Jaw cystic lesion identification method, device, equipment, medium and product

Abstract

The invention discloses a jaw cystic lesion recognition method, device and equipment, a medium and a product, and relates to the field of jaw cystic lesions. The method comprises the following steps: marking tooth and jaw regions in an oral cavity cone beam CT image, and selecting a region of interest to obtain an interested marked image; labeling cyst category labels generated by jaw cystic lesions in the interested labeling images, and forming a training set; constructing a deep learning model introducing an attention mechanism; the attention mechanism introduces relative positions of cysts and teeth as interested features; training a deep learning model by using the training set to obtain a jaw cystic lesion recognition model; and marking tooth and jaw regions in a to-be-identified oral cavity cone beam CT image of the patient with the jaw cystic lesion, inputting the marked image of interest after marking the tooth and jaw regions into the jaw cystic lesion identification model, and outputting the cyst category of the patient with the jaw cystic lesion. The accuracy and efficiency of jaw cystic lesion diagnosis can be improved.

Description

Technical Field

[0001] The present application relates to the field of jaw cystic lesions, and particularly to a method, device, equipment, medium and product for identifying jaw cystic lesions. Background Art

[0002] Jaw cystic lesions often present as painless bone swelling. Due to the lack of typical self-awareness symptoms in the early stage and children's inability to express their feelings accurately in a timely manner, the lesions have caused serious damage to the bone when most children seek medical treatment. It may involve the permanent tooth germ or the developing permanent teeth showing developmental disorders or displacements. In severe cases, there may even be tooth loss or occlusal disorders. For jaw lesions with similar imaging features but different behaviors, early and accurate imaging diagnosis is crucial for providing a reasonable treatment plan for patients, thereby effectively reducing the occurrence of bone defects and lowering the misdiagnosis rate of jaw tumors.

[0003] Traditional imaging diagnosis of jaw cystic lesions usually relies on doctors' experience and subjective judgment, resulting in diagnostic errors and differences among different doctors. Currently, Cone Beam Computed Tomography (CBCT) technology has been widely used in the field of oral medicine. CBCT has the advantages of high resolution, low radiation dose, three-dimensional imaging, etc. compared with panoramic films and CT, and can clearly display the anatomical structure and lesion characteristics of the jaw. Currently, the method of manual film reading is used to identify jaw cystic lesions. However, different types of cysts or tumors may have different density characteristics, and the density characteristics of these cysts are similar to those of the surrounding tissues, making it difficult to determine the boundary curve and thus difficult to distinguish. Moreover, the density and shape of cysts are diverse, the location is uncertain, and they may be interfered by artifacts. Therefore, the traditional manual film reading method has problems such as strong subjectivity, low efficiency, and inaccurate identification in identifying cysts. Summary of the Invention

[0004] The purpose of the present application is to provide a method, device, equipment, medium and product for identifying jaw cystic lesions, which can improve the accuracy and efficiency of diagnosing jaw cystic lesions.

[0005] To achieve the above purpose, the present application provides the following solutions.

[0006] In a first aspect, the present application provides a method for identifying jawbone cystic lesions, including: obtaining multiple oral cone beam CT images of multiple patients with jawbone cystic lesions; annotating the teeth and jawbone regions in each oral cone beam CT image; selecting regions of interest (ROIs) that include the annotated teeth and jawbone regions to obtain multiple annotated ROIs images; annotating the cyst category labels generated by the jawbone cystic lesions in each annotated ROIs image, and jointly forming a training set with all the annotated ROIs images with cyst category labels; constructing a deep learning model incorporating an attention mechanism; the attention mechanism introducing the relative positions of the cysts and teeth as the features of interest; training the deep learning model using the training set to obtain a jawbone cystic lesion identification model; annotating the teeth and jawbone regions in the oral cone beam CT image to be identified of a patient with jawbone cystic lesions, and inputting the annotated ROIs image of the oral cone beam CT image to be identified after annotating the teeth and jawbone regions into the jawbone cystic lesion identification model to output the cyst category of the patient with jawbone cystic lesions.

[0007] In a second aspect, the present application provides an apparatus for identifying jawbone cystic lesions, including: an image acquisition module, an annotation module, an ROI selection module, a training set formation module, a model construction module, a training module, and an identification module.

[0008] The image acquisition module is configured to obtain multiple oral cone beam CT images of multiple patients with jawbone cystic lesions; the annotation module is configured to annotate the teeth and jawbone regions in each oral cone beam CT image; the ROI selection module is configured to select ROIs that include the annotated teeth and jawbone regions to obtain multiple annotated ROIs images; the training set formation module is configured to annotate the cyst category labels generated by the jawbone cystic lesions in each annotated ROIs image, and jointly form a training set with all the annotated ROIs images with cyst category labels; the model construction module is configured to construct a deep learning model incorporating an attention mechanism; the attention mechanism introducing the relative positions of the cysts and teeth as the features of interest; the training module is configured to train the deep learning model using the training set to obtain a jawbone cystic lesion identification model; the identification module is configured to annotate the teeth and jawbone regions in the oral cone beam CT image to be identified of a patient with jawbone cystic lesions, and input the annotated ROIs image of the oral cone beam CT image to be identified after annotating the teeth and jawbone regions into the jawbone cystic lesion identification model to output the cyst category of the patient with jawbone cystic lesions.

[0009] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned method for identifying jawbone cystic lesions.

[0010] Fourthly, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned method for identifying jawbone cystic lesions is implemented.

[0011] Fifthly, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the above-mentioned method for identifying jawbone cystic lesions is implemented.

[0012] According to the specific embodiments provided by the present application, the present application has the following technical effects.

[0013] The present application provides a method, device, equipment, medium and product for identifying jawbone cystic lesions. By adding the relative position between the cyst and the teeth in the attention mechanism as a feature for defining and distinguishing cyst types, the identification method effectively focuses on the cyst and improves the identification accuracy of the diseased tissue; by identifying organs with obvious features such as teeth and jawbones and selecting the image interval of interest as the next training set, the detection range is narrowed and the efficiency of diagnosing jawbone cystic lesions is improved. Description of the Drawings

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0015] Figure 1 It is a schematic flowchart of a method for identifying jawbone cystic lesions provided by an embodiment of the present application.

[0016] Figure 2 It is a schematic diagram of common jawbone cysts provided by an embodiment of the present application.

[0017] Figure 3 It is a schematic flowchart of using a U-shaped network to identify the cyst type of a patient with jawbone cystic lesions provided by an embodiment of the present application.

[0018] Figure 4 It is a schematic simplified flowchart of a method for identifying jawbone cystic lesions provided by an embodiment of the present application.

[0019] Figure 5 It is a schematic diagram of the functional modules of an apparatus for identifying jawbone cystic lesions provided by an embodiment of the present application.

[0020] Figure 6 It is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed Embodiments

[0021] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0022] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0023] Using big data and deep learning algorithms can effectively identify and analyze the features in CBCT images, provide rapid diagnostic results, assist clinicians in formulating objective and accurate treatment plans, improve the efficiency of diagnosis, and also provide references for primary oral doctors and general oral doctors to select appropriate treatment plans.

[0024] Currently, most of the research on jaw cystic lesions is based on panoramic tomography or CT image analysis. There is no intelligent diagnosis of jaw cystic lesions based on deep learning, nor is there an intelligent diagnosis of jaw cystic lesions based on CBCT images.

[0025] In view of the above, in an exemplary embodiment, as Figure 1 shown, the present application provides a method for identifying jaw cystic lesions, including the following steps 101 to 107.

[0026] Step 101: Obtain multiple oral cone beam CT images of multiple patients with jaw cystic lesions.

[0027] Step 102: Mark the tooth and jaw regions in each oral cone beam CT image.

[0028] Step 103: Select the regions of interest containing the marked tooth and jaw regions to obtain multiple annotated images of the regions of interest.

[0029] Step 104: Mark the cyst category labels generated by the jaw cystic lesions in each annotated image of the region of interest, and together form a training set with all the annotated images of the regions of interest with cyst category labels.

[0030] Step 105: Construct a deep learning model incorporating an attention mechanism; the attention mechanism introduces the relative positions of the cyst and the teeth as the features of interest.

[0031] Step 106: Use the training set to train the deep learning model to obtain a jaw cystic lesion recognition model.

[0032] Step 107: Mark the teeth and jaw regions in the oral cone beam CT image to be recognized of the patient with jaw cystic lesions, and input the marked image of interest after marking the teeth and jaw regions in the oral cone beam CT image to be recognized into the jaw cystic lesion recognition model, and output the cyst category of the patient with jaw cystic lesions.

[0033] Implementing the above Step 101 to Step 107 is mainly aimed at multiple oral cone beam CT images (also known as jaw cystic lesion images) of multiple patients with jaw cystic lesions obtained by using CBCT technology, so as to accelerate the accuracy and speed of medical image recognition, and at the same time reduce the misdiagnosis rate of various doctors. The method of this application first identifies organs with obvious features such as teeth and jaws, then narrows the detection range according to the recognition results, selects appropriate image intervals of various cysts or tumors as the training data set for the next step, then constructs an appropriate deep learning model, trains it with the images around the teeth or jaws, and adds the feature relationship between the cyst and the teeth. Finally, a model that can be used to recognize jaw cystic lesions is obtained.

[0034] In another exemplary embodiment of this application, the dental CBCT training data set used consists of more than 300 patients, and the CBCT data of each patient are multiple dcm format images of 512×512.

[0035] In another exemplary embodiment of this application, jaw cystic lesions can be classified according to tissue origin and site of onset. Common jaw cysts are as Figure 2 shown. Those evolved from odontogenic tissues or the epithelium or epithelial remnants of teeth are called odontogenic jaw cysts, and cysts caused by residual epithelium in the embryonic period, blood extravasation cysts caused by injury, and aneurysmal bone cysts, etc. are called non-odontogenic jaw cysts. It can be seen from the definition of jaw cystic lesions that the relationship between cysts and teeth is close. For example: Periapical cysts are usually located around the apex of the infected tooth and are associated with pulp diseases. If a tooth has a history of long-term untreated pulpitis or periapical periodontitis, and a cyst is found near the root of the tooth, it is very likely to be a periapical cyst. And when a developing tooth is surrounded by a cyst, a dentigerous cyst is formed. This situation is common in impacted wisdom teeth or teeth that have not erupted normally. Therefore, if a cyst is found in a place where a tooth should grow but has not fully erupted, then this may be a sign of a dentigerous cyst. Keratocyst: Although keratocysts may also appear in any area containing tooth germs, they tend to occur more in the medial part of the mandible. So if the positional relationship and morphological relationship between the cyst and the teeth can be used as input parameters for deep learning, it will be beneficial to improve the accuracy of the final result.

[0036] Bone structures such as teeth and jaws are relatively easy to distinguish from other organs and tissues in CBCT images because their CT values are the highest, having high contrast in CBCT images. Teeth have fixed anatomical positions and morphological features. Under normal circumstances, the upper and lower teeth are adjacent row by row, forming a bell shape as a whole; the center point of each tooth is on the center line of the bell shape, and the variation angle is symmetric about the left and right incisors. Therefore, tooth recognition is usually relatively easy. Then, step 102 above can be replaced by the following steps 201 to 204.

[0037] Step 201: Mark the tooth regions in each oral cone beam CT image to obtain multiple oral cone beam CT images with marked tooth regions.

[0038] Exemplarily, the tooth region is obtained and marked through semantic segmentation, and the marked file is saved in the nii file format.

[0039] Step 202: Perform image enhancement on multiple oral cone beam CT images with marked tooth regions to obtain multiple enhanced images.

[0040] Enlarge the scale of the dataset and increase the diversity of the training dataset through conventional image enhancement methods.

[0041] Step 203: Use an image recognition algorithm to segment and mark each tooth from the tooth region of each enhanced image to obtain multiple enhanced images with marked teeth.

[0042] Exemplarily, traditional image recognition or deep learning methods can be used. For example, the Marker-Controlled Watershed (MCW) algorithm is used for tooth segmentation, the global threshold method is used for enamel segmentation, and then the MCW is used again for pulp segmentation.

[0043] Deep learning such as TensorFlow or PyTorch can also be used to build an instance segmentation model. Commonly used instance segmentation models include Mask R-CNN (Mask Region-based Convolutional Neural Networks), ResNet (Residual Network), U-Net (Convolutional Networks for Biomedical Image Segmentation), etc. During the training process, the images obtained in the previous step are used as inputs, and the manually marked tooth labels are used as the target outputs. Through the backpropagation algorithm and training samples, the model will learn the features of teeth and how to perform instance segmentation.

[0044] It is also possible to directly adopt a pre-trained model for tooth recognition, and then further train and fine-tune it according to the labeled data to obtain better performance.

[0045] Therefore, the marker-controlled watershed algorithm, instance segmentation model or pre-trained tooth recognition model is used to segment each tooth from the tooth region of each enhanced image.

[0046] In this application, the teeth in the CBCT image can be labeled using ITK-SNAP software, and the label file is saved in the nii file format.

[0047] Step 204: Locate and label the jaw region in each enhanced image with labeled teeth.

[0048] In another exemplary embodiment of the present application, in machine vision and image processing, the area to be processed is outlined in the form of a square, circle, ellipse, irregular polygon, etc. from the processed image, which is called the region of interest (ROI). In machine vision software, various operators and functions are commonly used to obtain the region of interest and perform the next step of image processing. In the field of image processing, the region of interest is an image area selected from the image, and this area is the focus of image analysis. The region of interest is circled for further processing. Using the ROI to circle the target to be read can reduce the processing time and increase the accuracy. For tooth segmentation, the ROI generation network first locates the foreground areas of the upper and lower jaws to reduce the computational cost of performing segmentation on high-resolution 3D CBCT images.

[0049] In another exemplary embodiment of the present application, the training data set is classified again according to the results of tooth and jaw recognition. The training data set of the deep learning semantic segmentation model (referring to the deep learning model introducing the attention mechanism) is labeled; ITK-SNAP software or other labeling software is used to set multiple regions of interest. By using rectangles of any size to enclose each lesion, and the coordinates of the upper left corner (x1, y1) and the lower right corner (x2, y2) of each ROI are recorded. x1 is the abscissa of the upper left corner of the ROI, y1 is the ordinate of the upper left corner of the ROI, x2 is the abscissa of the lower right corner of the ROI, and y2 is the ordinate of the lower right corner of the ROI. The width range of these ROIs is, for example, from 64 pixels to 450 pixels (median is 112 pixels), and the height range is from 70 pixels to 360 pixels (median is 122 pixels). The lesion label of each image data item, including these coordinates, is created in text form.

[0050] In another exemplary embodiment of the present application, converting the positional relationship or morphological relationship between the cyst and the tooth into a form that can be understood by a deep learning model requires a series of preprocessing steps to extract and represent this information. For example, three-dimensional coordinates are assigned to each tooth and cyst through spatial coordinates to quantify their specific positions within the jawbone. Calculate the relative distance from the cyst to the center of the nearest tooth and the shortest distance between the cyst edge and the tooth surface. Define a direction vector pointing from the tooth center to the cyst center to describe the relative orientation between the two. In addition to the above quantitative parameter forms, some qualitative indicators can also be established. For example, some qualitative proximity indicators are defined according to the relationship between the cyst and the tooth, such as "adjacent", "close", or "far away". That is, the relative position between the cyst and the tooth includes: the spatial coordinates of the tooth, the spatial coordinates of the cyst, the relative distance from the cyst to the center of the nearest tooth, the shortest distance between the cyst edge and the tooth surface, the direction vector pointing from the tooth center to the cyst center, and the proximity indicator between the cyst and the tooth.

[0051] In another exemplary embodiment of the present application, the deep learning model introducing the attention mechanism refers to a convolutional network model introducing the attention mechanism. The attention mechanism can help the model better focus on the key information in the input data. In image recognition, the attention mechanism can help the model better focus on the key features in the image, thereby improving the recognition accuracy.

[0052] Exemplarily, the convolutional network model introducing the attention mechanism includes: an input layer, a preprocessing layer, a convolutional layer, a residual block, an Inception module, an attention module (Convolutional Block Attention Module, CBAM), a pooling layer, a fully connected layer, a Dropout layer, and an output layer connected in sequence. The preprocessing layer is used to normalize and enhance the region of interest annotation image output by the input layer. The attention module includes a channel attention mechanism and a spatial attention mechanism; the relative position between the cyst and the tooth is introduced as the feature of interest in the spatial attention mechanism. The architecture of the convolutional network model introducing the attention mechanism will be introduced in more detail below.

[0053] 1. Input layer.

[0054] Input: The CT image is usually a grayscale image, the size depends on the actual situation, for example, 512×512 pixels, and the depth is 1 (single channel). If 3D CBCT images are used, the input dimension may be 128×128×128, and the depth is 1.

[0055] 2. Preprocessing layer.

[0056] Normalization: Scale the pixel values to between 0 and 1, which helps to accelerate the training process.

[0057] Enhancement: Operations such as random rotation, flipping, and scaling are performed to increase the diversity of the dataset and improve the generalization ability of the model.

[0058] The few-shot artificial intelligence method (e.g., N-way, K-shot) can also be used to optimize the most appropriate amount of data.

[0059] 3. Convolutional layer.

[0060] Multiple small convolutional kernels of 3×3 or 5×5 are used for feature extraction, with a stride of 1 and padding to keep the output size unchanged. Followed by the ReLU activation function to increase non-linearity.

[0061] Repeat multiple times: Stack multiple groups of convolutional layers and pooling layers as needed to gradually extract higher-level features.

[0062] 4. Residual block / Inception module.

[0063] Residual connection: Solve the problem of vanishing gradients in deep networks by directly connecting the previous layers to the subsequent layers to form a shortcut.

[0064] Inception module: Parallelly use convolutional kernels of different sizes (e.g., 1×1, 3×3, 5×5, etc.), and finally merge the outputs to enhance the model's adaptability to scale changes.

[0065] Attention module. Channel attention: Insert CBAM after each convolutional block. First, calculate the channel-level attention map, and by learning the importance of different feature channels, highlight the features that are more important for the task.

[0066] Spatial attention: Immediately followed by spatial-level attention to further refine the feature map and focus on the most relevant target regions. Introduce relative position parameters to highlight features such as the positional relationship between cysts and teeth.

[0067] Pooling layer. Use a 2×2 max pooling layer with a stride of 2 to reduce the spatial size of the feature map and lower the computational complexity.

[0068] 7. Fully connected layer.

[0069] Flatten: Flatten the output of the last convolutional layer or residual block / Inception module into a one-dimensional vector.

[0070] Fully connected layer: Contains a large number of neurons for further feature fusion.

[0071] 8. Dropout layer.

[0072] Dropout layer: Randomly discard a certain proportion of neurons to prevent overfitting.

[0073] 9. Output layer.

[0074] Output layer: Set the number of neurons according to the task requirements. For binary classification problems, it can be a single neuron plus a Sigmoid activation function; for multi-classification problems, the softmax activation function is used.

[0075] In another exemplary embodiment of the present application, the training and optimization of the convolutional network model incorporating the attention mechanism are introduced as follows.

[0076] Loss function: For classification tasks, the cross-entropy loss function is commonly used; for regression tasks, the mean squared error may be used, etc.

[0077] Optimizer: Adam, SGD (Stochastic Gradient Descent), etc., are used to adjust the model parameters and minimize the loss function.

[0078] Regularization: L1 / L2 regularization, which helps the model avoid overfitting.

[0079] Dataset: A large amount of labeled data is required to train the model to ensure that the model can learn sufficient patterns and rules.

[0080] The structure of the convolutional network model incorporating the attention mechanism is a typical design and may be adjusted according to specific data characteristics and business requirements in practical applications. At the same time, when necessary, auxiliary means can also be added to help accelerate the model learning process and improve efficiency, such as using the visualization method Gradient-weighted Class Activation Mapping (Grad-CAM) or comparing the images obtained from the self-attention layer with the original images (mean attention map). These two methods can verify each other, thereby accelerating the acquisition of information such as the important local positions of cysts in the images.

[0081] In one example, when the convolutional network model is a U-Net, the implementation process of step 107 above is as Figure 3 shown.

[0082] Figure 4The simple process of the method of the present application is shown: preprocess the dental and jaw CBCT image files (corresponding to step 102), establish a model to identify the high-density parts of the teeth and jaws (corresponding to step 102), preprocess the low-density CBCT images of the cysts around the teeth and jaws (corresponding to step 104), introduce the positional relationship between the teeth and jaws and the cysts into the model (corresponding to step 105), and achieve the identification of different types of cysts and the segmentation of cysts on the image (corresponding to step 107).

[0083] Compared with the prior art, the present application introduces important features such as the positional relationship between the cysts or masses to be identified and the teeth or other tissues. These important features are the features that define and distinguish the types of cysts and masses. Through the attention mechanism, the recognition algorithm can effectively focus on the cysts or masses, improve the recognition accuracy of the diseased tissues, and enhance the image recognition and segmentation capabilities based on the deep learning model.

[0084] The general steps of the method of the present application include: 1. Collection and preprocessing of dental CBCT images; 2. Construction of the model training set; 3. Construction of a convolutional network model with an attention layer; 4. Model training and evaluation. The results show that the method proposed in the present application provides an effective clinical application framework for digital dentistry.

[0085] The present application takes advantage of AI technology, gives full play to the advantage of applying CBCT as an auxiliary examination for jaw cystic lesions, applies the data set and multi-modal learning technology to the diagnosis and treatment plan selection of children's jaw cystic lesions, establishes an intelligent diagnosis of jaw cystic lesions based on CBCT images, improves the accuracy and efficiency of diagnosis, and provides strong support for clinical treatment.

[0086] Based on the same inventive concept, the embodiment of the present application also provides an identification device for jaw cystic lesions for implementing the above-mentioned identification method of jaw cystic lesions. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the identification device for jaw cystic lesions provided below can refer to the limitations on the identification method of jaw cystic lesions in the above text, and will not be repeated here.

[0087] In an exemplary embodiment, as Figure 5 shown, an identification device for jaw cystic lesions is provided, including: an image acquisition module, a labeling module, a region of interest selection module, a training set construction module, a model construction module, a training module, and an identification module.

[0088] An image acquisition module for acquiring multiple oral cone beam CT images of multiple patients with jaw cystic lesions; a labeling module for labeling the teeth and jaw regions in each oral cone beam CT image; a region of interest selection module for selecting regions of interest containing the labeled teeth and jaw regions to obtain multiple labeled images of regions of interest; a training set construction module for labeling the cyst class labels generated by jaw cystic lesions in each labeled image of the region of interest and constructing a training set together with all the labeled images of the region of interest with cyst class labels; a model construction module for constructing a deep learning model incorporating an attention mechanism; the attention mechanism introducing the relative positions of cysts and teeth as interesting features; a training module for training the deep learning model using the training set to obtain a jaw cystic lesion recognition model; and a recognition module for labeling the teeth and jaw regions in the oral cone beam CT image to be recognized of a patient with jaw cystic lesions, inputting the labeled image of the region of interest after labeling the teeth and jaw regions in the oral cone beam CT image to be recognized into the jaw cystic lesion recognition model, and outputting the cyst class of the patient with jaw cystic lesions.

[0089] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 6 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the cyst classes of patients with jaw cystic lesions. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for recognizing jaw cystic lesions.

[0090] Those skilled in the art can understand, Figure 6The structure shown is only a block diagram of some of the structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0091] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0092] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0093] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties.

[0094] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0095] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0096] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0097] In this article, specific examples are used to illustrate the principles and implementation modes of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation modes and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for identifying cystic lesions of the jawbone, characterized in that, Including: Obtaining multiple oral cone beam CT images of multiple patients with jaw cystic lesions; Labeling the teeth and jaw regions in each oral cone beam CT image; Selecting regions of interest containing the labeled teeth and jaw regions to obtain multiple labeled images of regions of interest; Labeling the cyst category labels generated by the jaw cystic lesions in each labeled image of the region of interest, and jointly constructing a training set with all the labeled images of the regions of interest with cyst category labels; Constructing a deep learning model incorporating an attention mechanism; the attention mechanism introduces the relative position of the cyst and the teeth as the region of interest feature; Training the deep learning model using the training set to obtain a jaw cystic lesion recognition model; Labeling the teeth and jaw regions in the oral cone beam CT image to be recognized of a patient with jaw cystic lesions, and inputting the labeled image of the region of interest after labeling the teeth and jaw regions in the oral cone beam CT image to be recognized into the jaw cystic lesion recognition model to output the cyst category of the patient with jaw cystic lesions.

2. The method for identifying jaw cystic lesions according to claim 1, wherein Labeling the teeth and jaw regions in each oral cone beam CT image specifically includes: Labeling the teeth regions in each oral cone beam CT image to obtain multiple oral cone beam CT images with labeled teeth regions; Performing image enhancement on multiple oral cone beam CT images with labeled teeth regions to obtain multiple enhanced images; Using an image recognition algorithm to segment and label each tooth from the teeth regions of each enhanced image to obtain multiple enhanced images with labeled teeth; Locating and labeling the jaw region in each enhanced image with labeled teeth.

3. The method for identifying jaw cystic lesions according to claim 2, wherein Segmenting each tooth from the teeth region of each enhanced image specifically includes: Using a marker-controlled watershed algorithm, an instance segmentation model, or a pre-trained tooth recognition model to segment each tooth from the teeth region of each enhanced image.

4. The method for identifying cystic lesions of the jawbone according to claim 1, characterized in that, The relative position of the cyst and the teeth includes: the spatial coordinates of the teeth, the spatial coordinates of the cyst, the relative distance from the cyst to the center of the nearest tooth, the shortest distance between the cyst edge and the tooth surface, the direction vector pointing from the tooth center to the cyst center, and the proximity index between the cyst and the teeth.

5. The method for identifying jawbone cystic lesions according to claim 1, characterized in that, The deep learning model incorporating an attention mechanism is a convolutional network model incorporating an attention mechanism.

6. The method for identifying cystic lesions of the jawbone according to claim 5, wherein The convolutional network model incorporating an attention mechanism includes: an input layer, a preprocessing layer, a convolutional layer, a residual block, an Inception module, an attention module, a pooling layer, a fully connected layer, a Dropout layer, and an output layer connected in sequence; The preprocessing layer is used to normalize and enhance the labeled image of the region of interest output by the input layer; The attention module includes a channel attention mechanism and a spatial attention mechanism; the relative position of the cyst and the teeth is introduced as the region of interest feature in the spatial attention mechanism.

7. An identification device for jaw cystic lesions, characterized in that, Including: An image acquisition module for obtaining multiple oral cone beam CT images of multiple patients with jaw cystic lesions; A labeling module for labeling the teeth and jaw regions in each oral cone beam CT image; A region of interest selection module for selecting regions of interest containing the labeled teeth and jaw regions to obtain multiple labeled images of regions of interest; A training set construction module, configured to label the cyst category labels generated by the jaw bone cystic lesions in each labeled image of interest, and together form a training set with all the labeled images of interest with the cyst category labels; A model construction module, configured to construct a deep learning model introducing an attention mechanism; the attention mechanism introduces the relative positions of the cyst and the teeth as the features of interest; A training module, configured to train the deep learning model using the training set to obtain a jaw bone cystic lesion recognition model; A recognition module, configured to label the tooth and jaw bone regions in the oral cone beam CT image to be recognized of a patient with a jaw bone cystic lesion, and input the labeled image of interest with the tooth and jaw bone regions in the oral cone beam CT image to be recognized into the jaw bone cystic lesion recognition model, and output the cyst category of the patient with the jaw bone cystic lesion.

8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the method for recognizing a jaw bone cystic lesion according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for recognizing a jaw bone cystic lesion according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for recognizing a jaw bone cystic lesion according to any one of claims 1-6.

Images