional image; the standardization processing includes: subtracting the normalized result after the normalization processing from the average value of the pixels of all images to obtain a difference value; and using the ratio of the obtained difference value to the standard deviation of the pixels of all images as the result after the standardization processing; The standardized three-dimensional image is subjected to rectangular region extraction using a set region range to obtain a local three-dimensional tooth region image.

6. The tooth detection method of claim 1, wherein, The method further includes: The three-dimensional image containing the tooth detection result is subjected to color point cloud conversion to obtain a corresponding three-dimensional model, and the three-dimensional model is displayed through a 3D display platform.

7. A tooth detection model acquisition apparatus characterized by comprising: The method includes: An image acquisition module is configured to acquire an original oral tomography three-dimensional image of a patient and perform preprocessing to obtain a three-dimensional tooth region image; The preprocessing includes locating the three-dimensional matrix range position where all the teeth are located and cropping a corresponding region image as the three-dimensional tooth region image; A detection module is configured to perform post-processing on the original oral tomography three-dimensional image and the three-dimensional tooth region image using a tooth detection model to obtain a tooth detection result of the patient; and The method further includes: According to whether the 32 categories of crowns exist in the crown classification result, if at least one is missing, the missing tooth position of the patient is located according to the category of the missing tooth and the annotation order of the crown. The preprocessing includes: The original oral tomography three-dimensional scanning image is subjected to normalization and standardization processing to obtain a standardized three-dimensional image; the standardization processing includes: subtracting the normalized result after the normalization processing from the average value of the pixels of all images to obtain a difference value; and using the ratio of the obtained difference value to the standard deviation of the pixels of all images as the result after the standardization processing; The standardized three-dimensional image is subjected to rectangular region extraction using a set region range to obtain a local three-dimensional tooth region image. The tooth detection model comprises a full tooth segmenter, a tooth crown and root segmenter, and a tooth crown classifier; the tooth detection model is trained by using an oral tomography image sample comprising labeled information of a full tooth category, the full tooth category comprising tooth roots labeled in sequence and tooth crowns comprising 32 categories; the full tooth segmenter adopts a MobileNetV3 network; the tooth crown and root segmenter adopts a first DeepLabV3 network, wherein a spatial gold tower pooling module is used in the pooling layer of the first DeepLabV3 network; the tooth crown classifier adopts a second DeepLabV3 network, wherein a compression excitation attention module is additionally arranged in the last two layers of the encoder of the second DeepLabV3 network; when the tooth detection model is used for post-processing, the following steps are included: The three-dimensional tooth region image is input into the full tooth segmenter for one-time segmentation to obtain a full tooth segmentation result; the full tooth segmentation result and the original oral tomography three-dimensional image are subjected to channel layer merging processing and then input into the tooth crown and root segmenter for two-time segmentation to obtain a tooth crown and root segmentation result; the tooth crown and root segmentation result and the original oral tomography three-dimensional image are subjected to channel layer merging processing and then input into the tooth crown classifier for classification to obtain a tooth crown classification result; A result obtaining module is configured to obtain tooth segmentation information and missing tooth site information of the patient based on the tooth detection result.

8. A terminal device, comprising: The terminal device comprises a processor and a memory, the memory stores a computer program, and the processor is configured to execute the computer program to implement the tooth detection method in any one of claims 1-6.

9. A readable storage medium, characterized by, The computer program is stored in the memory and is executed on the processor to implement the tooth detection method in any one of claims 1-6.

Images