A child tooth defect detection method, device, equipment, medium and product

Abstract

This application discloses a method, apparatus, device, medium, and product for detecting dental defects in children, relating to the field of image processing technology. The method includes preprocessing an acquired panoramic radiograph of a child to be detected to obtain a standard image of a child's teeth; extracting local and global feature vectors from the standard image of the child's teeth, and then weighted and fused the extracted local and global feature vectors to obtain a fused feature image of the teeth; extracting geometric anchor points and calibrating the coordinates of each tooth region based on the fused feature image of the teeth to obtain the positional information of each tooth; matching the defect type of a single tooth region according to the fused feature image of the teeth and the positional information of each tooth to obtain a defect identification result; and generating a detection result based on the defect identification result and the positional information of each tooth. This application improves the detection accuracy and positioning precision of permanent and deciduous teeth in children on panoramic radiographs, avoiding missed detection of small teeth.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus, device, medium and product for detecting dental defects in children. Background Technology

[0002] Childhood dental developmental disorders can negatively impact children's chewing, speech development, and oral health. Early and accurate detection and localization are crucial for effective intervention and treatment. Panoramic radiographs, as a core imaging tool for clinical screening of childhood dental developmental disorders, can clearly present the overall structure of the child's jawbone, teeth, and surrounding tissues, serving as an important basis for related diagnostic work. Traditional diagnostic methods rely on doctors' visual interpretation and clinical experience, which are prone to missed diagnoses and misdiagnoses due to the influence of doctors' subjective factors. Therefore, there is an urgent need to provide a method for detecting childhood dental defects that can assist doctors in improving diagnostic accuracy. Summary of the Invention

[0003] In view of the above-mentioned defects or deficiencies in the related technologies, the purpose of this application is to provide a method, device, equipment, medium and product for detecting dental defects in children, which can improve the detection accuracy and positioning precision of permanent teeth and deciduous teeth on panoramic radiographs, avoid missed detection of small teeth (especially permanent tooth germs), and provide data support for doctors' diagnosis of dental defects in children.

[0004] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for detecting dental defects in children, comprising: preprocessing an acquired pediatric surface slice to obtain a standard pediatric dental image; extracting local and global feature vectors from the standard pediatric dental image, and weighting and fusing the extracted local and global feature vectors to obtain a fused dental feature image; wherein the dental size parameter is the pixel size data presented by the pediatric surface slice; extracting geometric anchor points and calibrating coordinates for each dental region based on the fused dental feature image to obtain the positional information of each tooth; matching the defect type for a single dental region according to the fused dental feature image and the positional information of each tooth to obtain a defect identification result; and generating a detection result based on the defect identification result and the positional information of each tooth.

[0005] Optionally, the step of extracting local and global feature vectors from the standard children's dental image and weighting and fusing the extracted local and global feature vectors to obtain a fused dental feature image includes: using a CNN model to extract local feature vectors from the standard children's dental image to obtain image local feature vectors; introducing tooth anatomical location encoding information into the Token of the VisionTransformer model, and extracting global feature vectors to characterize tooth arrangement and adjacency relationships through a multi-layer self-attention mechanism to obtain image global feature vectors; concatenating the image local feature vectors and the dental global feature vectors to obtain a concatenated feature image; using the ratio of the tooth size parameter of a single tooth in the surface slice of the child to be tested to the average tooth size parameter of all teeth as a dynamic weight, and weighting the concatenated feature image through convolution to strengthen the feature expression of small tooth size parameters of permanent tooth germs, canines, and incisors to obtain the fused dental feature image.

[0006] Optionally, the step of extracting geometric anchor points and calibrating coordinates for each tooth region based on the tooth fusion feature image to obtain the position information of each tooth includes: extracting key anchor points for each tooth in the tooth fusion feature image to obtain tooth anchor points; extracting the tooth center, crown vertex, and minimum circumscribed rectangle vertex as key geometric anchor points in the tooth fusion feature image; parsing position encoding parameters from the Token of the Vision Transformer model that incorporates tooth anatomical position encoding information and converting them into pixel coordinates as initial values ​​for anchor point coordinates; geometrically calibrating the initial coordinate values ​​of the tooth anchor points according to the oral anatomy of children during the transition from deciduous to permanent teeth and the normal dentition arrangement structure, and removing abnormal coordinates caused by feature noise based on the pixel range of the circumscribed rectangle of a single tooth to obtain the position information of each tooth; the position information includes tooth center coordinates, circumscribed rectangle coordinates, and offset distance.

[0007] Optionally, the step of matching the defect type of a single tooth region based on the tooth fusion feature image and the position information of each tooth to obtain a defect identification result includes: segmenting the tooth fusion feature image into single tooth regions based on tooth position information, tooth size parameters, and the minimum bounding rectangle to obtain multiple single tooth images; matching the multiple single tooth images with a preset defect feature library to identify the defect categories in the multiple single tooth images to obtain the defect identification result; the defect categories include abnormal tooth number, abnormal tooth morphology, and abnormal tooth eruption; the abnormal tooth number includes congenital missing teeth and supernumerary teeth; the abnormal tooth morphology includes malformed central cusps, fused teeth, and conical teeth; the abnormal tooth eruption includes delayed eruption, impaction, and malpositioned eruption.

[0008] Optionally, generating detection results based on the defect identification results and the position information of each tooth includes: associating and labeling the defective teeth according to the defect identification results, tooth size parameters, and position offset to obtain target defective teeth bound with tooth identifiers, defect categories, defect levels, center coordinates, tooth size parameters, and eruption offset distances; the defect levels are divided into mild, moderate, and severe; and the target defective teeth are integrated according to the order of the upper and lower jaws and dentition to generate a visually labeled image and a standardized detection report.

[0009] Optionally, the step of preprocessing the obtained pediatric surface layer to obtain a standard pediatric dental image includes: normalizing the grayscale value of the pediatric surface layer from the range of [0,255] to the range of [-3,3] to obtain a first image; transforming the grayscale value of the first image to the range of [0,1] to obtain an enhanced second image; and scaling the second image to a preset pixel size to obtain the standard pediatric dental image.

[0010] Secondly, this application provides a device for detecting dental defects in children, comprising: The preprocessing module is used to preprocess the acquired pediatric surface slices to obtain standard pediatric dental images; The extraction and fusion module is used to extract the local feature vectors and global feature vectors of the standard children's teeth image, and to perform weighted fusion of the extracted local feature vectors and global feature vectors to obtain a fused tooth feature image; the tooth size parameter is the pixel size data presented by the surface layer slice of the child to be detected; The positioning module is used to extract geometric anchor points and calibrate coordinates for each tooth region based on the tooth fusion feature image to obtain the position information of each tooth. The identification module is used to match the defect type of a single tooth region based on the tooth fusion feature image and the position information of each tooth, and obtain the defect identification result. The generation module is used to generate detection results based on the defect identification results and the position information of each tooth.

[0011] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for detecting dental defects in children as described above.

[0012] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for detecting dental defects in children as described above.

[0013] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method for detecting dental defects in children as described above.

[0014] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, device, equipment, medium, and product for detecting dental defects in children. The method involves preprocessing a curved surface slice of the child to be tested to obtain a standard image of the child's teeth. Local and global feature vectors are extracted from the standard image, and then weighted and fused to obtain a fused tooth feature image, providing data support for subsequent precise positioning of each tooth. Geometric anchor points are extracted and coordinates are calibrated for each tooth region using the fused tooth feature image to obtain the positional information of each tooth, providing an independent and complete target region for subsequent defect identification of each tooth, avoiding interference from adjacent teeth. The method further utilizes the fused tooth feature image to obtain the positional information of each tooth. By matching the feature image with the positional information of each tooth, defect types are matched for individual tooth regions to obtain defect identification results. This allows for comprehensive and accurate localization and defect identification of each tooth, effectively improving the problems of missed detection and positioning deviation in small permanent tooth germs, canines, and incisors in existing technologies. It can significantly improve the detection accuracy and reliability of defect identification for children's teeth during the transition from deciduous to permanent dentition. By generating detection results from the defect identification results and the positional information of each tooth, a complete detection result containing both defect and positional information can be obtained, effectively reducing subjective interpretation errors by physicians, improving the efficiency and accuracy of early screening for dental developmental abnormalities in children, and better meeting the needs of precise clinical diagnosis in pediatric dentistry. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is an application environment diagram of a method for detecting dental defects in children according to an embodiment of this application; Figure 2 This is a flowchart illustrating a method for detecting dental defects in children, provided in one embodiment of this application. Figure 3 This is a schematic diagram of the structure of the CNN model provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of the Vision Transformer model provided in the embodiments of this application; Figure 5 This is a schematic diagram of children's teeth recognition provided in an embodiment of this application; Figure 6 This is a schematic diagram of the functional modules of a child dental defect detection device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] The method for detecting dental defects in children provided in this application can be applied to, for example... Figure 1 In the application environment shown, the image acquisition device 102 communicates with the server 104 via a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated into the server 104, or placed in the cloud or on another server. The image acquisition device 102 can send the surface-mount image of the child to be detected to the server 104. After receiving the surface-mount image, the server 104 preprocesses the acquired image to obtain a standard child's tooth image; extracts the local and global feature vectors of the standard child's tooth image, and weights and fuses the extracted local and global feature vectors to obtain a tooth fusion feature image; the tooth size parameter is the pixel size data presented by the surface-mount image of the child to be detected; based on the tooth fusion feature image, geometric anchor points are extracted and coordinates are calibrated for each tooth region to obtain the position information of each tooth; based on the tooth fusion feature image and the position information of each tooth, defect type matching is performed on a single tooth region to obtain defect identification results; based on the defect identification results and the position information of each tooth, a detection result is generated.

[0020] The image acquisition device 102 can be a 3D image acquisition device such as a dental panoramic X-ray machine, or it can be a combination of a 3D image acquisition device and various desktop computers, laptops, smartphones, tablets, etc., for temporarily storing the panoramic X-ray films of children to be tested. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0021] In one exemplary embodiment, such as Figure 2 As shown, a method for detecting dental defects in children is provided. This method is executed by a computer device. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S201 to S205. Wherein: Step S201: Preprocess the obtained pediatric surface slices to obtain standard pediatric teeth images.

[0022] In the example embodiment, the pediatric panoramic radiographs to be examined are acquired using a panoramic dental X-ray machine (oral panoramic X-ray machine). This device generates two-dimensional panoramic images of the entire mouth's teeth and maxillofacial region through rotational scanning, and is suitable for imaging examinations of children's dental development, permanent tooth germs, and oral lesions. Because the acquired pediatric panoramic radiographs have uneven grayscale distribution, imaging noise, and inconsistent sizes, and because the contrast of children's dental images is low and the features of small permanent tooth germs are not obvious, the accuracy of subsequent defect identification will be reduced. Therefore, preprocessing of the pediatric panoramic radiographs to be examined is necessary.

[0023] In this embodiment, the pediatric stromal slices to be tested undergo grayscale normalization, image enhancement, and size standardization processing sequentially to address the issues of uneven grayscale distribution, imaging noise, low contrast, and inconsistent sizes present in the pediatric stromal slices. Specifically: The grayscale values ​​of the child's surface layer to be tested are normalized from the range of [0,255] to the range of [-3,3] to obtain the first image; the grayscale values ​​of the first image are transformed to the range of [0,1] to obtain the second image after image enhancement; the second image is scaled to a preset pixel size to obtain a standard image of a child's teeth.

[0024] Understandably, formula (1) is used to normalize the gray values ​​of the pediatric surface slices to be tested from the interval [0, 255] to the interval [-3, 3], that is: (1) in, This represents the value after data scaling. mean ( x () represents the mean, or can be expressed as... express, The standard deviation of a variable.

[0025] The grayscale value range of the first image becomes [-3,3], which is the theoretical range. The actual range is generally smaller than [-3,3]. That is to say, the minimum value in the [0,1] interval is not exactly 0 and the maximum value is not exactly 1. At this time, the image of the first image is too dark or too bright after visualization, and the first image needs to be enhanced. This application uses the following formula (2) to enhance the first image, that is, to transform the grayscale value range of the first image back to [0,1].

[0026] (2) in, min ( x () represents the minimum value of the variable. max ( x ) indicates the maximum value of the variable.

[0027] It should be noted that the gray values ​​of the pediatric surface slices to be tested in this embodiment are naturally distributed in the [0, 255] interval. The Z-Score normalization algorithm is used to map the gray values ​​in the [0, 255] interval to the [-3, 3] interval, which can effectively suppress the interference of extreme gray values ​​(overly bright or overly dark areas) in the pediatric surface slices, initially unifying the image gray value distribution and laying the foundation for subsequent image enhancement. The Min-Max normalization algorithm is used to linearly map the gray values ​​of the first image from the [-3, 3] interval to the [0, 1] interval, which can significantly optimize image contrast, enhance the feature differentiation between the tooth region and background tissues such as the jawbone and gingiva, especially highlighting the gray value differences of subtle features such as small permanent tooth germs and canines, solving the problem of dark images and blurred details in pediatric surface slices caused by low imaging dose and incomplete jawbone development. The second image is uniformly scaled to a size of 224×224 pixels, which is consistent with the subsequent CNN model and Vision... The input size of the Transformer model is perfectly matched, eliminating the model input deviation caused by surface slices (with inconsistent resolution and size) acquired by different devices, and ensuring the consistency and stability of model feature extraction.

[0028] Step S202: Extract the local feature vector and global feature vector of the standard children's teeth image, and then perform weighted fusion of the extracted local feature vector and global feature vector to obtain the fused tooth feature image.

[0029] In the example implementation, a CNN model is used to extract the local feature vector f1 of the standard children's teeth image, focusing on detailed features such as tooth texture and contour; a Vision Transformer model is used to extract the global feature vector f2 of the standard children's teeth image, focusing on global features such as tooth arrangement and adjacency relationship; the local feature vector f1 and the global feature vector f2 are concatenated and weighted by 1×1 convolution with tooth size as the weight to obtain the tooth fusion feature image F, the specific formula is as follows (3).

[0030] F=cat(f1,f2) (3) Where f1 is the image feature extracted by the CNN model from a standard child's teeth image, and f2 is the image feature extracted by the VisionTransformer model from a standard child's teeth image.

[0031] Optionally, step 202 above may include: using a CNN model to extract local feature vectors from standard children's dental images to obtain image local feature vectors; introducing tooth anatomical location encoding information into the Token of the Vision Transformer model, and extracting global feature vectors to characterize tooth arrangement and adjacency relationships through multi-layer self-attention mechanism to obtain image global feature vectors; concatenating the image local feature vectors and the tooth global feature vectors to obtain a concatenated feature image; using the ratio of the tooth size parameter of a single tooth in the surface slice of the child to be tested to the average tooth size parameter of all teeth as dynamic weights, and performing weighted adjustment on the concatenated feature image through convolution to enhance the feature expression of small tooth size parameters of permanent tooth germs, canines, and incisors to obtain a tooth fusion feature image.

[0032] This application utilizes a dual-model collaborative feature extraction approach, combined with dynamic weighted fusion based on tooth size parameters. This overcomes the limitations of single-model feature extraction, highlighting features of small-sized teeth and providing high-quality feature support for subsequent tooth localization and defect identification. The fusion strategy employs a combination of splicing and dynamic weighting, using tooth size parameters as the core weight to enhance features of small-sized teeth while weakening redundant features from the background and normal-sized teeth, further providing high-quality feature support for subsequent tooth localization and defect identification.

[0033] In one specific implementation, such as Figure 3 As shown, the CNN model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a local connection layer, and two fully connected layers. The input layer image has a width and height of 224 and 3 channels. The first convolutional layer (kernel size 3×3×64) has a stride of 1 (i.e., s=1) and padding of 1 (i.e., p=1). The size of the output image is calculated using equation (4), and the convolution operation is calculated using equation (5), i.e.: (4) in, Out size Indicates the output image size. In size Indicates the size of the input image, in pixels; F size This indicates the kernel size, i.e., the width / height of the kernel in the convolutional layer; P This indicates the number of pixel rings used to fill the edges of the input image. S The step size represents the pixel distance that the convolution kernel slides across the input image each time, and is specifically calculated as shown in the following formula (5).

[0034] (5) in, S 1( i,j () represents the convolution calculation result of the output image; X Indicates the input image; K This represents the convolution kernel (filter).

[0035] The kernel size of the first average pooling layer is 2×2, i.e., s=2, the stride is 2, and p=2. The size of the output image is calculated according to the following formula (6): (6) in, S 2( i,j () represents the pixel in the i-th row and j-th column of the output image. H Where is the height of the input image, W is the width of the input image, and X is the input image height.

[0036] The kernel size of the second convolutional layer is 3×3×128, the stride is 1 (i.e., s=1), and the padding is 1 (i.e., p=1). The size of the output image is calculated according to the above formula (5): 112×112×128. The kernel size of the second average pooling layer is 4×4, i.e., s=4, and the step size is 4, i.e., p=4. The size of the output image is calculated according to the above formula (6): 28×28×128.

[0037] The kernel size of the third convolutional layer is 3×3×256, the stride is 1 (i.e., s=1), and the padding is 1 (i.e., p=1). The size of the output image is calculated according to the above formula (5): 56×56×256.

[0038] The kernel size of the third average pooling layer is 4×4, i.e., s=4, and the step size is 4, i.e., p=4. The size of the output image is calculated according to the above formula (6): 7×7×256.

[0039] To minimize the probability of overfitting, a locally connected layer is introduced, which is equivalent to not sharing weights. The parameters (i.e., weights) of the convolution kernel change every time the kernel moves to a different position.

[0040] Finally, two fully connected layers are applied, with dimensions of 12544×4096 and 4096×1000, respectively.

[0041] Furthermore, all activation operations involved in the CNN model are calculated according to the following equation (7), that is: Formula (7) Where a is the gray value of the output image and z is the gray value of the input image.

[0042] It should be noted that when using a CNN model to extract local features from standard children's teeth images, a 224×224×3 standard children's teeth image is input into the CNN model. The standard children's teeth image sequentially passes through a 3×3×64 first convolutional layer (stride 1, padding 1, sigmoid activation function), a 2×2 first average pooling layer, a 3×3×128 second convolutional layer (stride 1, padding 1), a 4×4 second average pooling layer, a 3×3×256 third convolutional layer (stride 1, padding 1), a 4×4 third average pooling layer, and a local connection layer. The system consists of a convolutional layer and two fully connected layers (dimensions 12544→4096→1000). The convolutional layer captures the local texture, contour, and edge details of teeth in standard children's dental images (such as the raised texture of malformed central cusps and the edge connection features of fused teeth). The pooling layer retains key features and reduces redundant information. The local connected layer avoids overfitting and preserves the subtle differences of individual teeth. Finally, the fully connected layer outputs a local feature vector with a fixed dimension, i.e., the image local feature vector, which can accurately represent the local detail features of individual teeth.

[0043] In one specific implementation, such as Figure 4As shown, the Vision Transformer model includes an image segmentation and embedding module, a Class Token and position encoding module, a 12-layer Transformer Block module, and a global feature output module. Specifically, the image segmentation and embedding module divides the image into 16×16 blocks, resulting in 196 image blocks, which are linearly mapped to a 196×768-dimensional token sequence. The Class Token and position encoding module adds one trainable class token, increasing the dimension to 197×768, and adds a trainable position encoding containing tooth anatomical location information. Each of the 12-layer Transformer Block modules consists of layer normalization, multi-head self-attention, residual connections, layer normalization, MLP, and residual connections. The MLP first increases the dimension from 768 to 3072, then restores it to 768 after GELU activation and Dropout. The global feature output module takes the class token output and obtains a 768-dimensional global feature vector, used to represent global contextual features such as tooth arrangement, adjacency relationships, and eruption location.

[0044] It should be noted that when using the Vision Transformer model to extract the global feature vector of standard children's dental images, the standard children's dental images are first divided into 16×16 blocks, resulting in 196 image blocks. Each image block is then converted into a 768-dimensional token sequence through linear mapping. Subsequently, a trainable tooth anatomical location code and a ClassToken are added, enabling the Vision Transformer model to capture global information such as the arrangement order and adjacency relationships of teeth in the dentition. After processing through a self-attention mechanism of 12 Transformer Blocks, the global contextual dependencies between teeth and the jawbone and adjacent teeth are captured. Finally, the ClassToken is extracted as the global feature vector, i.e., the global image feature vector. The global image feature vector can comprehensively represent the global arrangement and spatial location association features of teeth.

[0045] In addition, in practical applications, the CNN model and the VisionTransformer model can be trained using the pediatric surface slices annotated by physicians. During the training process, the stochastic gradient descent algorithm is used to optimize the CNN model and the VisionTransformer model. The base learning rate is set to 0.05, the weight decay factor is 1e-5, the momentum value is 0.8, and the epoch is set to 400. After training is completed, the model is saved. During this process, the loss function is used for backpropagation. The specific loss function is as follows (8): (8) in, Indicates the first The true label value of each sample Indicates the first The predicted value for each sample, defaulting to logarithmic. With natural numbers The base is equal to . Indicates the first The size of each tooth on a tomographic image, and The weights represent the average size of the teeth on the tomographic image. weight .

[0046] In one specific implementation, when weighting and fusing the local and global feature vectors of an image, a fusion strategy of concatenation and dynamic weighting is adopted. The tooth size parameter is used as the core weight to strengthen the features of small teeth and weaken redundant features of the background and normal-sized teeth. Specific steps may include: Step 1: Concatenate the local feature vectors extracted by the CNN model with the global feature vectors extracted by the Vision Transformer model to obtain a concatenated feature image. During the concatenation process, maintain consistent feature dimensions to ensure the complete preservation of local detailed features and globally related features, and avoid feature loss. Step 2: Based on the size parameters of each tooth, calculate the ratio of the pixel size of a single tooth to the average pixel size of all teeth, and use it as the dynamic weight coefficient for that tooth; among them, the ratio of small teeth (such as permanent tooth germs, canines, incisors) is greater than 1, and the ratio of normal-sized teeth is close to 1. Through this weight design, higher weight is given to the features of small teeth. Step 3: Use a 1×1 convolution kernel to perform weighted adjustment on the stitched feature image. According to the above dynamic weight coefficients, the stitched features are weighted channel by channel to enhance the local detail features and global correlation features of small teeth, and suppress background noise and redundant features of normal-sized teeth. After the weighted adjustment is completed, the final tooth fusion feature image is output. The tooth fusion feature image contains both local details and global correlation information of teeth, and the features of small teeth are significantly highlighted.

[0047] It should be noted that the tooth size parameters in this application are obtained by simultaneously extracting the pixel size (i.e., the pixel area of ​​the tooth in the image) of each tooth in the surface slice of the child to be tested (the original image before preprocessing to avoid size scaling deviation), calculating the average pixel size of all teeth, and determining the size parameters of each tooth, namely the pixel size of a single tooth and the average pixel size of all teeth, as the core basis for subsequent weighted fusion.

[0048] In this embodiment, dynamic weights are designed using tooth size parameters to assign higher weights to small teeth (permanent tooth germs, canines, and other easily missed targets). 1×1 convolution weighting is used to enhance their feature representation, solving the problem of missed detections caused by weak features of small teeth being easily masked by background or normal-sized tooth features in existing technologies. This significantly improves the feature recognition accuracy of small teeth. The fusion process is based on the actual tooth size rather than fixed weights, adapting to the large individual differences in children's tooth sizes. This allows the fused feature image to accurately match the actual features of each tooth, avoiding feature distortion caused by fixed weights, and further improving the accuracy of subsequent tooth localization and defect identification. Weighted fusion not only enhances the features of small teeth but also effectively suppresses background noise and interference from irrelevant tissues such as the jawbone, reducing localization deviations and misjudgments of defects caused by irrelevant interference, indirectly improving overall detection performance. The feature extraction process is precisely matched with the structural parameters of the CNN model and the Vision Transformer model, and the weighted fusion process is seamlessly integrated, ensuring that the features of the two models can be effectively fused. This fully leverages the synergistic advantages of the two models and provides high-quality fused feature support for subsequent tooth localization and defect identification, effectively promoting the efficient and accurate operation of the overall detection process.

[0049] Step S203: Based on the tooth fusion feature image, extract geometric anchor points and calibrate coordinates for each tooth region to obtain the position information of each tooth.

[0050] In example embodiments, such as Figure 5 As shown, the geometric center, contour corner points, and boundary anchor points of each tooth's region are extracted. Based on the standard tooth arrangement rules and image proportions, the coordinates of these anchor points are offset and spatially calibrated to remove positional deviations generated during feature extraction and localization. Finally, the positional information of each tooth on the curved surface layer is obtained. The tooth positional information in this application includes the tooth center coordinates, the coordinates of the circumscribed rectangle, and the offset distance.

[0051] Optionally, step S203 may include: extracting key anchor points of each tooth in the tooth fusion feature image to obtain tooth anchor points; extracting the tooth center, crown vertex, and minimum bounding rectangle vertex as key geometric anchor points in the tooth fusion feature image; parsing the position encoding parameters from the Token of the Vision Transformer model that incorporates tooth anatomical position encoding information and converting them into pixel coordinates as initial values ​​of anchor point coordinates; geometrically calibrating the initial values ​​of tooth anchor point coordinates according to the oral anatomy of children during the transition from deciduous to permanent teeth and the normal dentition arrangement structure, and removing abnormal coordinates caused by feature noise based on the pixel range of the bounding rectangle of a single tooth to obtain the position information of each tooth.

[0052] By extracting geometric anchor points and calibrating coordinates on the fused feature images, the enhanced local details and global arrangement information of teeth in the fused features can be fully utilized, significantly improving the stability and accuracy of tooth localization. Coordinate calibration effectively eliminates localization errors caused by image noise and feature offset, ensuring that the acquired tooth position information truly reflects the actual spatial position of the teeth in the jawbone. This provides a reliable positional basis for subsequent single tooth region segmentation, defect feature enhancement, and defect type matching, reducing the probability of missed detection and mislocalization of small permanent tooth germs, canines, and incisors.

[0053] Step S204: Based on the tooth fusion feature image and the position information of each tooth, perform defect type matching on a single tooth region to obtain defect identification results.

[0054] In the example embodiment, defect categories include abnormal tooth number, abnormal tooth morphology, and abnormal tooth eruption; abnormal tooth number includes congenital absence of teeth and supernumerary teeth; abnormal tooth morphology includes malformed central cusps, fused teeth, and conical teeth; abnormal tooth eruption includes delayed eruption, impaction, and malpositioned eruption. The defect identification result includes the corresponding tooth.

[0055] Optionally, step S204 may include: segmenting the tooth fusion feature image into single tooth regions based on tooth location information, tooth size parameters, and minimum bounding rectangle to obtain multiple single tooth images; matching the multiple single tooth images with a preset defect feature library to identify the defect categories in the multiple single tooth images and obtain defect identification results.

[0056] Based on the fusion feature images and location information of teeth, it can achieve precise segmentation of single tooth regions, effectively avoiding misjudgments caused by overlapping adjacent teeth and background interference. By enhancing and matching the defect features of single teeth, it can significantly improve the recognition accuracy of common dental defects in children, such as malformed central cusps, fused teeth, abnormal number of teeth, and eruption abnormalities. At the same time, relying on precise location information to associate defects with corresponding tooth positions, it ensures that the defect recognition results can be directly used for clinical diagnosis, further improving the practicality and reliability of the overall detection method.

[0057] Step S205: Based on the defect identification results and the position information of each tooth, generate the detection results.

[0058] In the example embodiment, the detection results include tooth position, tooth condition, defect type, defect level, and defect location; the defect level is divided into mild, moderate, and severe.

[0059] Optionally, step S205 may include: associating and annotating the defective teeth with defects based on the defect identification results, tooth size parameters, and positional offset to obtain target defective teeth bound with tooth identifiers, defect categories, defect levels, center coordinates, tooth size parameters, and eruption offset distances; integrating the target defective teeth in the order of the upper and lower jaws and dentition to generate a visually annotated image and a standardized inspection report.

[0060] By combining defect identification results with tooth location information to generate detection results, a one-to-one correspondence between defects and tooth positions can be achieved, making the detection results more intuitive, standardized, and in line with clinical reading habits. At the same time, it forms a complete, structured, and traceable detection information, which is convenient for doctors to quickly view, diagnose, and record, improving the ease of use and practical value of this method in the clinical screening of dental developmental abnormalities in children, and also providing reliable data support for the subsequent development of personalized treatment plans.

[0061] Steps S201 to S205 are performed as follows: Standard children's dental images are obtained by preprocessing the pediatric surface slices to be tested; local and global feature vectors are extracted from the standard children's dental images, and then weighted and fused to obtain a fused dental feature image, providing data support for subsequent precise localization of each tooth; geometric anchor points are extracted and coordinates are calibrated for each tooth region using the fused dental feature image to obtain the positional information of each tooth, providing an independent and complete target region for subsequent defect identification of each tooth, avoiding interference from adjacent teeth; and the fused dental feature image and the positional information of each tooth region are used to identify individual tooth regions. By matching defect types within the domain, defect identification results are obtained, enabling comprehensive and accurate localization and defect identification of each tooth. This effectively improves upon existing technologies that are prone to missed detection and localization errors in small permanent tooth germs, canines, and incisors, significantly enhancing the detection accuracy and reliability of defect identification for children during the transition from deciduous to permanent dentition. By generating detection results from the defect identification results and the location information of each tooth, a complete detection result containing both defect and location information is obtained. This effectively reduces subjective interpretation errors by physicians, improves the efficiency and accuracy of early screening for dental developmental abnormalities in children, better meets the needs of precise clinical diagnosis in pediatric dentistry, and provides accurate data support for early screening and intervention of dental developmental disorders in children.

[0062] Based on the same inventive concept, this application also provides a child dental defect detection device for implementing the aforementioned child dental defect detection method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more child dental defect detection device embodiments provided below can be found in the limitations of the child dental defect detection method described above, and will not be repeated here.

[0063] In one exemplary embodiment, such as Figure 6 As shown, a device for detecting dental defects in children is provided. This device 600 includes: a preprocessing module 601, an extraction and fusion module 602, a positioning module 603, a recognition module 604, and a result output module 605. Specifically, The preprocessing module 601 is used to preprocess the acquired pediatric surface slices to obtain standard pediatric dental images; The extraction and fusion module 602 is used to extract local feature vectors and global feature vectors from standard children's teeth images, and to perform weighted fusion of the extracted local feature vectors and global feature vectors to obtain a fused tooth feature image; the tooth size parameter is the pixel size data presented by the surface layer slice of the child to be detected; The positioning module 603 is used to extract geometric anchor points and calibrate coordinates of each tooth region based on the tooth fusion feature image to obtain the position information of each tooth; The recognition module 604 is used to match the defect type of a single tooth region based on the tooth fusion feature image and the position information of each tooth, and obtain the defect recognition result. The result output module 605 is used to generate detection results based on the defect identification results and the position information of each tooth.

[0064] As an optional implementation, the extraction and fusion module 602 is specifically used to: extract local feature vectors from standard children's dental images using a CNN model to obtain local image feature vectors; introduce tooth anatomical location encoding information into the Token of the Vision Transformer model, and extract global feature vectors to represent tooth arrangement and adjacency relationships through multi-layer self-attention mechanism processing to obtain global image feature vectors; concatenate the image local feature vectors and the tooth global feature vectors to obtain a concatenated feature image; and use the ratio of the tooth size parameter of a single tooth in the surface slice of the child to be tested to the average tooth size parameter of all teeth as a dynamic weight, and perform weighted adjustment on the concatenated feature image through convolution to enhance the feature expression of small tooth size parameters of permanent tooth germs, canines, and incisors to obtain a tooth fusion feature image.

[0065] As an optional implementation, the aforementioned positioning module 603 is specifically used to: extract key anchor points of each tooth in the tooth fusion feature image to obtain tooth anchor points; extract the tooth center, crown vertex, and minimum circumscribed rectangle vertex as key geometric anchor points in the tooth fusion feature image; parse the position encoding parameters from the Token of the Vision Transformer model that incorporates tooth anatomical position encoding information and convert them into pixel coordinates as initial values ​​of anchor point coordinates; perform geometric calibration on the initial coordinate values ​​of the tooth anchor points according to the oral anatomy of children during the transition from deciduous to permanent teeth and the normal dentition arrangement structure, and remove abnormal coordinates caused by feature noise based on the pixel range of the circumscribed rectangle of a single tooth to obtain the position information of each tooth; the position information includes tooth center coordinates, circumscribed rectangle coordinates, and offset distance.

[0066] As an optional implementation, the aforementioned recognition module 604 is specifically used to: segment the tooth fusion feature image into single-tooth regions based on tooth position information, tooth size parameters, and the minimum bounding rectangle to obtain multiple single-tooth images; match the multiple single-tooth images with a preset defect feature library to identify the defect categories in the multiple single-tooth images and obtain defect recognition results; the defect categories include abnormal tooth number, abnormal tooth morphology, and abnormal tooth eruption; abnormal tooth number includes congenital missing teeth and supernumerary teeth; abnormal tooth morphology includes malformed central cusps, fused teeth, and conical teeth; abnormal tooth eruption includes delayed eruption, impaction, and misaligned eruption.

[0067] As an optional implementation, the above-mentioned generator 605 is specifically used to associate and annotate the defective teeth according to the defect identification results, tooth size parameters and positional offset, to obtain target defective teeth bound with tooth identifiers, defect categories, defect levels, center coordinates, tooth size parameters and eruption offset distances; the defect levels are divided into mild, moderate and severe; the target defective teeth are integrated in the order of the upper and lower jaws and dentition to generate a visual annotated image and a standardized inspection report.

[0068] As an optional implementation, the preprocessing module 601 is specifically used to normalize the gray values ​​of the child surface layer to be detected from the range of [0,266] to the range of [-3,3] to obtain a first image; transform the gray values ​​of the first image to the range of [0,1] to obtain a second image after image enhancement; and scale the second image to a preset pixel size to obtain a standard child's teeth image.

[0069] This implementation method involves preprocessing the pediatric curved surface slices to obtain standard pediatric dental images. Local and global feature vectors are extracted from these images and then weighted and fused to obtain a fused dental feature image, providing data support for subsequent precise positioning of each tooth. Geometric anchor points are extracted and coordinates are calibrated using the fused feature image to obtain the positional information of each tooth, providing an independent and complete target area for subsequent defect identification and avoiding interference from adjacent teeth. Finally, the fused feature image and the positional information of each tooth are used to perform analysis on individual tooth regions. By matching defect types and obtaining defect identification results, the system can comprehensively and accurately locate and identify defects in each tooth. This effectively improves the problems of missed detection and positioning deviation in small permanent tooth germs, canines, and incisors in existing technologies, and can significantly improve the detection accuracy and defect identification reliability of teeth in children during the transition from deciduous to permanent dentition. By generating detection results from the defect identification results and the positional information of each tooth, a complete detection result containing both defect and positional information can be obtained, effectively reducing subjective interpretation errors by physicians, improving the efficiency and accuracy of early screening for dental developmental abnormalities in children, better meeting the needs of precise clinical diagnosis in pediatric dentistry, and providing accurate data support for early screening and intervention of dental developmental disorders in children.

[0070] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data on pediatric dental defects. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for detecting pediatric dental defects.

[0071] Those skilled in the art will understand that Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0072] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0073] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0074] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0075] 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 used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0076] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory 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), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0077] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0078] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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, they should be considered to be within the scope of this specification.

[0079] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for detecting dental defects in children, characterized in that, The method for detecting dental defects in children includes: The obtained pediatric surface radiographs were preprocessed to obtain standard pediatric dental images; Local and global feature vectors are extracted from the standard children's teeth image, and the extracted local and global feature vectors are weighted and fused to obtain a fused tooth feature image; the tooth size parameter is the pixel size data presented by the surface layer slice of the child to be tested; Based on the tooth fusion feature image, geometric anchor points are extracted and coordinates are calibrated for each tooth region to obtain the position information of each tooth; Based on the tooth fusion feature image and the position information of each tooth, defect type matching is performed on a single tooth region to obtain defect identification results; Based on the defect identification results and the location information of each tooth, a detection result is generated.

2. The method for detecting dental defects in children according to claim 1, characterized in that, The step of extracting local and global feature vectors from the standard children's teeth image, and then weighting and fusing the extracted local and global feature vectors to obtain a fused tooth feature image, includes: The local feature vectors of the standard children's teeth images are extracted using a CNN model to obtain the local feature vectors of the images; In the Token of the Vision Transformer model, tooth anatomical location encoding information is introduced. After processing by a multi-layer self-attention mechanism, global feature vectors for representing tooth arrangement and adjacency relationship are extracted to obtain the global feature vector of the image. The local feature vector of the image and the global feature vector of the teeth are concatenated to obtain a concatenated feature image; Using the ratio of the tooth size parameter of a single tooth in the surface slice of the child to be tested to the average tooth size parameter of all teeth as a dynamic weight, the stitched feature image is weighted by convolution to enhance the feature expression of small tooth size parameters of permanent tooth germs, canines and incisors, thus obtaining the tooth fusion feature image.

3. The method for detecting dental defects in children according to claim 2, characterized in that, The step of extracting geometric anchor points and calibrating coordinates for each tooth region based on the tooth fusion feature image to obtain the positional information of each tooth includes: Extract the key anchor points of each tooth in the tooth fusion feature image to obtain the tooth anchor points; The tooth center, crown vertex, and minimum bounding rectangle vertex are extracted from the tooth fusion feature image as key geometric anchor points. The position encoding parameters are parsed from the Token of the Vision Transformer model that incorporates tooth anatomical position encoding information and converted into pixel coordinates as initial values ​​for anchor coordinates. Based on the oral anatomy of children during the transition from deciduous to permanent teeth and the normal dentition arrangement, the initial coordinates of the tooth anchor points are geometrically calibrated, and abnormal coordinates caused by feature noise are eliminated based on the pixel range of the circumscribed rectangle of a single tooth to obtain the position information of each tooth; the position information includes the tooth center coordinates, the circumscribed rectangle coordinates, and the offset distance.

4. The method for detecting dental defects in children according to claim 3, characterized in that, The step of matching defect types in a single tooth region based on the tooth fusion feature image and the position information of each tooth to obtain defect identification results includes: Based on tooth location information, tooth size parameters, and minimum bounding rectangle, the tooth fusion feature image is segmented into individual tooth regions to obtain multiple individual tooth images. Multiple single-tooth images are matched with a preset defect feature library to identify the defect categories in the multiple single-tooth images, and the defect identification results are obtained. The defect categories include abnormal tooth number, abnormal tooth morphology, and abnormal tooth eruption. Abnormal tooth number includes congenital missing teeth and supernumerary teeth. Abnormal tooth morphology includes malformed central cusps, fused teeth, and conical teeth. Abnormal tooth eruption includes delayed eruption, impaction, and misaligned eruption.

5. The method for detecting dental defects in children according to claim 4, characterized in that, The generation of detection results based on the defect identification results and the position information of each tooth includes: Based on the defect identification results, tooth size parameters, and positional offset, the defective teeth are associated and labeled to obtain target defective teeth bound with tooth identifiers, defect categories, defect levels, center coordinates, tooth size parameters, and eruption offset distances; the defect levels are divided into mild, moderate, and severe. The target defective teeth are integrated according to the order of the upper and lower jaws and the dental arch to generate a visually labeled image and a standardized test report.

6. The method for detecting dental defects in children according to claim 1, characterized in that, The process of preprocessing the acquired pediatric surface radiographs to obtain standard pediatric dental images includes: The gray values ​​of the child surface layer to be detected are normalized from the range of [0,255] to the range of [-3,3] to obtain the first image; The grayscale values ​​of the first image are transformed to the [0,1] range to obtain the enhanced second image; The second image is scaled to a preset pixel size to obtain the standard child's teeth image.

7. A device for detecting dental defects in children, characterized in that, The device for detecting dental defects in children includes: The preprocessing module is used to preprocess the acquired pediatric surface slices to obtain standard pediatric dental images; The extraction and fusion module is used to extract the local feature vectors and global feature vectors of the standard children's teeth image, and to perform weighted fusion of the extracted local feature vectors and global feature vectors to obtain a fused tooth feature image; the tooth size parameter is the pixel size data presented by the surface layer slice of the child to be detected; The positioning module is used to extract geometric anchor points and calibrate coordinates for each tooth region based on the tooth fusion feature image to obtain the position information of each tooth. The identification module is used to match the defect type of a single tooth region based on the tooth fusion feature image and the position information of each tooth, and obtain the defect identification result. The generation module is used to generate detection results based on the defect identification results and the position information of each tooth.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the method for detecting dental defects in children according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method for detecting dental defects in children as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method for detecting dental defects in children as described in any one of claims 1-6.

Images