Tooth segmentation method, device, equipment and storage medium for oral medical images

Abstract

The embodiment of the present application provides a method, device, equipment and storage medium for tooth segmentation of oral medical images. The method includes: determining a first training set and a second training set according to the acquired oral medical image samples; generating a first pipeline fingerprint for training a tooth classification prediction module according to the first data fingerprint of the first training set, and training the tooth classification prediction module according to the first training set and the first pipeline fingerprint; generating a second pipeline fingerprint for training a tooth partition prediction module according to the second data fingerprint of the second training set, and training the tooth partition prediction module according to the second training set and the second pipeline fingerprint; respectively predicting the classification and partition of teeth in the target oral medical image; and determining the tooth segmentation result according to the tooth classification prediction result and the tooth partition prediction result. This scheme can effectively improve the segmentation efficiency and segmentation accuracy of teeth in oral medical images.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence, and in particular to a method, device, electronic device and computer-readable medium for tooth segmentation of oral medical images. Background Art

[0002] In the field of oral surgical robots, tooth segmentation is an important step for oral surgical robots to accurately implant teeth. During the dental implantation process, the oral surgical robot needs to locate the vacant tooth position and place the implant. At the same time, it needs to confirm the specific position, depth and angle of the implant based on the teeth adjacent to the vacant tooth position. This requires the oral surgical robot to be able to segment the teeth in the mouth as accurately as possible to identify the position of each tooth in the mouth.

[0003] In the prior art, tooth segmentation methods are mainly divided into two categories. The first category of methods is based on traditional algorithms, such as threshold segmentation, region growing, etc., and the second category of methods is deep learning methods, such as the PointNet model. The first category of methods requires artificial prior knowledge, and when oral medical images become more and more complex and diverse, they cannot adapt well to all oral medical images, thereby reducing the accuracy of tooth segmentation in oral medical images. Compared with the first category of methods, the second category of methods can learn the optimal model by processing a large number of oral medical images, and can segment teeth end-to-end, and the accuracy and generalization ability are effectively improved. However, before model training, it is necessary to manually prepare training parameters for the model to be trained. Different training parameters will affect the accuracy of the model's segmentation of teeth in oral medical images. In addition, it also affects the model's efficiency in segmenting teeth in oral medical images.

[0004] It can be seen that how to effectively improve the segmentation efficiency and accuracy of teeth in oral medical images has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present application is to propose a tooth segmentation method, device, electronic device and computer-readable medium for oral medical images, so as to solve the technical problem existing in the prior art of how to effectively improve the segmentation efficiency and segmentation accuracy of teeth in oral medical images.

[0006] According to a first aspect of an embodiment of the present application, a method for tooth segmentation of an oral medical image is provided. The method comprises: determining a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model according to an oral medical image sample obtained by scanning an oral cavity with an oral scanning device; adaptively generating a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set by a first adaptive module in the tooth classification prediction model, and training the tooth classification prediction module according to the first training set and the first pipeline fingerprint; adaptively generating a second pipeline fingerprint for training the tooth partition prediction module according to a second data fingerprint of the second training set by a second adaptive module in the tooth partition prediction model, and training the tooth partition prediction module according to the second training set and the second pipeline fingerprint; predicting the classification of teeth in a target oral medical image by the trained tooth classification prediction module, and predicting the partition of teeth in the target oral medical image by the trained tooth partition prediction module; determining a tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image.

[0007] According to the second aspect of the embodiment of the present application, a tooth segmentation device for oral medical images is provided. The device includes: a first determination module, which is used to determine a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model according to an oral medical image sample obtained by scanning an oral cavity with an oral scanning device; a first training module, which is used to adaptively generate a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth classification prediction model, and train the tooth classification prediction module according to the first training set and the first pipeline fingerprint; a second training module, which is used to adaptively generate a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth partition prediction model; and train the tooth classification prediction module according to the first training set and the first pipeline fingerprint; A second adaptive module is provided, which adaptively generates a second pipeline fingerprint for training the tooth partition prediction module according to the second data fingerprint of the second training set, and trains the tooth partition prediction module according to the second training set and the second pipeline fingerprint; a prediction module is used to predict the classification of teeth in the target oral medical image through the trained tooth classification prediction module, and predict the partition of teeth in the target oral medical image through the trained tooth partition prediction module; a second determination module is used to determine the tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image.

[0008] According to the third aspect of the embodiments of the present application, an electronic device is provided, comprising: one or more processors; a storage device configured to store one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the tooth segmentation method of oral medical images as described in the first aspect of the embodiments of the present application.

[0009] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method for tooth segmentation of oral medical images as described in the first aspect of an embodiment of the present application is implemented.

[0010] Through the tooth segmentation scheme for oral medical images provided by the embodiment of the present application, there is no need to manually prepare pipeline fingerprints for the tooth classification prediction module to be trained and the tooth partition prediction module to be trained in advance. Instead, the first adaptive module in the tooth classification prediction model adaptively generates the first pipeline fingerprint for training the tooth classification prediction module according to the first data fingerprint of the first training set, and the second adaptive module in the tooth partition prediction model adaptively generates the second pipeline fingerprint for training the tooth partition prediction module according to the second data fingerprint of the second training set, thereby avoiding the trouble of manually preparing pipeline fingerprints, reducing the interference of human factors, and effectively improving the efficiency of tooth segmentation in oral medical images. In addition, the diversity of tooth segmentation results in oral medical images caused by manually preparing pipeline fingerprints is effectively avoided, thereby effectively improving the accuracy of tooth segmentation in oral medical images.

[0011] Furthermore, the classification of teeth in the target oral medical image is predicted by the trained tooth classification prediction module, and the partitioning of teeth in the target oral medical image is predicted by the trained tooth partitioning prediction module, and then the tooth segmentation result of the target oral medical image is determined based on the tooth classification prediction result and the tooth partitioning prediction result of the target oral medical image, which can effectively enable the trained tooth classification prediction module and the trained tooth partitioning prediction module to more fully utilize the limited data of the target oral medical image, thereby effectively improving the segmentation accuracy of teeth in the target oral medical image. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Hereinafter, some specific embodiments of the present application will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings:

[0013] Figure 1 This is a flowchart of the steps of the method for teeth segmentation in oral medical images according to the first embodiment of the present application;

[0014] Figure 2 This is a schematic diagram of the structure of a tooth segmentation device for oral medical images according to the second embodiment of the present application;

[0015] Figure 3 This is a schematic diagram of the structure of the electronic device in the third embodiment of the present application;

[0016] Figure 4 This is the hardware structure of the electronic device in Example 4 of the present application. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.

[0018] Reference Figure 1 , shows a flowchart of the steps of the method for teeth segmentation of oral medical images according to the first embodiment of the present invention.

[0019] Specifically, the tooth segmentation method of the oral medical image provided in this embodiment includes the following steps:

[0020] In step S101, a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model are determined based on oral medical image samples obtained by scanning an oral cavity with an oral scanning device.

[0021] In this embodiment, the oral scanning device can be understood as a device for scanning the patient's oral cavity to obtain an oral medical image, and the oral scanning device can be a cone beam CT (CBCT) or an oral scanner. The oral medical image can be understood as a three-dimensional image obtained by imaging the oral cavity. The oral medical image sample can be understood as an oral medical image used as a training sample for the model. The tooth classification prediction model may include a first adaptive module and a tooth classification prediction module connected to the first adaptive module. The tooth classification prediction model may be nnU-Net (Self-Adapting Framework for U-Net-Based Medical Image Segmentation, an adaptive framework for medical image segmentation based on U-Net). This is a deep learning network model in the field of medical segmentation, referred to as nnU-Net. It is improved on the basis of traditional U-Net. The traditional U-Net consists of an encoding part and a decoding part, and its appearance is like "U", so it is called U-Net. nnU-Net does not change the model structure of U-Net, but adds an adaptive module to analyze the data preprocessing, model scale and parameters, and post-processing in detail and automatically select appropriate parameters. The tooth classification prediction module may be U-Net (Convolutional Networks for Biomedical Image Segmentation). The tooth partition prediction model may include a second adaptive module and a tooth partition prediction module connected to the second adaptive module. The tooth partition prediction model may be nnU-Net. The tooth partition prediction module may be U-Net. The second training set includes oral medical image samples and their tooth partition labels for training the tooth partition prediction module. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0022] In a specific example, oral medical images can be first loaded from CBCT to obtain a data set of oral medical image samples. Then, the data format of the data set is converted into the image format required by the training model. For example, the dicom format of the data set is converted into the nii.gz format required by the training model, and the data set is divided into a training set and a test set at a ratio of 8:2. The training set covers as many oral medical image samples as possible, which is helpful for model training and learning different features to complete the segmentation of oral medical images. The test set is used to evaluate the model obtained by training. For the tooth classification task, the first training set includes oral medical image samples and their tooth classification labels for training the tooth classification prediction module. For the tooth partitioning task, the second training set includes oral medical image samples and their tooth partitioning labels for training the tooth partitioning prediction module. The oral medical image samples in the first training set are the same as the oral medical image samples in the second training set, but the tooth classification labels are different from the tooth partitioning labels. nnU-NET is used to complete tooth classification and tooth partitioning respectively. The tooth classification labels include 7 categories, namely 1, 2, 3, 4, 5, 6, and 7. The tooth partitioning labels include 4 categories, namely 1, 2, 3, and 4. Specifically, the 28 teeth are numbered 1-4, the upper left 7 teeth are numbered 1, the lower left 7 teeth are numbered 2, the upper right 7 teeth are numbered 3, and the lower right 7 teeth are numbered 4. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0023] In step S102, a first pipeline fingerprint for training the tooth classification prediction module is adaptively generated according to the first data fingerprint of the first training set through the first adaptive module in the tooth classification prediction model, and the tooth classification prediction module is trained according to the first training set and the first pipeline fingerprint.

[0024] In this embodiment, the first adaptive module can be understood as a program module for adaptive selection. The first data fingerprint of the first training set can be understood as the attributes of the first training set, including the attributes of the oral medical image samples in the first training set and the attributes of the training tasks associated with the first training set. The first data fingerprint of the first training set may include the size of the oral medical image samples in the first training set, the number of voxels of different spatial dimensions of the oral medical image samples in the first training set, the voxel spacing of the oral medical image samples in the first training set, the grayscale information of the oral medical image samples in the first training set, and the task category of the training tasks associated with the first training set. Among them, the task category of the training tasks associated with the first training set can be tooth classification. The first pipeline fingerprint can be understood as the training parameters related to the training of the tooth classification prediction module. When the tooth classification prediction module is U-Net, the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet. Therefore, the first pipeline fingerprint includes the identification information of the network to be trained in the tooth classification prediction module and the network hyperparameters of the network to be trained. The identification information can be understood as information for indicating the network category, which can indicate that the network category is 2D-Unet, 3D-Unet, or 3D cascade Unet. The network hyperparameters of the network to be trained may include the learning rate of the network to be trained, the loss function of the network to be trained, the optimizer of the network to be trained, the training batch size of the network to be trained, the number of iterations of the network to be trained, etc.

[0025] During the training process of the network to be trained, the learning rate of the network to be trained can be dynamically adjusted:

[0026]

[0027] Wherein, lr represents the current learning rate, baselr represents the initial learning rate, which can be 0.01, epoch represents the number of iterations of the network to be trained, maxepoch represents the maximum number of iterations, and power (taken as 0.99) controls the shape of the curve.

[0028] Among them, the loss function of the network to be trained may be the sum of Dice loss and cross entropy loss. The optimizer adopts stochastic gradient descent (SGD). The number of iterations of the network to be trained may be 1000 times per round. The training batch size of the network to be trained may be 250 iteration samples each time. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0029] In a specific example, when adaptively generating identification information of the network to be trained in the tooth classification prediction module, the identification information of the fixed network to be trained can be determined as the identification information of the adaptively generated network to be trained through the first adaptive module. When adaptively generating the learning rate of the network to be trained in the tooth classification prediction module, the learning rate of the fixed network to be trained can be determined as the learning rate of the adaptively generated network to be trained through the first adaptive module. When adaptively generating the loss function of the network to be trained in the tooth classification prediction module, the loss function of the fixed network to be trained can be determined as the loss function of the adaptively generated network to be trained through the first adaptive module. When adaptively generating the optimizer of the network to be trained in the tooth classification prediction module, the optimizer of the fixed network to be trained can be determined as the optimizer of the adaptively generated network to be trained through the first adaptive module. When adaptively generating the training batch size of the network to be trained in the tooth classification prediction module, the training batch size of the network to be trained in the tooth classification prediction module can be adaptively generated through the first adaptive module according to the size of the oral medical image samples in the first training set and the GPU memory size. When adaptively generating the number of iterations of the network to be trained in the tooth classification prediction module, the first adaptive module can determine the fixed number of iterations of the network to be trained as the number of iterations of the network to be trained generated adaptively. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0030] In some optional embodiments, the method further includes: adaptively selecting a preprocessing method for the first training set according to the first data fingerprint of the first training set through the first adaptive module; preprocessing the first training set according to the preprocessing method to obtain the preprocessed first training set, and training the tooth classification prediction module according to the first training set and the first pipeline fingerprint includes: training the tooth classification prediction module according to the preprocessed first training set and the first pipeline fingerprint. Thereby, there is no need to manually rely on prior knowledge to select a preprocessing method for the first training set, but adaptively selecting a preprocessing method for the first training set according to the first data fingerprint of the first training set through the first adaptive module, and preprocessing the first training set according to the preprocessing method, which not only effectively improves the efficiency of preprocessing the first training set, but also effectively improves the accuracy of preprocessing the first training set. It can be understood that the above description is only exemplary, and this embodiment does not make any limitation to this.

[0031] In a specific example, the preprocessing method includes cropping, resampling and normalization. When adaptively selecting a cropping method for the first training set, the first adaptive module can be used to determine the fixed cropping method of the oral medical image samples in the first training set as the adaptively selected cropping method. When adaptively selecting a resampling method for the first training set, the first adaptive module can be used to adaptively select a resampling method for the first training set based on the voxel spacing of the oral medical image samples in the first training set. When adaptively selecting a normalization method for the first training set, the first adaptive module can be used to adaptively select a normalization method for the first training set based on the voxel size of the oral medical image samples in the first training set. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0032] In a specific example, the preprocessing method includes cropping, resampling and normalization. The preprocessing method is used to preprocess the first training set to obtain the first training set after preprocessing, including: cropping the oral medical image samples in the first training set to obtain the cropped oral medical image samples; resampling the cropped oral medical image samples according to the voxel spacing of the cropped oral medical image samples to obtain the resampled oral medical image samples; normalizing the resampled oral medical image samples according to the voxel size of the resampled oral medical image samples to obtain the normalized oral medical image samples, and obtaining the first training set after preprocessing according to the normalized oral medical image samples. The cropping operation is used to crop the zero-value areas of all oral medical image samples in the first training set to retain the non-zero-value areas, reduce the image size, and reduce the amount of model calculation. Among them, different oral medical image samples may have different voxel spacings due to different oral scanning devices or device acquisition protocols. In order to ensure that the model can learn the correct spatial semantics, the voxel spacing of the oral medical image samples is resampled to the median voxel spacing. Among them, the oral medical image samples after resampling are normalized by the voxel size of the oral medical image samples after resampling, which can avoid the adverse effects of singular data and accelerate model convergence. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0033] In a specific example, when resampling the cropped oral medical image sample according to the voxel spacing of the cropped oral medical image sample, the ratio of the maximum axis spacing of the voxels of the cropped oral medical image sample to the minimum axis spacing of the voxels of the cropped oral medical image sample is calculated; if the ratio is less than or equal to 3, cubic linear interpolation is used to resample the cropped oral medical image sample; if the ratio is greater than 3, cubic linear interpolation is used to resample the plane of the cropped oral medical image sample, and the nearest neighbor algorithm is used to resample the plane of the cropped oral medical image sample. When normalizing the resampled oral medical image sample according to the voxel size of the resampled oral medical image sample, the mean value and standard deviation of the voxel size of the resampled oral medical image sample are calculated; the resampled oral medical image sample is normalized according to the mean value and standard deviation of the voxel size of the sampled oral medical image sample to obtain the normalized oral medical image sample. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0034] In some optional embodiments, the method further includes: adaptively selecting a data enhancement method for the first training set through the first adaptive module; performing data enhancement on the first training set according to the data enhancement method to obtain the first training set after data enhancement, and training the tooth classification prediction module according to the first training set and the first pipeline fingerprint, including: training the tooth classification prediction module according to the first training set after data enhancement and the first pipeline fingerprint. Thereby, there is no need to manually rely on prior knowledge to select a data enhancement method for the first training set, but the first adaptive module is used to adaptively select a data enhancement method for the first training set, and the first training set is data enhanced according to the data enhancement method, which not only effectively improves the efficiency of data enhancement on the first training set, but also effectively improves the accuracy of data enhancement on the first training set. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0035] In a specific example, the data enhancement method includes rotation scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing. When the rotation scaling method is adaptively selected for the first training set, the first adaptive module can be used to determine the fixed rotation scaling method of the first training set as the adaptively generated rotation scaling method of the first training set. When the Gaussian noise method is adaptively selected for the first training set, the first adaptive module can be used to determine the fixed Gaussian noise method of the first training set as the adaptively generated Gaussian noise method of the first training set. When the Gaussian blur method is adaptively selected for the first training set, the first adaptive module can be used to determine the fixed Gaussian blur method of the first training set as the adaptively generated Gaussian blur method of the first training set. When the brightness processing method is adaptively selected for the first training set, the first adaptive module can be used to determine the fixed brightness processing method of the first training set as the adaptively generated brightness processing method of the first training set. When the contrast processing method is adaptively selected for the first training set, the first adaptive module can be used to determine the fixed contrast processing method of the first training set as the adaptively generated contrast processing method of the first training set. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0036] In a specific example, the data enhancement method includes rotation and scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing. When data enhancement is performed on the first training set according to the data enhancement method, the oral medical image samples in the first training set are rotated and scaled to obtain the oral medical image samples after rotation and scaling, and the first training set after data enhancement is obtained according to the oral medical image samples after rotation and scaling; or, Gaussian noise is performed on the oral medical image samples in the first training set to obtain the oral medical image samples after Gaussian noise, and the first training set after data enhancement is obtained according to the oral medical image samples after Gaussian noise; or, Gaussian blur is performed on the oral medical image samples in the first training set to obtain the oral medical image samples after Gaussian blur, and the first training set after data enhancement is obtained according to the oral medical image samples after Gaussian blur; or, brightness processing is performed on the oral medical image samples in the first training set to obtain the oral medical image samples after brightness processing, and the first training set after data enhancement is obtained according to the oral medical image samples after brightness processing; or, contrast processing is performed on the oral medical image samples in the first training set to obtain the oral medical image samples after contrast processing, and the first training set after data enhancement is obtained according to the oral medical image samples after contrast processing. Among them, the Gaussian noise addition can be understood as adding noise to the oral medical image samples, and the probability density function of the noise obeys the Gaussian distribution. The Gaussian blur can be understood as actually filtering the oral medical image samples. The filtering method is Gaussian filtering, which is image blurring to extract important information of the oral medical image samples. The brightness processing can be understood as changing the brightness of the oral medical image samples, and the contrast processing can be understood as changing the contrast of the oral medical image samples. The more and more diverse the oral medical image samples in the first training set, the more conducive to model training and improving the generalization ability of the model. In order to train the best possible model on the limited first training set, the oral medical image samples in the first training set are expanded by data enhancement to improve the quality of the training samples, and increase the noise data to improve the robustness of the model. It can be understood that the above description is only exemplary, and this embodiment does not make any limitation to this.

[0037] In some optional embodiments, the first pipeline fingerprint includes identification information of the network to be trained in the tooth classification prediction module and network hyperparameters of the network to be trained. When the tooth classification prediction module is trained according to the first training set and the first pipeline fingerprint, the network to be trained is determined according to the identification information, and the network to be trained is set according to the network hyperparameters to obtain the set network to be trained; the classification of teeth in the oral medical image samples in the first training set is predicted by the set network to be trained to obtain the tooth classification prediction results of the oral medical image samples; the set network to be trained is trained according to the tooth classification prediction results of the oral medical image samples and the tooth classification labeling results of the oral medical image samples. Thereby, the trained network can accurately classify the teeth of the target oral medical image. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0038] In a specific example, the tooth classification prediction result of the oral medical image sample can be understood as the predicted probability data of the tooth classification label to which the voxels in the oral medical image sample belong, and the tooth classification annotation result of the oral medical image sample can be understood as the annotation probability data of the tooth classification label to which the voxels in the oral medical image sample belong. When the network to be trained after setting is trained according to the tooth classification prediction result of the oral medical image sample and the tooth classification annotation result of the oral medical image sample, the difference value between the predicted probability data and the annotation probability data is determined by the target loss function; based on the difference value, the parameters of the network to be trained after setting are adjusted. Among them, the target loss function can be the sum of Dice loss and cross entropy loss. When adjusting the parameters of the network to be trained after setting, the back propagation algorithm or the stochastic gradient descent algorithm can be used to adjust the parameters of the network to be trained after setting. It can be understood that the above description is only exemplary, and the embodiments of the present application do not impose any limitation on this.

[0039] In a specific example, by determining the difference value between the predicted probability data and the labeled probability data, the currently obtained predicted probability data is evaluated as a basis for the subsequent training of the network to be trained after the setting. Specifically, the difference value can be transmitted back to the network to be trained after the setting, so as to iteratively train the network to be trained after the setting. The training of the network to be trained after the setting is an iterative process. This embodiment only describes one training process, but those skilled in the art should understand that this training method can be used for each training of the network to be trained after the setting until the training of the network to be trained after the setting is completed. It can be understood that the above description is only exemplary, and the embodiments of the present application do not impose any limitation on this.

[0040] In step S103, a second pipeline fingerprint for training the dental partition prediction module is adaptively generated according to the second data fingerprint of the second training set through the second adaptive module in the dental partition prediction model, and the dental partition prediction module is trained according to the second training set and the second pipeline fingerprint.

[0041] In this embodiment, the second adaptive module can be understood as a program module for adaptive selection. The second data fingerprint of the second training set can be understood as the attributes of the second training set, including the attributes of the oral medical image samples in the second training set and the attributes of the training tasks associated with the second training set. The second data fingerprint of the second training set may include the size of the oral medical image samples in the second training set, the number of voxels of different spatial dimensions of the oral medical image samples in the second training set, the voxel spacing of the oral medical image samples in the second training set, the grayscale information of the oral medical image samples in the second training set, and the task category of the training tasks associated with the second training set. Among them, the task category of the training tasks associated with the second training set can be tooth partitioning. The second pipeline fingerprint can be understood as the training parameters related to the training of the tooth partition prediction module. When the tooth partition prediction module is U-Net, the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet. Therefore, the second pipeline fingerprint includes the identification information of the network to be trained in the tooth partition prediction module and the network hyperparameters of the network to be trained. The identification information can be understood as information for indicating the network category, which can indicate that the network category is 2D-Unet, 3D-Unet, or 3D cascade Unet. The network hyperparameters of the network to be trained may include the learning rate of the network to be trained, the loss function of the network to be trained, the optimizer of the network to be trained, the training batch size of the network to be trained, the number of iterations of the network to be trained, etc.

[0042] During the training process of the network to be trained, the learning rate of the network to be trained can be dynamically adjusted:

[0043]

[0044] Wherein, lr represents the current learning rate, baselr represents the initial learning rate, which can be 0.01, epoch represents the number of iterations of the network to be trained, maxepoch represents the maximum number of iterations, and power (taken as 0.99) controls the shape of the curve.

[0045] Among them, the loss function of the network to be trained may be the sum of Dice loss and cross entropy loss. The optimizer adopts stochastic gradient descent (SGD). The number of iterations of the network to be trained may be 1000 times per round. The training batch size of the network to be trained may be 250 iteration samples each time. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0046] In a specific example, when adaptively generating identification information of the network to be trained in the tooth partition prediction module, the identification information of the fixed network to be trained can be determined as the identification information of the adaptively generated network to be trained through the second adaptive module. When adaptively generating the learning rate of the network to be trained in the tooth partition prediction module, the learning rate of the fixed network to be trained can be determined as the learning rate of the adaptively generated network to be trained through the second adaptive module. When adaptively generating the loss function of the network to be trained in the tooth partition prediction module, the loss function of the fixed network to be trained can be determined as the loss function of the adaptively generated network to be trained through the second adaptive module. When adaptively generating the optimizer of the network to be trained in the tooth partition prediction module, the optimizer of the fixed network to be trained can be determined as the optimizer of the adaptively generated network to be trained through the second adaptive module. When adaptively generating the training batch size of the network to be trained in the tooth partition prediction module, the training batch size of the network to be trained in the tooth partition prediction module can be adaptively generated through the second adaptive module according to the size of the oral medical image samples in the second training set and the GPU memory size. When adaptively generating the number of iterations of the network to be trained in the tooth partition prediction module, the second adaptive module can determine the fixed number of iterations of the network to be trained as the number of iterations of the network to be trained generated adaptively. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0047] In some optional embodiments, the method further includes: adaptively selecting a preprocessing method for the second training set according to the second data fingerprint of the second training set through the second adaptive module; preprocessing the second training set according to the preprocessing method to obtain the preprocessed second training set, and training the tooth partition prediction module according to the second training set and the second pipeline fingerprint includes: training the tooth partition prediction module according to the preprocessed second training set and the second pipeline fingerprint. Thereby, there is no need to manually rely on prior knowledge to select a preprocessing method for the second training set, but adaptively selecting a preprocessing method for the second training set according to the second data fingerprint of the second training set through the second adaptive module, and preprocessing the second training set according to the preprocessing method, which not only effectively improves the efficiency of preprocessing the second training set, but also effectively improves the accuracy of preprocessing the second training set. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0048] In a specific example, the preprocessing method includes cropping, resampling and normalization. When adaptively selecting a cropping method for the second training set, the second adaptive module can be used to determine the cropping method of the fixed oral medical image samples in the second training set as the adaptively selected cropping method. When adaptively selecting a resampling method for the second training set, the second adaptive module can be used to adaptively select a resampling method for the second training set based on the voxel spacing of the oral medical image samples in the second training set. When adaptively selecting a normalization method for the second training set, the second adaptive module can be used to adaptively select a normalization method for the second training set based on the voxel size of the oral medical image samples in the second training set. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0049] In a specific example, the preprocessing method includes cropping, resampling and normalization, and the preprocessing of the second training set according to the preprocessing method to obtain the preprocessed second training set includes: cropping the oral medical image samples in the second training set to obtain the cropped oral medical image samples; resampling the cropped oral medical image samples according to the voxel spacing of the cropped oral medical image samples to obtain the resampled oral medical image samples; normalizing the resampled oral medical image samples according to the voxel size of the resampled oral medical image samples to obtain the normalized oral medical image samples, and obtaining the preprocessed second training set according to the normalized oral medical image samples. The cropping operation is used to crop the zero-value areas of all oral medical image samples in the second training set to retain the non-zero-value areas, reduce the image size, and reduce the amount of model calculation. Among them, different oral medical image samples may have different voxel spacings due to different oral scanning devices or device acquisition protocols. In order to ensure that the model can learn the correct spatial semantics, the voxel spacing of the oral medical image samples is resampled to the median voxel spacing. Among them, the oral medical image samples after resampling are normalized by the voxel size of the oral medical image samples after resampling, which can avoid the adverse effects of singular data and accelerate model convergence. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0050] In a specific example, when resampling the cropped oral medical image sample according to the voxel spacing of the cropped oral medical image sample, the ratio of the maximum axis spacing of the voxels of the cropped oral medical image sample to the minimum axis spacing of the voxels of the cropped oral medical image sample is calculated; if the ratio is less than or equal to 3, cubic linear interpolation is used to resample the cropped oral medical image sample; if the ratio is greater than 3, cubic linear interpolation is used to resample the plane of the cropped oral medical image sample, and the nearest neighbor algorithm is used to resample the plane of the cropped oral medical image sample. When normalizing the resampled oral medical image sample according to the voxel size of the resampled oral medical image sample, the mean value and standard deviation of the voxel size of the resampled oral medical image sample are calculated; the resampled oral medical image sample is normalized according to the mean value and standard deviation of the voxel size of the sampled oral medical image sample to obtain the normalized oral medical image sample. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0051] In some optional embodiments, the method further includes: adaptively selecting a data enhancement method for the second training set through the second adaptive module; performing data enhancement on the second training set according to the data enhancement method to obtain the second training set after data enhancement, and training the tooth partition prediction module according to the second training set and the second pipeline fingerprint, including: training the tooth partition prediction module according to the second training set after data enhancement and the second pipeline fingerprint. Thereby, there is no need to manually rely on prior knowledge to select a data enhancement method for the second training set, but the second adaptive module adaptively selects a data enhancement method for the second training set, and performs data enhancement on the second training set according to the data enhancement method, which not only effectively improves the efficiency of data enhancement on the second training set, but also effectively improves the accuracy of data enhancement on the second training set. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0052] In a specific example, the data enhancement method includes rotation scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing. When the rotation scaling method is adaptively selected for the second training set, the rotation scaling method of the second training set that is fixed and set can be determined as the rotation scaling method of the second training set that is adaptively generated through the second adaptive module. When the Gaussian noise method is adaptively selected for the second training set, the Gaussian noise method of the second training set that is fixed and set can be determined as the Gaussian noise method of the second training set that is adaptively generated through the second adaptive module. When the Gaussian blur method is adaptively selected for the second training set, the Gaussian blur method of the second training set that is fixed and set can be determined as the Gaussian blur method of the second training set that is adaptively generated through the second adaptive module. When the brightness processing method is adaptively selected for the second training set, the brightness processing method of the second training set that is fixed and set can be determined as the brightness processing method of the second training set that is adaptively generated through the second adaptive module. When the contrast processing method is adaptively selected for the second training set, the contrast processing method of the second training set that is fixed and set can be determined as the contrast processing method of the second training set that is adaptively generated through the second adaptive module. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0053] In a specific example, the data enhancement method includes rotation and scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing. When data enhancement is performed on the second training set according to the data enhancement method, the oral medical image samples in the second training set are rotated and scaled to obtain the oral medical image samples after rotation and scaling, and the second training set after data enhancement is obtained according to the oral medical image samples after rotation and scaling; or, Gaussian noise is performed on the oral medical image samples in the second training set to obtain the oral medical image samples after Gaussian noise, and the second training set after data enhancement is obtained according to the oral medical image samples after Gaussian noise; or, Gaussian blur is performed on the oral medical image samples in the second training set to obtain the oral medical image samples after Gaussian blur, and the second training set after data enhancement is obtained according to the oral medical image samples after Gaussian blur; or, brightness processing is performed on the oral medical image samples in the second training set to obtain the oral medical image samples after brightness processing, and the second training set after data enhancement is obtained according to the oral medical image samples after brightness processing; or, contrast processing is performed on the oral medical image samples in the second training set to obtain the oral medical image samples after contrast processing, and the second training set after data enhancement is obtained according to the oral medical image samples after contrast processing. Among them, the Gaussian noise addition can be understood as adding noise to the oral medical image samples, and the probability density function of the noise obeys the Gaussian distribution. The Gaussian blur can be understood as actually filtering the oral medical image samples. The filtering method is Gaussian filtering, which is image blurring to extract important information of the oral medical image samples. The brightness processing can be understood as changing the brightness of the oral medical image samples, and the contrast processing can be understood as changing the contrast of the oral medical image samples. The more and more diverse the oral medical image samples in the second training set, the more conducive to model training and improving the generalization ability of the model. In order to train the best possible model on the limited second training set, the oral medical image samples in the second training set are expanded by data enhancement to improve the quality of the training samples, and increase the noise data to improve the robustness of the model. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0054] In some optional embodiments, the second pipeline fingerprint includes identification information of the network to be trained in the tooth partition prediction module and network hyperparameters of the network to be trained. When the tooth partition prediction module is trained according to the second training set and the second pipeline fingerprint, the network to be trained is determined according to the identification information, and the network to be trained is set according to the network hyperparameters to obtain the set network to be trained; the tooth partition in the oral medical image sample in the second training set is predicted by the set network to be trained to obtain the tooth partition prediction result of the oral medical image sample; the set network to be trained is trained according to the tooth partition prediction result of the oral medical image sample and the tooth partition labeling result of the oral medical image sample. Thereby, the trained network can accurately perform tooth partitioning on the target oral medical image. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0055] In a specific example, the tooth partition prediction result of the oral medical image sample can be understood as the predicted probability data of the tooth partition label to which the voxel in the oral medical image sample belongs, and the tooth partition labeling result of the oral medical image sample can be understood as the labeling probability data of the tooth partition label to which the voxel in the oral medical image sample belongs. When the network to be trained after setting is trained according to the tooth partition prediction result of the oral medical image sample and the tooth partition labeling result of the oral medical image sample, the difference value between the predicted probability data and the labeling probability data is determined by the target loss function; based on the difference value, the parameters of the network to be trained after setting are adjusted. Among them, the target loss function can be the sum of Dice loss and cross entropy loss. When adjusting the parameters of the network to be trained after setting, the back propagation algorithm or the stochastic gradient descent algorithm can be used to adjust the parameters of the network to be trained after setting. It can be understood that the above description is only exemplary, and the embodiments of the present application do not impose any limitation on this.

[0056] In a specific example, by determining the difference value between the predicted probability data and the labeled probability data, the currently obtained predicted probability data is evaluated as a basis for the subsequent training of the network to be trained after the setting. Specifically, the difference value can be transmitted back to the network to be trained after the setting, so as to iteratively train the network to be trained after the setting. The training of the network to be trained after the setting is an iterative process. This embodiment only describes one training process, but those skilled in the art should understand that this training method can be used for each training of the network to be trained after the setting until the training of the network to be trained after the setting is completed. It can be understood that the above description is only exemplary, and the embodiments of the present application do not impose any limitation on this.

[0057] In step S104, the classification of teeth in the target oral medical image is predicted by the trained tooth classification prediction module, and the partition of teeth in the target oral medical image is predicted by the trained tooth partition prediction module.

[0058] In some optional embodiments, the method further includes: adaptively selecting an inference mode for the trained tooth classification prediction module through the first adaptive module, wherein the classification of teeth in the target oral medical image is predicted through the trained tooth classification prediction module, including: predicting the classification of teeth in the target oral medical image according to the inference mode through the trained tooth classification prediction module to obtain a tooth classification prediction result of the target oral medical image. Thus, there is no need to manually rely on prior knowledge to select an inference mode for the trained tooth classification prediction module, but the classification of teeth in the target oral medical image is predicted according to the inference mode through the trained tooth classification prediction module, which not only effectively improves the efficiency of the inference of the trained tooth classification prediction module, but also effectively improves the accuracy of the inference of the trained tooth classification prediction module. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0059] In a specific example, when the tooth classification prediction module is U-Net, the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet. You can select the best / certain network among the three networks for inference, or you can select the average of the results of the three networks for inference; the best network is selected for inference by default. When the inference mode is adaptively selected for the trained tooth classification prediction module through the first adaptive module, the fixed inference mode is determined as the inference mode of the trained tooth classification prediction module through the first adaptive module. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0060] In some optional embodiments, the method further includes: adaptively selecting an inference mode for the trained tooth partition prediction module through the second adaptive module, wherein the tooth partition prediction module predicts the tooth partition in the target oral medical image, including: predicting the tooth partition in the target oral medical image according to the inference mode through the trained tooth partition prediction module to obtain the tooth partition prediction result of the target oral medical image. Thereby, there is no need to manually rely on prior knowledge to select an inference mode for the trained tooth partition prediction module, but the tooth partition in the target oral medical image is predicted according to the inference mode through the trained tooth partition prediction module, which not only effectively improves the efficiency of the inference of the trained tooth partition prediction module, but also effectively improves the accuracy of the inference of the trained tooth partition prediction module. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0061] In a specific example, when the tooth partition prediction module is U-Net, the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet. The best / certain network among the three networks can be selected for inference, or the average value of the results of the three networks can be selected for inference; the best network is selected for inference by default. When the inference mode is adaptively selected for the trained tooth partition prediction module through the second adaptive module, the fixed inference mode is determined as the inference mode of the trained tooth partition prediction module through the second adaptive module. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0062] In some optional embodiments, when the classification of teeth in the target oral medical image is predicted by the trained tooth classification prediction module, the target oral medical image is downsampled by multiple cascaded downsampling units in the trained tooth classification prediction module to obtain multiple first feature maps of different scales of the target oral medical image; and the last first feature map obtained by downsampling the target oral medical image is upsampled by multiple cascaded upsampling units in the trained tooth classification prediction module according to the multiple first feature maps of different scales of the target oral medical image to obtain the tooth classification prediction result of the target oral medical image. The downsampling unit includes a cascaded convolution layer and a pooling layer, and the upsampling unit includes a cascaded upsampling layer and a deconvolution layer. In this way, the tooth classification prediction result of the target oral medical image can be accurately obtained. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0063] In some optional embodiments, when predicting the tooth partition in the target oral medical image through the trained tooth partition prediction module, the target oral medical image is downsampled through multiple cascaded downsampling units in the trained tooth partition prediction module to obtain multiple second feature maps of different scales of the target oral medical image; and the last second feature map obtained by downsampling the target oral medical image is upsampled according to the multiple second feature maps of different scales of the target oral medical image through multiple cascaded upsampling units in the trained tooth partition prediction module to obtain the tooth partition prediction result of the target oral medical image. The downsampling unit includes a cascaded convolution layer and a pooling layer, and the upsampling unit includes a cascaded upsampling layer and a deconvolution layer. In this way, the tooth partition prediction result of the target oral medical image can be accurately obtained. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0064] In step S105, a tooth segmentation result of the target oral medical image is determined according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image.

[0065] In some optional embodiments, the method further includes: adaptively selecting a first post-processing method for the tooth classification prediction result through the first adaptive module; adaptively selecting a second post-processing method for the tooth partition prediction result through the second adaptive module, and determining the tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image, including: post-processing the tooth classification prediction result of the target oral medical image through the first post-processing method to obtain the post-processed tooth classification prediction result; post-processing the tooth partition prediction result of the target oral medical image through the second post-processing method to obtain the post-processed tooth partition prediction result; fusing the post-processed tooth classification prediction result and the post-processed tooth partition prediction result to obtain the tooth segmentation result of the target oral medical image. In this way, the tooth segmentation result of the target oral medical image can be accurately obtained. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0066] In a specific example, the first post-processing method includes a connected domain analysis, the tooth classification prediction result includes probability data of the tooth classification label to which the voxels in the target oral medical image belong, and the post-processed tooth classification prediction result includes the tooth classification label to which the tooth area of ​​the target oral medical image belongs. When the tooth classification prediction result of the target oral medical image is post-processed by the first post-processing method, a connected domain analysis is performed on the probability data of the tooth classification label to which the voxels in the target oral medical image belong, so as to obtain the tooth classification label to which the tooth area of ​​the target oral medical image belongs. Thereby, the tooth classification label to which the tooth area of ​​the target oral medical image belongs can be accurately obtained. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0067] In a specific example, the second post-processing method includes a connected domain analysis, the tooth partition prediction result includes probability data of the tooth partition label to which the voxels in the target oral medical image belong, and the post-processed tooth partition prediction result includes the tooth partition label to which the tooth area of ​​the target oral medical image belongs. When the tooth partition prediction result of the target oral medical image is post-processed by the second post-processing method, a connected domain analysis is performed on the probability data of the tooth partition label to which the voxels in the target oral medical image belong, so as to obtain the tooth partition label to which the tooth area of ​​the target oral medical image belongs. Thereby, the tooth partition label to which the tooth area of ​​the target oral medical image belongs can be accurately obtained. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0068] In a specific example, the post-processed tooth classification prediction result includes the tooth classification label to which the tooth region of the target oral medical image belongs, and the post-processed tooth partition prediction result includes the tooth partition label to which the tooth region of the target oral medical image belongs. When the post-processed tooth classification prediction result and the post-processed tooth partition prediction result are fused, the tooth classification label to which the tooth region of the target oral medical image belongs and the tooth partition label to which the tooth region of the target oral medical image belongs are fused to obtain the tooth segmentation label to which the tooth region of the target oral medical image belongs; the target oral medical image is segmented according to the tooth segmentation label to which the tooth region of the target oral medical image belongs to obtain the tooth segmentation result of the target oral medical image. Thus, by segmenting the target oral medical image according to the tooth segmentation label to which the tooth region of the target oral medical image belongs, the tooth segmentation result of the target oral medical image can be obtained more accurately. It can be understood that the above description is only exemplary, and this embodiment does not impose any limitation on this.

[0069] In a specific example, the tooth classification label of the tooth region of the target oral medical image and the tooth partition label of the tooth region of the target oral medical image are fused, the tooth classification label is 7 categories, the labels are 1-7, the tooth partition label is 4 categories, the labels are 1-4, and the tooth segmentation label obtained after fusion is 28 categories, the labels are 11-17, 21-27, 31-37, 41-47. Specifically, the upper left 7 teeth are labeled 11-17, the lower left 7 teeth are labeled 21-27, the upper right 7 teeth are labeled 31-37, and the lower right 7 teeth are labeled 41-47. Divide tooth segmentation into two tasks: 7-category tooth classification and 4-category tooth partitioning. This makes better use of data than direct 28-category classification. The 7-category tooth classification task can extract 4 groups of teeth from one image, while 28 categories can only extract one group of teeth from one image. The amount of data for 7-category tooth segmentation is much larger than that for 28 categories. This is more conducive to the learning and training of deep models. Segmenting all teeth by classification and partitioning is more accurate than direct segmentation. By fusing tooth classification labels and tooth partitioning labels to perform tooth segmentation, the deep model can make better use of limited data, thereby improving the accuracy of the model. Convert the fused data format from nii.gz format to dicom format that can be read by the software, and display the results of tooth segmentation in the software. It can be understood that the above description is only exemplary and this embodiment does not impose any limitation on this.

[0070] In specific practical applications, the tooth segmentation method for oral medical images provided in this embodiment can use the nnU-NET deep model to automatically select appropriate pre-processing and post-processing methods for oral medical images, and at the same time can select appropriate networks to train oral medical image samples, avoiding reliance on prior knowledge like traditional tooth segmentation algorithms or other deep learning algorithms that require manual intervention, thereby improving the efficiency and accuracy of tooth segmentation, enabling oral surgical robots to automatically segment and display teeth in medical images, and helping the robot to accurately locate teeth in subsequent tasks and perform high-precision dental implant operations.

[0071] The tooth segmentation method for oral medical images provided by this embodiment does not require manual preparation of pipeline fingerprints for the tooth classification prediction module to be trained and the tooth partition prediction module to be trained. Instead, the first adaptive module in the tooth classification prediction model adaptively generates the first pipeline fingerprint for training the tooth classification prediction module according to the first data fingerprint of the first training set, and the second adaptive module in the tooth partition prediction model adaptively generates the second pipeline fingerprint for training the tooth partition prediction module according to the second data fingerprint of the second training set, thereby avoiding the trouble of manually preparing pipeline fingerprints, reducing the interference of human factors, and effectively improving the efficiency of tooth segmentation in oral medical images. In addition, the diversity of tooth segmentation results in oral medical images caused by manually preparing pipeline fingerprints is effectively avoided, thereby effectively improving the accuracy of tooth segmentation in oral medical images.

[0072] Furthermore, the classification of teeth in the target oral medical image is predicted by the trained tooth classification prediction module, and the partitioning of teeth in the target oral medical image is predicted by the trained tooth partitioning prediction module, and then the tooth segmentation result of the target oral medical image is determined based on the tooth classification prediction result and the tooth partitioning prediction result of the target oral medical image, which can effectively enable the trained tooth classification prediction module and the trained tooth partitioning prediction module to more fully utilize the limited data of the target oral medical image, thereby effectively improving the segmentation accuracy of teeth in the target oral medical image.

[0073] The tooth segmentation method of oral medical images provided in this embodiment can be executed by any appropriate device with data processing capabilities, including but not limited to: cameras, terminals, mobile terminals, PCs, servers, vehicle-mounted devices, entertainment equipment, advertising equipment, personal digital assistants (PDAs), tablet computers, laptop computers, handheld game consoles, smart glasses, smart watches, wearable devices, virtual display devices or display enhancement devices, etc.

[0074] Reference Figure 2, shows a schematic structural diagram of a tooth segmentation device for oral medical images according to a second embodiment of the present application.

[0075] The tooth segmentation device for oral medical images provided in this embodiment includes: a first determination module 201, which is used to determine a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model according to an oral medical image sample obtained by scanning an oral cavity with an oral scanning device; a first training module 202, which is used to adaptively generate a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth classification prediction model, and train the tooth classification prediction module according to the first training set and the first pipeline fingerprint; a second training module 203, which is used to adaptively generate a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth classification prediction model, and train the tooth classification prediction module according to the first training set and the first pipeline fingerprint; The second adaptive module in the tooth partition prediction model adaptively generates a second pipeline fingerprint for training the tooth partition prediction module according to the second data fingerprint of the second training set, and trains the tooth partition prediction module according to the second training set and the second pipeline fingerprint; the prediction module 204 is used to predict the classification of teeth in the target oral medical image through the trained tooth classification prediction module, and predict the partition of teeth in the target oral medical image through the trained tooth partition prediction module; the second determination module 205 is used to determine the tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image.

[0076] Optionally, the device also includes: a first selection module, used to adaptively select a preprocessing method for the first training set according to the first data fingerprint of the first training set through the first adaptive module; a first preprocessing module, used to preprocess the first training set according to the preprocessing method to obtain the preprocessed first training set, and the first training module 202 is specifically used to train the tooth classification prediction module according to the preprocessed first training set and the first pipeline fingerprint.

[0077] Optionally, the preprocessing method includes cropping, resampling and normalization, and the first preprocessing module is specifically used to: crop the oral medical image samples in the first training set to obtain the cropped oral medical image samples; resample the cropped oral medical image samples according to the voxel spacing of the cropped oral medical image samples to obtain the resampled oral medical image samples; normalize the resampled oral medical image samples according to the voxel size of the resampled oral medical image samples to obtain the normalized oral medical image samples, and obtain the preprocessed first training set based on the normalized oral medical image samples.

[0078] Optionally, the device also includes: a second selection module, used to adaptively select a preprocessing method for the second training set according to the second data fingerprint of the second training set through the second adaptive module; a second preprocessing module, used to preprocess the second training set according to the preprocessing method to obtain the preprocessed second training set; the second training module 203 is specifically used to train the tooth partition prediction module according to the preprocessed second training set and the second pipeline fingerprint.

[0079] Optionally, the preprocessing method includes cropping, resampling and normalization, and the second preprocessing module is specifically used to: crop the oral medical image samples in the second training set to obtain the cropped oral medical image samples; resample the cropped oral medical image samples according to the voxel spacing of the cropped oral medical image samples to obtain the resampled oral medical image samples; normalize the resampled oral medical image samples according to the voxel size of the resampled oral medical image samples to obtain the normalized oral medical image samples, and obtain the preprocessed second training set based on the normalized oral medical image samples.

[0080] Optionally, the device also includes: a third selection module, used to adaptively select a data enhancement method for the first training set through the first adaptive module; a first data enhancement module, used to perform data enhancement on the first training set according to the data enhancement method to obtain the first training set after data enhancement, and the first training module 202 is specifically used to: train the tooth classification prediction module according to the first training set after data enhancement and the first pipeline fingerprint.

[0081] Optionally, the data enhancement method includes rotation and scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing, and the first data enhancement module is specifically used to: rotate and scale the oral medical image samples in the first training set to obtain the rotated and scaled oral medical image samples, and obtain the first training set after data enhancement according to the rotated and scaled oral medical image samples; or, perform Gaussian noise on the oral medical image samples in the first training set to obtain the Gaussian noised oral medical image samples, and obtain the first training set after data enhancement according to the Gaussian noised oral medical image samples; or, The oral medical image samples are Gaussian blurred to obtain the oral medical image samples after Gaussian blur, and the first training set after data enhancement is obtained according to the oral medical image samples after Gaussian blur; or, the oral medical image samples in the first training set are brightness processed to obtain the oral medical image samples after brightness processing, and the first training set after data enhancement is obtained according to the oral medical image samples after brightness processing; or, the oral medical image samples in the first training set are contrast processed to obtain the oral medical image samples after contrast processing, and the first training set after data enhancement is obtained according to the oral medical image samples after contrast processing.

[0082] Optionally, the device also includes: a fourth selection module, used to adaptively select a data enhancement method for the second training set through the second adaptive module; a second data enhancement module, used to perform data enhancement on the second training set according to the data enhancement method to obtain the second training set after data enhancement; the second training module 203 is specifically used to train the tooth partition prediction module according to the second training set after data enhancement and the second pipeline fingerprint.

[0083] Optionally, the data enhancement method includes rotation and scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing, and the second data enhancement module is specifically used to: rotate and scale the oral medical image samples in the second training set to obtain the rotated and scaled oral medical image samples, and obtain the second training set after data enhancement based on the rotated and scaled oral medical image samples; or, perform Gaussian noise on the oral medical image samples in the second training set to obtain the Gaussian noised oral medical image samples, and obtain the second training set after data enhancement based on the Gaussian noised oral medical image samples; or, perform Gaussian noise on the oral medical image samples in the second training set to obtain the Gaussian noised oral medical image samples, and obtain the Gaussian noised oral medical image samples; The oral medical image samples are Gaussian blurred to obtain the oral medical image samples after Gaussian blur, and the second training set after data enhancement is obtained according to the oral medical image samples after Gaussian blur; or, the oral medical image samples in the second training set are brightness processed to obtain the oral medical image samples after brightness processing, and the second training set after data enhancement is obtained according to the oral medical image samples after brightness processing; or, the oral medical image samples in the second training set are contrast processed to obtain the oral medical image samples after contrast processing, and the second training set after data enhancement is obtained according to the oral medical image samples after contrast processing.

[0084] Optionally, the first pipeline fingerprint includes identification information of the network to be trained in the tooth classification prediction module and network hyperparameters of the network to be trained. The first training module 202 is specifically used to: determine the network to be trained according to the identification information, and set the network to be trained according to the network hyperparameters to obtain the set network to be trained; predict the classification of teeth in the oral medical image samples in the first training set through the set network to be trained to obtain the tooth classification prediction results of the oral medical image samples; train the set network to be trained according to the tooth classification prediction results of the oral medical image samples and the tooth classification labeling results of the oral medical image samples.

[0085] Optionally, the second pipeline fingerprint includes identification information of the network to be trained in the tooth partition prediction module and network hyperparameters of the network to be trained. The second training module 203 is specifically used to: determine the network to be trained according to the identification information, and set the network to be trained according to the network hyperparameters to obtain the set network to be trained; predict the partition of teeth in the oral medical image samples in the second training set through the set network to be trained to obtain the tooth partition prediction results of the oral medical image samples; train the set network to be trained according to the tooth partition prediction results of the oral medical image samples and the tooth partition labeling results of the oral medical image samples.

[0086] Optionally, the device also includes: a fifth selection module, which is used to adaptively select an inference mode for the trained tooth classification prediction module through the first adaptive module, and the prediction module 204 is specifically used to: predict the classification of teeth in the target oral medical image according to the inference mode through the trained tooth classification prediction module to obtain a tooth classification prediction result of the target oral medical image.

[0087] Optionally, the device also includes: a sixth selection module, which is used to adaptively select an inference mode for the trained tooth partition prediction module through the second adaptive module, and the prediction module 204 is specifically used to: predict the partition of teeth in the target oral medical image according to the inference mode through the trained tooth partition prediction module to obtain the tooth partition prediction result of the target oral medical image.

[0088] Optionally, the prediction module 204 is specifically used to: downsample the target oral medical image through multiple cascaded downsampling units in the trained tooth classification prediction module to obtain multiple first feature maps of different scales of the target oral medical image; upsample the last first feature map obtained by downsampling the target oral medical image through multiple cascaded upsampling units in the trained tooth classification prediction module according to the multiple first feature maps of different scales of the target oral medical image to obtain a tooth classification prediction result of the target oral medical image.

[0089] Optionally, the prediction module 204 is specifically used to: downsample the target oral medical image through multiple cascaded downsampling units in the trained dental partition prediction module to obtain multiple second feature maps of different scales of the target oral medical image; upsample the last second feature map obtained by downsampling the target oral medical image through multiple cascaded upsampling units in the trained dental partition prediction module according to the multiple second feature maps of different scales of the target oral medical image to obtain a dental partition prediction result of the target oral medical image.

[0090] Optionally, the device also includes: a seventh selection module, used to adaptively select a first post-processing method for the tooth classification prediction result through the first adaptive module; an eighth selection module, used to adaptively select a second post-processing method for the tooth partition prediction result through the second adaptive module, and the second determination module 205 includes: a first post-processing submodule, used to post-process the tooth classification prediction result of the target oral medical image through the first post-processing method to obtain the post-processed tooth classification prediction result; a second post-processing submodule, used to post-process the tooth partition prediction result of the target oral medical image through the second post-processing method to obtain the post-processed tooth partition prediction result; a fusion submodule, used to fuse the post-processed tooth classification prediction result and the post-processed tooth partition prediction result to obtain the tooth segmentation result of the target oral medical image.

[0091] Optionally, the first post-processing method includes connected domain analysis, the tooth classification prediction result includes probability data of the tooth classification label to which the voxels in the target oral medical image belong, and the post-processed tooth classification prediction result includes the tooth classification label to which the tooth area of ​​the target oral medical image belongs. The first post-processing submodule is specifically used to perform connected domain analysis on the probability data of the tooth classification label to which the voxels in the target oral medical image belong, so as to obtain the tooth classification label to which the tooth area of ​​the target oral medical image belongs.

[0092] Optionally, the second post-processing method includes connected domain analysis, the tooth partition prediction result includes probability data of the tooth partition label to which the voxels in the target oral medical image belong, and the post-processed tooth partition prediction result includes the tooth partition label to which the tooth area of ​​the target oral medical image belongs. The second post-processing submodule is specifically used to perform connected domain analysis on the probability data of the tooth partition label to which the voxels in the target oral medical image belong, so as to obtain the tooth partition label to which the tooth area of ​​the target oral medical image belongs.

[0093] Optionally, the post-processed tooth classification prediction result includes the tooth classification label to which the tooth area of ​​the target oral medical image belongs, and the post-processed tooth partitioning prediction result includes the tooth partitioning label to which the tooth area of ​​the target oral medical image belongs, and the fusion submodule is specifically used to: fuse the tooth classification label to which the tooth area of ​​the target oral medical image belongs and the tooth partitioning label to which the tooth area of ​​the target oral medical image belongs to obtain the tooth segmentation label to which the tooth area of ​​the target oral medical image belongs; and segment the target oral medical image according to the tooth segmentation label to which the tooth area of ​​the target oral medical image belongs to obtain the tooth segmentation result of the target oral medical image.

[0094] The tooth segmentation device for oral medical images provided in this embodiment is used to implement the corresponding tooth segmentation methods for oral medical images in the aforementioned multiple method embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0095] Figure 3 This is a schematic diagram of the structure of an electronic device in Embodiment 3 of the present application; the electronic device may include:

[0096] One or more processors 301;

[0097] The computer readable medium 302 may be configured to store one or more programs,

[0098] When the one or more programs are executed by the one or more processors, the one or more processors implement the tooth segmentation method for oral medical images as described in the first embodiment.

[0099] Figure 4 The hardware structure of the electronic device in the fourth embodiment of the present application; Figure 4 As shown, the hardware structure of the electronic device may include: a processor 401, a communication interface 402, a computer readable medium 403 and a communication bus 404;

[0100] The processor 401, the communication interface 402, and the computer readable medium 403 communicate with each other via a communication bus 404;

[0101] Optionally, the communication interface 402 may be an interface of a communication module, such as an interface of a GSM module;

[0102] The processor 401 can be specifically configured to: determine a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model according to an oral medical image sample obtained by scanning an oral cavity with an oral scanning device; adaptively generate a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth classification prediction model, and train the tooth classification prediction module according to the first training set and the first pipeline fingerprint; adaptively generate a second pipeline fingerprint for training the tooth partition prediction module according to a second data fingerprint of the second training set through a second adaptive module in the tooth partition prediction model, and train the tooth partition prediction module according to the second training set and the second pipeline fingerprint; predict the classification of teeth in a target oral medical image through the trained tooth classification prediction module, and predict the partition of teeth in the target oral medical image through the trained tooth partition prediction module; determine a tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image.

[0103] The processor 401 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0104] The computer readable medium 403 can be, but is not limited to, a random access memory medium (RAM), a read-only memory medium (ROM), a programmable read-only memory medium (PROM), an erasable programmable read-only memory medium (EPROM), an electrically erasable read-only memory medium (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.

[0105] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code configured to execute the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed. It should be noted that the computer-readable medium described in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more conductors, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, an apparatus or a device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program configured to be used by or in combination with an instruction execution system, an apparatus or a device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0106] Computer program code configured to perform the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0107] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of the code contains one or more executable instructions configured to implement the specified logical function. There are specific sequential relationships in the above-mentioned specific embodiments, but these sequential relationships are only exemplary. When the specific implementation is implemented, these steps may be less, more, or the execution order may be adjusted. That is, in some implementations as replacements, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0108] The modules involved in the embodiments described in the present application may be implemented by software or by hardware. The modules described may also be set in a processor, for example, may be described as: a processor includes a first determination module, a first training module, a second training module, a prediction module, and a second determination module. Among them, the names of these modules do not constitute a limitation on the module itself under certain circumstances. For example, the first determination module may also be described as "a module for determining a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model based on oral medical image samples obtained by scanning the oral cavity with an oral scanning device."

[0109] As another aspect, the present application further provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method for teeth segmentation of oral medical images as described in the first embodiment above.

[0110] As another aspect, the present application further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the device, the device: determines a first training set for training a tooth classification prediction module in a tooth classification prediction model and a second training set for training a tooth partition prediction module in a tooth partition prediction model according to an oral medical image sample obtained by scanning an oral cavity with an oral scanning device; adaptively generates a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth classification prediction model, and trains the tooth classification prediction module according to the first training set and the first pipeline fingerprint; adaptively generates a second pipeline fingerprint for training the tooth partition prediction module according to a second data fingerprint of the second training set through a second adaptive module in the tooth partition prediction model, and trains the tooth partition prediction module according to the second training set and the second pipeline fingerprint; predicts the classification of teeth in a target oral medical image through the trained tooth classification prediction module, and predicts the partition of teeth in the target oral medical image through the trained tooth partition prediction module; determines a tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image.

[0111] The expressions "first", "second", "the first" or "the second" used in various embodiments of the present disclosure may modify various components regardless of order and / or importance, but these expressions do not limit the corresponding components. The above expressions are only configured for the purpose of distinguishing an element from other elements. For example, a first user device and a second user device represent different user devices, although both are user devices. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element without departing from the scope of the present disclosure.

[0112] When one element (e.g., a first element) is referred to as being "(operably or communicatively) coupled" or "(operably or communicatively) coupled to" or "connected to" another element (e.g., a second element), it is understood that the one element is directly connected to the other element or the one element is indirectly connected to the other element via yet another element (e.g., a third element). Conversely, it is understood that when an element (e.g., a first element) is referred to as being "directly connected" or "directly coupled" to another element (the second element), no element (e.g., a third element) is interposed between the two.

[0113] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A method for teeth segmentation in oral medical images, characterized in that: The method comprises: Determine, based on oral medical image samples obtained by scanning the oral cavity with an oral scanning device, a first training set for training a tooth classification prediction module in a tooth classification prediction module and a second training set for training a tooth partition prediction module in a tooth partition prediction module; Load oral medical images from CBCT to obtain a dataset of oral medical image samples; A first pipeline fingerprint for training the tooth classification prediction module is adaptively generated according to a first data fingerprint of the first training set by a first adaptive module in the tooth classification prediction module, and the tooth classification prediction module is trained according to the first training set and the first pipeline fingerprint, wherein the first training set includes oral medical image samples and tooth classification labels thereof for training the tooth classification prediction module, the tooth classification labels include 7 categories, and the first data fingerprint is an attribute of the first training set; The tooth classification prediction module is U-Net, and the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet; The first pipeline fingerprint includes identification information of the network to be trained in the tooth classification prediction module and network hyperparameters of the network to be trained; By means of a second adaptive module in the tooth partition prediction module, a second fingerprint of the second training set is adaptively generated for training the tooth partition prediction module. Second pipeline fingerprint , and according to the second training set and the second pipeline fingerprint, Tooth Partition Prediction Module Training is performed, wherein the tooth partition prediction module is U-Net, and the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet, the second training set includes oral medical image samples and tooth partition labels thereof for training the tooth partition prediction module, the tooth partition labels include 4 categories, the second data fingerprint is the attribute of the second training set, and the second pipeline fingerprint includes identification information of the network to be trained in the tooth partition prediction module and network hyperparameters of the network to be trained; Predicting the classification of teeth in the target oral medical image by the trained tooth classification prediction module, and predicting the partition of teeth in the target oral medical image by the trained tooth partition prediction module; According to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image, the tooth segmentation result of the target oral medical image is determined, and the tooth segmentation result is the tooth segmentation label obtained after fusion, which is 28 categories.

2. The method for teeth segmentation of oral medical images according to claim 1, characterized in that: The method further comprises: Adaptively selecting a preprocessing method for the first training set according to a first data fingerprint of the first training set by the first adaptive module; Preprocessing the first training set according to the preprocessing method to obtain the preprocessed first training set, The step of training the tooth classification prediction module according to the first training set and the first pipeline fingerprint includes: The tooth classification prediction module is trained according to the preprocessed first training set and the first pipeline fingerprint.

3. The method for teeth segmentation of oral medical images according to claim 2, characterized in that: The preprocessing methods include cropping, resampling and normalization. Preprocessing the first training set according to the preprocessing method to obtain the preprocessed first training set includes: Cropping the oral medical image samples in the first training set to obtain the cropped oral medical image samples; Resampling the cropped oral medical image sample according to the voxel spacing of the cropped oral medical image sample to obtain the resampled oral medical image sample; According to the voxel size of the resampled oral medical image samples, the resampled oral medical image samples are normalized to obtain the normalized oral medical image samples, and according to the normalized oral medical image samples, the first preprocessed training set is obtained.

4. The method for teeth segmentation of oral medical images according to claim 1, characterized in that: The method further comprises: Adaptively selecting a data enhancement method for the first training set through the first adaptive module; According to the data enhancement method, data enhancement is performed on the first training set to obtain the first training set after data enhancement, The step of training the tooth classification prediction module according to the first training set and the first pipeline fingerprint includes: The tooth classification prediction module is trained according to the first training set and the first pipeline fingerprint after data enhancement.

5. The method for teeth segmentation of oral medical images according to claim 4, characterized in that: The data enhancement method includes rotation and scaling, Gaussian noise, Gaussian blur, brightness processing, or contrast processing. The step of performing data enhancement on the first training set according to the data enhancement method to obtain the first training set after data enhancement includes: Rotate and scale the oral medical image samples in the first training set to obtain the rotated and scaled oral medical image samples, and obtain the first training set after data enhancement based on the rotated and scaled oral medical image samples; or Performing Gaussian noise addition on the oral medical image samples in the first training set to obtain the oral medical image samples after Gaussian noise addition, and obtaining the first training set after data enhancement based on the oral medical image samples after Gaussian noise addition; or, Performing Gaussian blurring on the oral medical image samples in the first training set to obtain the oral medical image samples after Gaussian blurring, and obtaining the first training set after data enhancement based on the oral medical image samples after Gaussian blurring; or Performing brightness processing on the oral medical image samples in the first training set to obtain the oral medical image samples after brightness processing, and obtaining the first training set after data enhancement based on the oral medical image samples after brightness processing; or Performing contrast processing on the oral medical image samples in the first training set to obtain the oral medical image samples after contrast processing, and obtaining the first training set after data enhancement based on the oral medical image samples after contrast processing.

6. The method for teeth segmentation of oral medical images according to claim 1, characterized in that: The step of training the tooth classification prediction module according to the first training set and the first pipeline fingerprint includes: Determine the network to be trained according to the identification information, and set the network to be trained according to the network hyperparameters to obtain the set network to be trained; Predicting the classification of teeth in the oral medical image samples in the first training set by using the set network to be trained, so as to obtain a tooth classification prediction result of the oral medical image samples; The set network to be trained is trained according to the tooth classification prediction result of the oral medical image sample and the tooth classification labeling result of the oral medical image sample.

7. The method for teeth segmentation of oral medical images according to claim 1, characterized in that: The method further comprises: Adaptively selecting an inference mode for the trained tooth classification prediction module through the first adaptive module, The trained tooth classification prediction module predicts the classification of teeth in the target oral medical image, including: The trained tooth classification prediction module is used to predict the classification of teeth in the target oral medical image according to the inference mode to obtain a tooth classification prediction result of the target oral medical image.

8. The method for teeth segmentation of oral medical images according to claim 1, characterized in that: The trained tooth classification prediction module predicts the classification of teeth in the target oral medical image, including: Downsampling the target oral medical image through a plurality of cascaded downsampling units in the trained tooth classification prediction module to obtain a plurality of first feature maps of different scales of the target oral medical image; Through the multiple cascaded upsampling units in the trained tooth classification prediction module, the last first feature map obtained by downsampling the target oral medical image is upsampled according to the multiple first feature maps of different scales of the target oral medical image to obtain the tooth classification prediction result of the target oral medical image.

9. The method for teeth segmentation of oral medical images according to claim 1, characterized in that: The method further comprises: Adaptively selecting a first post-processing method for the tooth classification prediction result through the first adaptive module; Adaptively selecting a second post-processing method for the tooth partition prediction result through the second adaptive module, The step of determining the tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image comprises: Post-processing the tooth classification prediction result of the target oral medical image by the first post-processing method to obtain the post-processed tooth classification prediction result; Post-processing the tooth partition prediction result of the target oral medical image by the second post-processing method to obtain the post-processed tooth partition prediction result; The post-processed tooth classification prediction result and the post-processed tooth partition prediction result are fused to obtain a tooth segmentation result of the target oral medical image.

10. The method for teeth segmentation of oral medical images according to claim 9, characterized in that: The first post-processing method includes connected domain analysis, the tooth classification prediction result includes probability data of tooth classification labels to which voxels in the target oral medical image belong, and the post-processed tooth classification prediction result includes tooth classification labels to which tooth regions in the target oral medical image belong, The post-processing of the tooth classification prediction result of the target oral medical image by the first post-processing method to obtain the post-processed tooth classification prediction result includes: Connected domain analysis is performed on probability data of tooth classification labels to which voxels in the target oral medical image belong, so as to obtain tooth classification labels to which tooth regions in the target oral medical image belong.

11. The method for teeth segmentation of oral medical images according to claim 9, characterized in that: The second post-processing method includes connected domain analysis, the tooth partition prediction result includes probability data of tooth partition labels to which voxels in the target oral medical image belong, and the post-processed tooth partition prediction result includes tooth partition labels to which tooth regions in the target oral medical image belong, The post-processing of the tooth partition prediction result of the target oral medical image by the second post-processing method to obtain the post-processed tooth partition prediction result includes: Connected domain analysis is performed on probability data of tooth partition labels to which voxels in the target oral medical image belong, so as to obtain tooth partition labels to which the tooth regions of the target oral medical image belong.

12. The method for teeth segmentation of oral medical images according to claim 9, characterized in that: The post-processed tooth classification prediction result includes a tooth classification label to which the tooth region of the target oral medical image belongs, and the post-processed tooth partition prediction result includes a tooth partition label to which the tooth region of the target oral medical image belongs, The step of fusing the post-processed tooth classification prediction result and the post-processed tooth partition prediction result to obtain the tooth segmentation result of the target oral medical image includes: Fusing the tooth classification label of the tooth region of the target oral medical image and the tooth partition label of the tooth region of the target oral medical image to obtain the tooth segmentation label of the tooth region of the target oral medical image; The target oral medical image is segmented according to the tooth segmentation label to which the tooth region of the target oral medical image belongs, so as to obtain a tooth segmentation result of the target oral medical image.

13. A tooth segmentation device for oral medical images, characterized in that: The device comprises: A first determination module is used to determine a first training set for training a tooth classification prediction module in a tooth classification prediction module and a second training set for training a tooth partition prediction module in a tooth partition prediction module according to an oral medical image sample obtained by scanning an oral cavity with an oral scanning device; A first training module is used to adaptively generate a first pipeline fingerprint for training the tooth classification prediction module according to a first data fingerprint of the first training set through a first adaptive module in the tooth classification prediction module, and train the tooth classification prediction module according to the first training set and the first pipeline fingerprint, wherein the first training set includes oral medical image samples and tooth classification labels thereof for training the tooth classification prediction module, wherein the tooth classification labels include 7 categories, and the first data fingerprint is an attribute of the first training set; The tooth classification prediction module is U-Net, and the U-Net includes three networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet; The first pipeline fingerprint includes identification information of the network to be trained in the tooth classification prediction module and network hyperparameters of the network to be trained; A second training module is used to adaptively generate a second pipeline fingerprint for training the tooth partition prediction module according to the second data fingerprint of the second training set through the second adaptive module in the tooth partition prediction module, and train the tooth partition prediction module according to the second training set and the second pipeline fingerprint, wherein the tooth partition prediction module is U-Net, and the U-Net includes three types of networks, namely 2D-Unet, 3D-Unet and 3D cascade Unet, the second training set includes oral medical image samples and tooth partition labels for training the tooth partition prediction module, and the tooth partition labels include 4 categories, the second data fingerprint is the attribute of the second training set, and the second pipeline fingerprint includes identification information of the network to be trained in the tooth partition prediction module and network hyperparameters of the network to be trained; A prediction module, used to predict the classification of teeth in the target oral medical image through the trained tooth classification prediction module, and to predict the partition of teeth in the target oral medical image through the trained tooth partition prediction module; The second determination module is used to determine the tooth segmentation result of the target oral medical image according to the tooth classification prediction result and the tooth partition prediction result of the target oral medical image, wherein the tooth segmentation result is the tooth segmentation label obtained after fusion, which is 28 categories.

14. An electronic device, characterized in that: The device comprises: one or more processors; A computer readable medium configured to store one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the tooth segmentation method for oral medical images as described in any one of claims 1 to 12.

15. A computer readable medium, characterized in that A computer program is stored thereon, and when the program is executed by a processor, the method for tooth segmentation of oral medical images as claimed in any one of claims 1 to 12 is implemented.

Images