Tooth health preliminary screening method and device based on image recognition, equipment and medium

Abstract

The invention relates to a tooth health primary screening method and device based on image recognition, equipment and a medium. The method comprises the following steps: based on acquired tooth multi-source image data, carrying out image preprocessing on the tooth multi-source image data to obtain preprocessed tooth image data; extracting multi-scale tooth features from the preprocessed tooth image data to obtain tooth feature map data; performing multi-feature fusion processing on the tooth feature map data to obtain fused tooth feature data; on the basis of the fused tooth feature data, combining a preset tooth anatomical structure template, performing feature region division and mapping to obtain tooth region feature data; performing health condition analysis on the tooth region feature data to obtain tooth health condition data; and generating a tooth health screening result according to the tooth health condition data. By adopting the method, multi-modal information can be fully mined, and the comprehensiveness and accuracy of preliminary screening are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image recognition, and particularly relates to a tooth health preliminary screening method and device based on image recognition, equipment and a medium. BACKGROUND

[0002] With the development of digital technology in the field of oral health, computer-aided diagnosis technology based on single modal images (such as visible light photos or X-ray films) has emerged. This technology assists doctors in finding specific oral problems through image analysis, thereby leading to the current tooth health screening method.

[0003] In traditional technology, visual inspection by dentists and imaging analysis for specific problems are usually relied on. For example, caries is preliminarily judged by observing the color and shape changes of the tooth surface with the naked eye, or the condition of the tooth root and alveolar bone is evaluated by separately interpreting X-ray films. This separate and single diagnosis mode constitutes the basis of the current oral health screening.

[0004] However, the current screening method or traditional method has significant problems. First, it highly depends on the personal experience of doctors, and is highly subjective, making it difficult to achieve standardized risk assessment. Second, the information of different imaging modalities (such as surface, internal, and transmission) is fragmented and cannot be integrated to form a comprehensive and three-dimensional understanding of the tooth health condition, resulting in a high rate of missed diagnosis or misdiagnosis of early or complex lesions. Finally, the traditional screening process is time-consuming and difficult to meet the needs of efficient and low-cost health preliminary screening for large populations. SUMMARY

[0005] Therefore, it is necessary to provide a tooth health preliminary screening method, device, equipment and medium based on image recognition to solve the above technical problems.

[0006] In a first aspect, the application provides a tooth health preliminary screening method based on image recognition, comprising:

[0007] Based on the obtained tooth multi-source image data, image preprocessing is performed on the tooth multi-source image data to obtain preprocessed tooth image data; the tooth multi-source image data includes visible light tooth surface images, X-ray tooth root and jaw bone images, and near-infrared caries detection images;

[0008] Multi-scale tooth features are extracted from the preprocessed tooth image data to obtain tooth feature map data; the tooth feature map data is used to represent the morphological and structural characteristics of the tooth under different imaging modalities;

[0009] The tooth feature map data is subjected to multi-feature fusion processing to obtain fused tooth feature data;

[0010] Based on the fused tooth feature data, the preset tooth anatomical structure template is combined to perform feature region division and mapping, to obtain tooth region feature data; the tooth region feature data is used to represent the features of each independent tooth or a specific tooth surface region;

[0011] The tooth region feature data is analyzed for health condition to obtain tooth health condition data; the tooth health condition data is used to represent the potential risks of the tooth in terms of caries, periodontitis and malocclusion;

[0012] According to the tooth health condition data, a tooth health screening result is generated.

[0013] In one of the embodiments, multi-scale tooth features are extracted from the preprocessed tooth image data to obtain tooth feature map data, including:

[0014] The preprocessed tooth image data is input into a preset convolutional neural network;

[0015] Local texture features of the tooth are extracted by a first layer of convolutional kernels of the convolutional neural network to obtain a first feature map;

[0016] Morphological features of the tooth are extracted by intermediate layer convolutional kernels of the convolutional neural network to obtain a second feature map;

[0017] Global semantic features of the tooth are extracted by deep layer convolutional kernels of the convolutional neural network to obtain a third feature map;

[0018] According to the first feature map, the second feature map and the third feature map, tooth feature map data is obtained.

[0019] In one of the embodiments, multi-feature fusion processing is performed on the tooth feature map data to obtain fused tooth feature data, including:

[0020] Based on a preset multi-scale feature pyramid, the first feature map, the second feature map and the third feature map are fused to obtain a multi-scale feature set;

[0021] The multi-scale feature set is subjected to cross-scale feature enhancement to obtain the fused tooth feature data.

[0022] In one of the embodiments, based on the fused tooth feature data, the preset tooth anatomical structure template is combined to perform feature region division and mapping, to obtain tooth region feature data, including:

[0023] The fused tooth feature data is aligned with the preset tooth anatomical structure template to obtain an aligned feature map;

[0024] According to the aligned feature map, each independent tooth region is divided to obtain a tooth region mask;

[0025] statistical aggregation is performed on the features in the tooth region mask to obtain a feature vector of each tooth region;

[0026] The feature vector of each tooth region is mapped with a preset dental surface region template to obtain a dental surface region feature.

[0027] The feature vector of each tooth region is integrated with the dental surface region feature to obtain tooth region feature data.

[0028] In one embodiment, the tooth region feature data is analyzed for health status to obtain tooth health status data, including:

[0029] The tooth region feature data is classified for caries risk to obtain a caries risk probability;

[0030] The tooth region feature data is predicted for periodontitis risk to obtain a periodontitis risk score;

[0031] The tooth region feature data is detected for malocclusion to obtain a malocclusion degree;

[0032] The caries risk probability, periodontitis risk score and malocclusion degree are weighted and fused to obtain a comprehensive health score;

[0033] According to the comprehensive health score, tooth health status data is generated.

[0034] In one embodiment, the tooth region feature data is classified for caries risk to obtain a caries risk probability, including:

[0035] The tooth region feature data is extracted for enamel layer microstructure change features to obtain an enamel demineralization feature vector; the enamel demineralization feature vector is used to quantitatively represent the severity of loss of enamel density and increase in porosity;

[0036] Based on the tooth region feature data, a feature component corresponding to a dentin structure region is separated to obtain a dentin permeability feature vector; the dentin permeability feature vector is used to assist in determining whether caries has penetrated the enamel layer and affected the dentin;

[0037] Based on the enamel demineralization feature vector and the dentin permeability feature vector, the following formula is used to obtain a preliminary probability of caries:

[0038]

[0039] wherein, is the preliminary probability of caries, is the enamel demineralization feature vector, is the dentin permeability feature vector, represents a vector concatenation operation, is a weight matrix, is a bias term, is a Sigmoid activation function;

[0040] According to the initial caries probability, attention weight distribution is performed on the tooth region feature data to obtain weighted region feature data;

[0041] Based on the weighted region feature data, a caries depth level is obtained;

[0042] According to the initial caries probability and the caries depth level, the following formula is used to obtain the caries risk probability:

[0043]

[0044] wherein, is a caries risk probability, is an initial caries probability, is a caries depth level, is a weight coefficient for balancing the initial caries probability, is a weight coefficient for balancing the influence of the caries depth level, and , is a maximum value of the caries depth level.

[0045] In one of the embodiments, periodontitis risk prediction is performed on the tooth region feature data to obtain a periodontitis risk score, including:

[0046] The tooth region feature data is subjected to alveolar bone height feature extraction to obtain an alveolar crest to cement-enamel junction alveolar bone height feature vector; the alveolar bone height feature vector is used to represent the severity of alveolar bone absorption;

[0047] The tooth region feature data is subjected to gingival margin morphology analysis to obtain a gingival morphology feature vector; the gingival morphology feature vector is used to represent the inflammation state of the gingival tissue;

[0048] According to the alveolar bone height feature vector and the gingival morphology feature vector, the following formula is used to obtain an initial periodontitis risk index:

[0049]

[0050] wherein, is an initial periodontitis risk index, is a feature dimension, represents an alveolar bone height feature of the i-th dimension, represents a gingival morphology feature of the i-th dimension, is a weight of the alveolar bone height feature of the i-th dimension, is a weight of the gingival morphology feature of the i-th dimension, is a weight of the alveolar bone height feature of the i-th dimension, is a weight of the gingival morphology feature of the i-th dimension, is a weight of the alveolar bone height feature of the i-th dimension, a weight of a gingival morphology feature in the one dimension;

[0051] Based on the preliminary periodontitis risk index, combined with the periodontal pocket depth features extracted from the adjacent tooth area feature data, the following formula is used to obtain the modified risk value, and the calculation formula is:

[0052]

[0053] wherein, is the modified risk value, is the preliminary periodontitis risk index, is the risk correction coefficient of the periodontal pocket depth, is the actual periodontal pocket depth feature, is the threshold value of the normal periodontal pocket depth;

[0054] The modified risk value is standardized to obtain a periodontitis risk score.

[0055] In a second aspect, the application also provides a tooth health preliminary screening device based on image recognition, comprising:

[0056] A data acquisition and preprocessing module is configured to perform image preprocessing on the tooth multi-source image data based on the acquired tooth multi-source image data, to obtain preprocessed tooth image data. The tooth multi-source image data includes visible light tooth surface images, X-ray tooth root and jaw bone images, and near-infrared caries detection images.

[0057] A tooth feature extraction module is configured to extract multi-scale tooth features from the preprocessed tooth image data to obtain tooth feature map data. The tooth feature map data is used to represent the morphological and structural characteristics of the tooth under different imaging modalities.

[0058] A tooth feature fusion module is configured to perform multi-feature fusion processing on the tooth feature map data to obtain fused tooth feature data.

[0059] A tooth region division module is configured to perform feature region division and mapping based on the fused tooth feature data and a pre-set tooth anatomical structure template to obtain tooth region feature data. The tooth region feature data is used to represent the features of each independent tooth or specific tooth surface region.

[0060] A tooth health analysis module is configured to analyze the tooth region feature data to obtain tooth health status data. The tooth health status data is used to represent the potential risks of the tooth in terms of caries, periodontitis, and malocclusion.

[0061] A tooth screening result module is configured to generate a tooth health screening result based on the tooth health status data.

[0062] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0063] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0064] The aforementioned image recognition-based method, device, equipment, and medium for initial screening of dental health integrate multimodal dental image data. After preprocessing and optimizing data quality, it uses convolutional neural networks to extract multi-scale features such as local texture, morphology, and global semantics. Then, through multi-scale feature pyramid fusion and cross-scale enhancement, it fully explores the complementary information of different imaging modalities. Combined with dental anatomical templates, it achieves precise segmentation and feature mapping of independent teeth and tooth surface regions. This allows for targeted health analysis of caries, periodontitis, and malocclusion. Through scientific feature extraction, weight allocation, and formula calculation, it quantifies various risks and performs weighted fusion, comprehensively covering key dimensions of dental health. It overcomes the limitations of single images and single features, achieving depth and comprehensiveness in feature extraction, accuracy in region segmentation, and scientific rigor in health assessment. It effectively identifies early potential health risks, providing reliable and comprehensive assessment basis for initial oral health screening, assisting in the accurate formulation of subsequent treatment decisions, improving the efficiency and targeting of oral health management, and meeting the initial screening needs in different scenarios. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying 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.

[0066] Figure 1 This is a flowchart illustrating an image recognition-based initial screening method for dental health in one embodiment.

[0067] Figure 2 This is a schematic diagram of an image recognition-based dental health screening device in one embodiment. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0069] In one embodiment, such asFigure 1 As shown, an image recognition-based method for initial screening of dental health is provided. This embodiment illustrates the application of this method to a dental screening terminal (hereinafter referred to as the terminal). It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0070] S1. Based on the acquired multi-source image data of teeth, perform image preprocessing on the multi-source image data of teeth to obtain preprocessed tooth image data.

[0071] The multi-source dental image data includes visible light tooth surface images, X-ray images of tooth roots and jawbones, and near-infrared caries detection images. Visible light tooth surface images are used to capture the appearance features of the tooth surface, such as plaque distribution and enamel surface defects; X-ray images of tooth roots and jawbones can penetrate the surface tissues of the teeth, revealing the morphology of the tooth roots, root canal structure, and the condition of the jawbone, which can help detect deeper problems such as tooth root resorption and jawbone lesions; near-infrared caries detection images are based on the difference in near-infrared light absorption between carious tissue and healthy tissue, and can be used to identify early signs of caries such as enamel demineralization.

[0072] Specifically, after acquiring multi-source image data of teeth through image acquisition equipment, the tooth screening terminal first performs standardized preprocessing. For visible light tooth surface images, a color correction algorithm is used to eliminate ambient lighting differences, and white balance processing is used to restore the true color of the teeth. For X-ray images of tooth roots and jawbones, a contrast-limited adaptive histogram equalization technique is applied to enhance the display effect of trabecular bone structure and periodontal ligament space. For near-infrared caries detection images, a non-local mean denoising algorithm is used to suppress optical scattering noise, while Laplacian pyramid fusion is used to enhance the texture features of early caries lesions. All modal images are spatially aligned using a feature point registration algorithm, with enamel boundary points and apical points as reference markers. Finally, bicubic interpolation is used to uniformly scale to standard resolution, and Z-score normalization is used to normalize pixel values, forming preprocessed tooth image data with consistent dimensions. The image acquisition equipment may include an intraoral scanner, a dental SLR camera, a dental X-ray imaging device, and a near-infrared caries detector, etc.

[0073] S2, extract multi-scale tooth features from the preprocessed tooth image data to obtain tooth feature map data.

[0074] Specifically, the tooth screening terminal uses a Convolutional Neural Network (CNN, a deep neural network with convolutional layers that extracts local features and gradually abstracts them through sliding convolutional kernels) as a feature extraction tool. During the extraction process, small-scale feature extraction (e.g., 3×3 convolutional kernels) is used to capture local texture features of the tooth surface (e.g., pits and fissures on the enamel surface, and traces of minor defects). Then, medium-scale feature extraction (e.g., 5×5 or 7×7 convolutional kernels) is used to obtain morphological features of the teeth (e.g., crown contours, cervical curves, and adjacency relationships between teeth). Finally, large-scale feature extraction (e.g., 11×11 or larger convolutional kernels or feature fusion after multiple rounds of pooling) is used to obtain global semantic features (e.g., the positional distribution of teeth and jawbone, and the overall arrangement trend of the dentition). The resulting tooth feature map data can fully characterize the morphological and structural properties of teeth under different imaging modalities. Each feature map corresponds to feature information of a specific scale or modality, laying the foundation for subsequent multi-feature fusion.

[0075] S3. Perform multi-feature fusion processing on the tooth feature map data to obtain fused tooth feature data.

[0076] For example, the tooth screening terminal first performs correlation analysis on tooth feature map data from different modalities (visible light, X-ray, near-infrared) and different scales (local, mesoscale, global) to identify the intrinsic relationships between features (such as the correlation between caries features in near-infrared images and texture changes in corresponding areas in visible light images). Subsequently, it integrates the features using methods such as feature stitching, weighted summation, or attention mechanisms: assigning higher weights to highly complementary features and suppressing redundant information. Through this processing, the fused tooth feature data can simultaneously contain multi-dimensional information such as tooth surface morphology, internal structure, and early lesions, avoiding the limitations of single features.

[0077] S4, based on the fused tooth feature data and combined with the preset tooth anatomy template, performs feature region division and mapping to obtain tooth region feature data.

[0078] For example, the tooth screening terminal, based on fused tooth feature data and combined with a preset tooth anatomical structure template, performs feature region segmentation and mapping to obtain tooth region feature data. The preset tooth anatomical structure template is a digital model constructed based on human tooth anatomy standards, including the standard number and arrangement of permanent / deciduous teeth, the morphological parameters of each tooth, and the definition of tooth surface regions (such as occlusal surface, buccal surface, and lingual surface). The terminal first aligns the fused feature data with the template spatially using a feature matching algorithm to ensure that the feature positions correspond to the anatomical structure; then, based on the tooth boundary information in the template, it divides the region of each independent tooth to form a tooth region mask; next, it performs statistical analysis (such as feature mean and variance) on the features within each mask to obtain the feature vector of a single tooth; finally, it maps the feature vector to the tooth surface region template to extract the features of each tooth surface. The final tooth region feature data can accurately represent the features of each tooth and specific tooth surfaces, providing support for targeted health analysis.

[0079] S5 analyzes the health status of the tooth region feature data to obtain tooth health status data.

[0080] For example, the dental screening terminal analyzes the health status of tooth region feature data to generate dental health status data. The analysis process revolves around three common problems: dental caries, periodontitis, and malocclusion. For dental caries, the risk of disease is assessed by identifying features such as enamel demineralization and dentin exposure. For periodontitis, the degree of inflammation is assessed based on features such as changes in alveolar bone height and abnormal gingival margin morphology. For malocclusion, the alignment is assessed by analyzing parameters such as tooth spacing and tilt angle. The terminal uses classification and regression algorithms to quantify the above features, obtaining risk indicators for each type of problem, such as the probability of dental caries risk, the periodontitis risk score, and the degree of malocclusion. Subsequently, weights for each indicator are set based on clinical experience, and a comprehensive health score is obtained through weighted calculation. The final dental health status data comprehensively reflects the potential risks of teeth in the three types of problems.

[0081] S6 generates dental health screening results based on dental health data.

[0082] Specifically, the dental screening terminal generates dental health screening results based on dental health status data. The core is to transform quantified health risk data into intuitive and understandable information to meet the usage needs in initial screening scenarios (such as user self-checks and preliminary assessments by primary healthcare providers). First, the terminal presets a health risk grading standard (e.g., low risk, medium risk, high risk), based on clinical guidelines for oral medicine and a large amount of initial screening data. Then, it compares various risk indicators in the dental health data with the grading standard to determine the overall health risk level. Next, the terminal integrates detailed risk information, providing specific descriptions of high-risk or medium-risk items (e.g., high-risk item: 0.6 probability of proximal caries in tooth #2, corresponding to superficial dentin caries depth; medium-risk item: 45 risk score for periodontitis in the lower posterior teeth region, mainly due to mild alveolar bone resorption). It also provides maintenance suggestions for low-risk items (e.g., low-risk item: malocclusion degree 0.1, recommending regular dental checkups and maintaining correct brushing techniques). Finally, the terminal outputs the screening results in the form of a text report or visual charts. The report includes the overall risk level, detailed risk descriptions for each health problem, and subsequent treatment suggestions (e.g., high-risk: recommend visiting a dentist within one week; medium-risk: recommend a follow-up examination within three months; low-risk: recommend routine checkups every six months). This ensures that users or medical personnel can clearly understand their dental health status, quickly identify potential health problems, and guide subsequent actions.

[0083] The aforementioned image recognition-based initial screening method for dental health integrates multimodal dental image data from visible light, X-rays, and near-infrared sources. After preprocessing, multi-scale feature extraction and fusion, and combining with dental anatomical templates, precise region segmentation is achieved. This allows for a comprehensive analysis of dental caries, periodontitis, and malocclusion, generating screening results. By fully leveraging the morphological and structural characteristics of teeth under different imaging modalities, it enables comprehensive and accurate initial screening of various dental health problems, effectively identifying early potential risks. This provides a reliable basis for early oral health screening and subsequent treatment decisions, facilitating timely intervention in dental health issues and improving the efficiency and effectiveness of oral health management.

[0084] In an optional embodiment, multi-scale tooth features are extracted from the preprocessed tooth image data to obtain tooth feature map data, including the following steps:

[0085] S11, input the preprocessed tooth image data into the preset convolutional neural network.

[0086] For example, the tooth screening terminal inputs preprocessed tooth image data into a pre-defined Convolutional Neural Network (CNN, a deep neural network containing convolutional layers, adept at processing grid-structured data such as images). This network is trained on a large amount of labeled tooth image data, and its network structure includes an input layer, multiple convolutional layers, pooling layers, and an output layer. The parameters of each layer are optimized through training to meet the needs of tooth feature extraction. Before input, the terminal needs to convert the preprocessed image data into a network-compatible format (e.g., adjusting the number of channels and pixel value range) to ensure that the data can be correctly read by the network. The input process essentially transforms the pixel information of the image into tensor data that the network can process, providing the raw input for feature extraction in subsequent layers.

[0087] S12 extracts local texture features of the teeth through the first convolutional kernel of the convolutional neural network to obtain the first feature map.

[0088] Specifically, the tooth screening terminal extracts local texture features of teeth using the first convolutional kernel of a CNN to obtain a first feature map. The first convolutional kernel typically uses a small size (e.g., 3×3) and its function is to capture local details in the image, such as enamel texture, minute depressions or protrusions, and traces of plaque adhesion on the tooth surface. During convolution, the kernel slides across the image, performing convolution operations with the pixel values ​​of corresponding regions to generate feature values ​​reflecting local texture differences. Each pixel in the first feature map corresponds to the texture features of a local region in the original image, enabling precise characterization of the microstructure of the tooth surface and providing a foundation for subsequent identification of lesions that rely on surface details, such as early caries.

[0089] S13, extract the morphological features of the teeth through the intermediate convolutional kernel of the convolutional neural network to obtain the second feature map.

[0090] Specifically, the tooth screening terminal extracts morphological features of teeth through intermediate convolutional kernels in a CNN to obtain a second feature map. The size of the intermediate convolutional kernels is typically larger than that of the first layer (e.g., 5×5), and due to feature aggregation from preceding layers, it can capture a wider range of feature information. This layer primarily focuses on the overall morphological contour of the tooth, such as the shape of the crown, the curvature of the neck, and the orientation of the root (for X-ray images), while also identifying the adjacency relationships between teeth (e.g., contact point location, gap size). Through convolutional operations in the intermediate layer, local texture features are combined into more abstract morphological features. The second feature map reflects the macroscopic morphological characteristics of the tooth, providing a basis for determining whether there are morphological abnormalities (e.g., malformed teeth).

[0091] S14: Extract global semantic features of teeth through deep convolutional kernels of convolutional neural networks to obtain the third feature map.

[0092] Specifically, the tooth screening terminal extracts global semantic features of teeth using deep convolutional kernels in a CNN to obtain a third feature map. The deep layers of the CNN are the part of the network closest to the output layer. After multiple rounds of convolution, pooling, and feature fusion, the dimensionality of the deep feature map is significantly reduced, but the receptive field covers the entire tooth image or most of its area, enabling it to capture global contextual information and meet the extraction requirements of global semantic features of teeth (such as the positional relationship between teeth and jawbone, the overall arrangement trend of the dentition, the boundary features between teeth and gingiva, and the spatial distribution among multiple teeth, all of which belong to global information). The role of the deep convolutional kernels is to perform higher-level abstraction and association of the morphological features extracted from the intermediate layers. For example, it combines the morphological features of a single tooth with the morphological features of adjacent teeth and jawbone features to analyze whether the overall position of the tooth in the oral cavity is normal; it combines the edge features of the tooth with the edge features of the gingiva to determine whether there is gingival recession or hyperplasia. The features in the third feature map are no longer isolated local or morphological information, but global features containing semantic relationships (such as a tooth with a normal shape but located in an abnormal position in the jawbone, or the overall dental arch showing a crowded arrangement). It can represent the overall condition of the teeth and surrounding tissues from a global perspective, providing feature support from a global perspective for subsequent multi-feature fusion, and avoiding the one-sidedness that may be caused by local feature analysis.

[0093] S15. Based on the first feature map, the second feature map, and the third feature map, obtain the tooth feature map data.

[0094] For example, the tooth screening terminal obtains tooth feature map data based on the first feature map, the second feature map, and the third feature map. The integration process is not a simple stitching together, but rather uses a feature fusion mechanism (such as skip connections) to associate features at different scales, allowing local texture features and global semantic features to complement each other. For instance, local texture details in the first feature map are mapped to the morphological region of the second feature map, and then combined with the global information of the third feature map to form a hierarchical feature representation. The final tooth feature map data contains microscopic texture information, covers mesoscopic morphological features, and macroscopic semantic features, comprehensively representing the characteristics of teeth at different scales and providing complete feature input for subsequent multi-feature fusion processing.

[0095] In an optional embodiment, multi-feature fusion processing is performed on the tooth feature map data to obtain fused tooth feature data, including the following steps:

[0096] S21, based on the preset multi-scale feature pyramid, the first feature map, the second feature map, and the third feature map are fused to obtain a multi-scale feature set.

[0097] For example, the tooth screening terminal fuses a first feature map, a second feature map, and a third feature map based on a preset multi-scale feature pyramid to obtain a multi-scale feature set. The multi-scale feature pyramid is a hierarchical feature structure, with the bottom layer corresponding to high-resolution local features (the first feature map), the middle layer corresponding to medium-resolution morphological features (the second feature map), and the top layer corresponding to low-resolution global features (the third feature map). The terminal upsamples the high-level features (e.g., bilinear interpolation) to the same resolution as the low-level features, and then fuses the features from different levels through element-wise addition or concatenation. This process preserves the advantages of features at each scale, ensuring that the fused multi-scale feature set contains both fine local details and global semantic information, laying the foundation for subsequent feature enhancement.

[0098] S22, cross-scale feature enhancement is performed on the multi-scale feature set to obtain fused tooth feature data.

[0099] For example, the tooth screening terminal performs cross-scale feature enhancement on a multi-scale feature set to obtain fused tooth feature data. Cross-scale feature enhancement aims to improve the discriminative ability of features by strengthening the correlation between features at different scales. The terminal uses channel attention and spatial attention to process the multi-scale feature set. Channel attention is used to highlight feature channels that are more important for health assessment (such as feature channels reflecting caries), and spatial attention is used to enhance the feature response of suspected lesion areas. Through this processing, redundant information is effectively suppressed, the expression of key features is enhanced, and the final fused tooth feature data retains multi-scale information while having stronger relevance and discriminative power, making it more suitable for subsequent feature region segmentation and health analysis.

[0100] In an optional embodiment, based on fused tooth feature data and combined with a preset tooth anatomy template, feature regions are divided and mapped to obtain tooth region feature data, including the following steps:

[0101] S31, Align the fused tooth feature data with the preset tooth anatomy template to obtain the aligned feature map.

[0102] For example, the tooth screening terminal aligns the fused tooth feature data with a preset tooth anatomical structure template to obtain an aligned feature map. The alignment process is based on anatomical landmarks (such as cusps, root apex, cementoenamel junction, etc.) in the template. The terminal identifies the corresponding landmarks in the fused feature data using a feature matching algorithm, and then adjusts the spatial position of the feature data through affine transformation or projection transformation so that the tooth structure in the feature is consistent with the standard structure in the template in spatial coordinates.

[0103] The dental anatomical structure template was obtained through the following methods: collecting thousands of standard dental panoramic X-ray images, intraoral scanning 3D models, and cone-beam CT data annotated by professional physicians; extracting dental arch morphological feature vectors and tooth spatial distribution patterns through principal component analysis; establishing statistical priors for crown morphology, root inclination, and adjacency relationships using a probabilistic graphical model; and mapping individual tooth positions to the standard dental arch coordinate system using a non-rigid registration algorithm. Finally, a parametric template library containing tooth anatomical boundaries, tooth surface partitions, and spatial topological relationships was generated. This template supports adaptation to jawbone morphological variations in different patients through affine transformation.

[0104] S32, based on the aligned feature map, divide each independent tooth region to obtain the tooth region mask.

[0105] For example, the tooth screening terminal divides the aligned feature map into individual tooth regions, obtaining a tooth region mask. Region division employs a semantic segmentation algorithm, based on the feature differences (such as shape, texture, and grayscale values) of different teeth in the aligned feature map, combined with the boundary information of each tooth in the template, to determine the pixel range of each tooth. The tooth region mask is a binary image where pixels belonging to a specific tooth are labeled with specific values, and other areas are background values. The mask clearly defines the spatial range of each individual tooth, providing clear region boundaries for subsequent feature aggregation.

[0106] S33: Statistically aggregate the features within the tooth region mask to obtain the feature vector of each tooth region.

[0107] For example, the tooth screening terminal performs statistical aggregation on the features within the tooth region mask to obtain feature vectors for each tooth region. Statistical aggregation includes calculating statistics such as the mean, variance, maximum, and minimum values ​​of the features within the mask, and compressing high-dimensional features within the region into fixed-dimensional vectors through pooling operations (such as global average pooling and global max pooling). This process integrates local features within the region into overall features that can represent the entire tooth. Each element in the feature vector corresponds to a quantitative indicator of a certain characteristic of the tooth (such as average texture complexity or overall morphological parameters), facilitating subsequent mapping with tooth surface templates and health analysis.

[0108] S34: Map the feature vectors of each tooth region to the preset tooth surface region template to obtain the tooth surface region features.

[0109] For example, the tooth screening terminal maps the feature vectors of each tooth region to a preset tooth surface region template to obtain tooth surface region features. During the mapping process, the terminal first determines the tooth type based on the morphological features (such as tooth length, width, and number of cusps) in the tooth region feature vectors (e.g., determining it as a molar based on the number of cusps); then, it calls the tooth surface region template for the corresponding tooth type and, through coordinate mapping (converting the standard coordinate range of the tooth surface in the template to the actual coordinates corresponding to the tooth region feature vector), locates the specific range of each tooth surface within the tooth region; next, it extracts features within each tooth surface range (such as pit and fissure texture features of the occlusal surface, contact gap features of the proximal surfaces, and morphological features of the buccal surface); finally, it combines the features of each tooth surface to form tooth surface region features. These features can more precisely reflect the characteristics of different parts of the tooth. For example, occlusal surface features can be used to assess the risk of caries (pits and fissures easily trap food debris, leading to a high incidence of caries), while proximal surface features can be used to assess the risk of periodontitis (proximal surfaces easily form plaque, leading to gingivitis), providing support for targeted analysis of health problems on each tooth surface.

[0110] The tooth surface templates are constructed based on the definition of tooth surfaces in dental anatomy. Customized templates are designed for different types of teeth (such as incisors, canines, premolars, and molars). For example, the molar template includes standard parameters for five tooth surfaces: occlusal, mesial, distal, buccal, and lingual. The incisor template includes standard parameters for four tooth surfaces: labial, lingual, mesial, and distal. Each tooth surface is marked in the template with its standard location range (e.g., the occlusal surface is located at the occlusal end of the tooth, occupying 1 / 3 of the tooth height), anatomical features (e.g., the occlusal surface has cusps and fissures, while the proximal surface is planar), and corresponding standard features (e.g., the occlusal surface is predominantly characterized by fissures, while the proximal surface is predominantly smooth).

[0111] S35 integrates the feature vectors of each tooth region with the features of the tooth surface region to obtain tooth region feature data.

[0112] For example, the tooth screening terminal integrates the feature vectors of each tooth region with the features of the tooth surface region to obtain tooth region feature data. The integration process associates and stores tooth-level features with tooth surface-level features, forming a hierarchical feature structure: each tooth corresponds to a feature vector, which also includes the features of its subordinate tooth surfaces. This structure retains the overall feature information of the tooth while including the detailed features of each tooth surface, meeting the needs of health analysis at different granularities (such as comprehensively assessing the health status of a tooth or specifically analyzing the disease risk of a particular tooth surface), providing comprehensive and refined feature input for subsequent health status analysis.

[0113] In an optional embodiment, a health status analysis is performed on the dental region feature data to obtain dental health status data, including the following steps:

[0114] S41, classify dental caries risk based on tooth region feature data to obtain dental caries risk probability.

[0115] For example, the tooth screening terminal classifies dental caries risk based on tooth region feature data to obtain the caries risk probability. The caries risk classification is based on the pathological characteristics of caries development, focusing on analyzing features such as enamel demineralization and dentin exposure. The terminal processes the feature data using a pre-set classification model (such as logistic regression or support vector machine). The model learns from the differences between lesion and healthy features in historical labeled data, outputting the caries risk probability for each tooth and tooth surface. This probability reflects the degree of matching between the feature data and caries features; a higher value indicates a greater likelihood of caries, providing a quantitative indicator for the caries dimension in subsequent comprehensive health assessments.

[0116] S42, periodontitis risk prediction is performed on tooth region feature data to obtain periodontitis risk score.

[0117] For example, the tooth screening terminal performs periodontitis risk prediction on tooth region feature data to obtain a periodontitis risk score. Periodontitis risk prediction focuses on features related to periodontal tissues, including changes in alveolar bone height, gingival margin morphology, and periodontal pocket depth. The terminal uses a regression model to quantify these features. The model outputs a corresponding risk value based on the degree of abnormality of the feature (such as alveolar bone resorption or the extent of gingival redness and swelling). This risk value is then standardized to convert it into a periodontitis risk score. A higher score indicates a potentially more severe degree of periodontal tissue inflammation and damage. This score provides a basis for comprehensive assessment from the perspective of periodontal health.

[0118] S43, perform malocclusion detection on the dental region feature data to obtain the degree of malocclusion.

[0119] For example, a dental screening terminal detects malocclusion by analyzing regional dental feature data to determine the degree of malocclusion. Malocclusion detection is achieved by analyzing tooth alignment parameters, including the size of interdental spaces, tilt angles, rotational degree, and overall symmetry. The terminal calculates the deviations of these parameters from corresponding parameters in a standard dental arch template to obtain quantified malocclusion indicators. These indicators are then integrated into a degree of malocclusion (e.g., mild, moderate, severe). This degree reflects the regularity of the dental alignment and provides a reference for assessing dental aesthetics and function (such as chewing and speech), making it an important component of comprehensive health assessment.

[0120] S44 is a weighted fusion of caries risk probability, periodontitis risk score, and malocclusion degree to obtain a comprehensive health score.

[0121] For example, the dental screening terminal weights and integrates the probability of dental caries risk, periodontitis risk score, and degree of malocclusion to obtain a comprehensive health score. First, the terminal determines the weighting coefficients: if dental caries and periodontitis are not intervened in time, they may lead to serious complications such as pulpitis and alveolar bone atrophy, posing a greater threat to oral health. The weights are determined using the Delphi Method (weights are determined through multiple rounds of expert consultation) combined with clinical data statistics (analyzing the impact of different health problems on overall oral health) to ensure the scientific validity of the weights. Then, the terminal standardizes the three indicators: mapping the probability of dental caries risk, periodontitis risk score, and degree of malocclusion to a uniform 0-1 range to avoid any single indicator dominating the score due to dimensional differences. Finally, the comprehensive health score is calculated according to the weighted formula. This score comprehensively reflects the overall status of teeth in terms of disease risk (dental caries, periodontitis) and morphological arrangement (malocclusion).

[0122] S45 generates dental health data based on the comprehensive health score.

[0123] For example, the dental screening terminal generates dental health data based on a comprehensive health score. First, the terminal presets a grading standard for the comprehensive health score, which is based on clinical guidelines in oral medicine and a large amount of initial screening data to ensure the rationality and clinical reference value of the grading. Then, the terminal compares the calculated comprehensive health score with the grading standard to determine the overall dental health risk level. Simultaneously, the terminal supplements the risk details based on the specific circumstances of three individual indicators: if the comprehensive score is medium risk, and the main risk is a high periodontitis risk score, it is described as: Comprehensive health level: Medium risk; Main risk sources: Periodontitis 0.6, Caries risk probability 0.2, Malocclusion degree 0.1; if the comprehensive score is high risk, and both the Caries risk probability and the periodontitis risk score are high, the specific values ​​of the two high-risk indicators and the corresponding tooth regions are detailed (e.g., Caries risk probability on the occlusal surface of the 3rd tooth 0.8, and a standardized periodontitis risk score of 0.9 in the lower posterior teeth region). In addition, the terminal also records abnormal characteristics of each indicator (e.g., high Caries risk corresponds to significant enamel demineralization, and high Periodontitis risk corresponds to mild alveolar bone resorption). The final dental health data includes overall risk level, details of individual indicators, sources of risk, and abnormal characteristics, providing comprehensive and detailed health assessment information for the subsequent generation of screening results, ensuring the accuracy and practicality of the screening results.

[0124] In an optional embodiment, dental caries risk classification is performed on tooth region feature data to obtain caries risk probabilities, including the following steps:

[0125] S51, extract the microstructural changes of the enamel layer from the tooth region feature data to obtain the enamel demineralization feature vector.

[0126] For example, the tooth screening terminal extracts enamel microstructure change features from tooth region feature data to obtain an enamel demineralization feature vector. Enamel demineralization is a typical feature of early caries, manifested as decreased enamel density and increased porosity. The terminal analyzes the light reflection characteristics of the enamel region in near-infrared images (the reflectivity of demineralized areas differs from that of normal areas) and texture changes in visible light images (demineralized areas may show chalky white spots and coarse texture), extracting relevant features (such as the area of ​​abnormal reflectivity regions and texture complexity), and quantifying these features into vector form. Each element of the enamel demineralization feature vector corresponds to a quantitative index of the degree of demineralization, used to accurately characterize the severity of the impact of caries on the enamel.

[0127] S52, based on tooth region feature data, separates the feature components corresponding to the dentin structure region to obtain the dentin permeability feature vector.

[0128] For example, the tooth screening terminal separates feature components corresponding to dentin structural regions based on tooth region feature data, obtaining a dentin permeability feature vector. Dentin permeability reflects the degree of opening of dentinal tubules; caries penetrating the enamel layer leads to increased dentin permeability. The terminal analyzes density changes in dentin regions using X-ray images (density decreases in caries-affected areas) and, combined with feature separation algorithms (such as independent component analysis), extracts dentin-related feature components from the tooth region feature data, including the extent of density abnormalities and the rate of change in permeability. The dentin permeability feature vector quantifies these components, helping to determine whether caries has progressed to the dentin layer and providing a basis for assessing caries depth.

[0129] S53, based on the enamel demineralization feature vector and the dentin permeability feature vector, the preliminary probability of dental caries is obtained using the following formula:

[0130]

[0131] in, This is a preliminary probability of dental caries. This is the characteristic vector of glaze demineralization. This is the dentin permeability feature vector. This represents a vector concatenation operation. This is the weight matrix. For bias terms, This is the Sigmoid activation function.

[0132] In the above formula, This represents the preliminary probability of tooth decay, indicating the initial likelihood that tooth decay exists in the current tooth area. Its value ranges from [value missing]. Between 1 and 0, the closer the value is to 1, the higher the probability of dental caries; the closer the value is to 0, the lower the probability. This is a feature vector for enamel demineralization, used to quantify the microstructural changes in tooth enamel caused by caries, such as the degree of enamel density loss and the severity of increased porosity. These features are usually extracted from near-infrared and X-ray images and are key markers of early caries (enamel layer). This is a dentin permeability feature vector used to help determine whether caries has penetrated the enamel layer and affected the dentin, such as changes in the permeability of dentinal tubules or abnormal dentin density. It can supplement the shortcomings of enamel features in determining caries depth. This indicates a vector concatenation operation that directly connects the enamel demineralization feature vector and the dentin permeability feature vector in terms of dimension, forming a longer joint feature vector. The weight matrix is ​​a parameter matrix obtained through training, used to measure the influence of each dimension of the enamel demineralization feature vector and dentin permeability feature vector on the caries probability. The bias term is a parameter learned through training, used to adjust the result of linear combinations and improve the model's ability to fit complex situations. The sigmoid activation function is used to map the result of a linear combination to... The interval visually reflects the initial probability of caries development. For example, the tooth screening terminal uses the formula above to obtain the initial probability of caries based on the enamel demineralization feature vector and the dentin permeability feature vector. By integrating the features of enamel and dentin, the risk of caries is preliminarily quantified.

[0133] S54. Based on the preliminary probability of dental caries, attention weights are assigned to the tooth region feature data to obtain weighted region feature data.

[0134] For example, the tooth screening terminal assigns attention weights to tooth region feature data based on the initial probability of caries, resulting in weighted region feature data. The core of this attention weight allocation is to strengthen the features of high-risk areas and weaken the influence of low-risk areas. The terminal assigns a corresponding attention weight to each location in the tooth region feature data based on the distribution of the initial caries probability (the higher the probability of a region, the greater its corresponding weight), and then applies the weights to the feature data through element-wise multiplication. This process makes the feature data more focused on areas where caries may exist, improving the accuracy of subsequent depth level judgments and reducing interference from irrelevant region features.

[0135] S55, based on weighted regional feature data, yields the caries depth grade.

[0136] For example, the dental screening terminal uses weighted regional feature data as input and combines the depth representation capabilities of different imaging modalities to determine the caries depth level. The terminal employs a semantic segmentation model (such as U-Net, a encoder-decoder network commonly used for medical image segmentation) or a regression model. Based on the pixel-level classification results output by the model (such as superficial enamel, deep enamel, superficial dentin, and deep dentin), the caries depth is divided into corresponding levels. The level classification follows the oral medicine caries classification standards, with each level corresponding to a specific pathological stage. For example, superficial enamel caries corresponds to early caries, and deep dentin caries corresponds to the caries progression stage, providing a quantitative basis for subsequent risk probability correction in terms of depth dimension.

[0137] S56. Based on the preliminary probability of caries and the caries depth grade, the caries risk probability is obtained using the following formula:

[0138]

[0139] in, This represents the probability of dental caries risk. This is a preliminary probability of dental caries. The degree of caries is graded. To balance the weighting coefficients of the initial probability of dental caries, To balance the weighting coefficients of the impact of caries depth grade, and , This represents the maximum value for the caries depth grade.

[0140] In the above formula, The caries risk probability represents the final caries risk probability of the current tooth area, with a value range of [value missing]. The closer it is to 1, the higher the risk. This is a preliminary probability of dental caries. The caries depth grade indicates the pathological depth of the caries (such as superficial enamel, deep enamel, superficial dentin, deep dentin, etc.). The higher the value, the deeper and more severe the lesion. The weighting coefficients for balancing the initial probabilities are determined from clinical data or model training. The weighting coefficients for balancing the impact of caries depth severity are used to adjust the influence of caries depth severity on the final risk, and satisfy the following conditions: , This represents the maximum value for the caries depth grade, a preset upper limit for the depth grade (e.g., if caries depth is divided into 4 grades, then...). , used to Standardization to The interval is adjusted to align with the dimension of the initial caries probability. The dental screening terminal uses the above formula to calculate the caries risk probability based on the initial caries probability and the caries depth grade.

[0141] In an optional embodiment, periodontitis risk prediction is performed on tooth region feature data to obtain a periodontitis risk score, including the following steps:

[0142] S61, alveolar bone height features are extracted from the tooth region feature data to obtain the alveolar bone height feature vector from the alveolar ridge crest to the cementoenamel boundary.

[0143] For example, the tooth screening terminal extracts alveolar bone height features from tooth region feature data to obtain an alveolar bone height feature vector from the alveolar ridge crest to the cementoenamel junction. Reduced alveolar bone height is a typical manifestation of periodontitis, and the distance between the alveolar ridge crest and the cementoenamel junction is a key indicator for assessing alveolar bone resorption. The terminal analyzes changes in this distance using X-ray images, extracts relevant features (such as distance values, rate of change of distance, and distribution range of the resorption area), and quantifies these features into vector form. Each element of the alveolar bone height feature vector corresponds to a quantitative index of alveolar bone resorption, used to accurately characterize the severity of alveolar bone impaction by periodontitis.

[0144] S62, perform gingival margin morphology analysis on tooth region feature data to obtain gingival morphology feature vector.

[0145] For example, the tooth screening terminal performs gingival margin morphology analysis on tooth region feature data to obtain gingival morphology feature vectors. Abnormal gingival margin morphology (such as redness, recession, and irregularity) is an important sign of periodontitis. The terminal analyzes the position, contour, and color changes of the gingival margin (inflammatory areas may appear red and swollen, with a darker color) using visible light images, extracts relevant features (such as the amount of gingival recession, margin irregularity, and area of ​​abnormal color regions), and quantifies these features into vector form. The gingival morphology feature vector is used to characterize the inflammatory state of the gingival tissue, providing a basis for assessing the activity level of periodontitis.

[0146] S63, based on the alveolar bone height feature vector and the gingival morphology feature vector, the preliminary periodontitis risk index is obtained using the following formula:

[0147]

[0148] in, This serves as a preliminary risk index for periodontitis. For feature dimension, Indicates the first Alveolar bone height characteristics in several dimensions Indicates the first Gingival morphological features in 3 dimensions For the first The weights of alveolar bone height features in each dimension. For the first The weights of each dimension of gingival morphological features.

[0149] In the above formula, This is a preliminary periodontitis risk index, used to initially characterize the risk level of periodontitis. The higher the value, the higher the risk of periodontitis. The feature dimension is the total number of dimensions of the alveolar bone height features and gingival morphology features involved in the calculation. Indicates the first The alveolar bone height features in 3 dimensions are used to quantify the pathological changes in the alveolar bone caused by periodontitis. Indicates the first The morphological characteristics of the gums are analyzed in multiple dimensions to quantify the morphological abnormalities of the gums caused by inflammation. For the first The weights of the alveolar bone height features in each dimension are used to reflect the importance of these features in the diagnosis of periodontitis. For the first The weights of each dimension of gingival morphological feature are used to reflect the importance of that gingival feature in the diagnosis of periodontitis. The tooth screening terminal uses the above formula to obtain a preliminary periodontitis risk index based on the alveolar bone height feature vector and the gingival morphological feature vector.

[0150] S64, based on the preliminary periodontitis risk index and combined with the periodontal pocket depth features extracted from the feature data of adjacent teeth, the corrected risk value is obtained using the following formula:

[0151]

[0152] in, This is the corrected risk value. This serves as a preliminary risk index for periodontitis. The risk correction factor for periodontal pocket depth. It is a feature of the actual periodontal pocket depth. This is the threshold for normal periodontal pocket depth.

[0153] In the above formula, The revised risk value is the final quantitative value of periodontitis risk after integrating alveolar bone, gingival morphology, and periodontal pocket depth. The higher the value, the higher the risk. This serves as a preliminary risk index for periodontitis. This is a risk correction factor for periodontal pocket depth, used to adjust the degree of influence of periodontal pocket depth on the final risk, and is determined by statistical analysis of clinical data. It is the actual periodontal pocket depth feature; the actual periodontal pocket depth value (such as millimeters) extracted from the feature data of adjacent tooth regions is the direct pathological manifestation of the periodontal pocket formation stage in periodontitis. The threshold for normal periodontal pocket depth, set according to oral medicine standards, is used to distinguish between physiological and pathological depths. The tooth screening terminal uses a preliminary periodontitis risk index, combined with periodontal pocket depth features extracted from adjacent tooth region feature data, to obtain a corrected risk value using the above formula.

[0154] S65 standardizes the corrected risk values ​​to obtain a periodontitis risk score.

[0155] For example, the dental screening terminal standardizes the corrected risk values ​​to obtain a periodontitis risk score. The purpose of standardization is to convert the corrected risk values ​​into a uniform order of magnitude for easier understanding and comparison. The terminal uses a min-max normalization method to map the corrected risk values ​​to a preset scoring range. The standardized score clearly reflects the relative level of periodontitis risk, providing a standardized periodontal health indicator for comprehensive health assessment.

[0156] The aforementioned image recognition-based initial screening method for dental health utilizes preprocessing of multi-source dental images, multi-scale feature extraction and fusion, and precise region segmentation using dental anatomical templates. This allows for comprehensive analysis of potential risks related to dental caries, periodontitis, and malocclusion, generating screening results. By employing multi-scale feature and multi-feature fusion techniques, the method fully leverages the morphological, structural, and textural features of teeth. Combined with precise region segmentation and mapping, it achieves comprehensive and accurate initial screening for various dental health problems. This effectively identifies early potential risks, providing a reliable basis for early oral health screening and subsequent treatment decisions, facilitating timely intervention in dental health issues, and improving the efficiency and effectiveness of oral health management.

[0157] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0158] Based on the same inventive concept, this application also provides an image recognition-based dental health screening device for implementing the image recognition-based dental health screening method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more image recognition-based dental health screening device embodiments provided below can be found in the limitations of the image recognition-based dental health screening method described above, and will not be repeated here.

[0159] In one exemplary embodiment, such as Figure 2 As shown, an image recognition-based dental health screening device 200 is provided, comprising:

[0160] The data acquisition and preprocessing module 201 is used to perform image preprocessing on the acquired multi-source image data of teeth to obtain preprocessed tooth image data; the multi-source image data of teeth includes visible light tooth surface images, X-ray images of tooth roots and jawbones, and near-infrared caries detection images;

[0161] The tooth feature extraction module 202 is used to extract multi-scale tooth features from the preprocessed tooth image data to obtain tooth feature map data; the tooth feature map data is used to characterize the morphological and structural characteristics of teeth under different imaging modalities.

[0162] The tooth feature fusion module 203 is used to perform multi-feature fusion processing on the tooth feature map data to obtain fused tooth feature data.

[0163] The tooth region segmentation module 204 is used to segment and map feature regions based on fused tooth feature data and a preset tooth anatomy template to obtain tooth region feature data; the tooth region feature data is used to characterize the features of each independent tooth or a specific tooth surface region.

[0164] The dental health analysis module 205 is used to analyze the health status of dental regional feature data to obtain dental health status data; the dental health status data is used to characterize the potential risks of teeth in terms of caries, periodontitis and malocclusion.

[0165] The dental screening results module 206 is used to generate dental health screening results based on dental health status data.

[0166] Furthermore, the tooth feature extraction module 202 is also used for:

[0167] The preprocessed tooth image data is input into a pre-defined convolutional neural network;

[0168] The first feature map is obtained by extracting local texture features of the teeth through the first convolutional kernel of the convolutional neural network.

[0169] The morphological features of teeth are extracted by the intermediate convolutional kernels of the convolutional neural network to obtain the second feature map;

[0170] The global semantic features of teeth are extracted using deep convolutional kernels of a convolutional neural network to obtain a third feature map;

[0171] Based on the first feature map, the second feature map, and the third feature map, tooth feature map data is obtained.

[0172] Furthermore, the tooth feature fusion module 203 is also used for:

[0173] Based on the preset multi-scale feature pyramid, the first feature map, the second feature map, and the third feature map are fused to obtain a multi-scale feature set;

[0174] Cross-scale feature enhancement is performed on the multi-scale feature set to obtain fused tooth feature data.

[0175] Furthermore, the tooth region segmentation module 204 is also used for:

[0176] The fused tooth feature data is aligned with a preset tooth anatomy template to obtain an aligned feature map.

[0177] Based on the aligned feature map, each independent tooth region is divided, and the tooth region mask is obtained.

[0178] Statistical aggregation is performed on the features within the tooth region mask to obtain the feature vector of each tooth region;

[0179] The feature vectors of each tooth region are mapped to a preset tooth surface region template to obtain the tooth surface region features;

[0180] By integrating the feature vectors of each tooth region with the features of the tooth surface region, tooth region feature data is obtained.

[0181] Furthermore, the dental health analysis module 205 is also used for:

[0182] Dental caries risk classification is performed on tooth region feature data to obtain the caries risk probability;

[0183] Periodontitis risk is predicted using tooth region feature data to obtain a periodontitis risk score;

[0184] The degree of malocclusion is determined by analyzing the characteristic data of the tooth region.

[0185] A comprehensive health score is obtained by weighting and integrating the caries risk probability, periodontitis risk score, and malocclusion degree.

[0186] Dental health data is generated based on a comprehensive health score.

[0187] Furthermore, the dental health analysis module 205 is also used for:

[0188] The enamel microstructure change features were extracted from the tooth region feature data to obtain the enamel demineralization feature vector; the enamel demineralization feature vector is used to quantitatively characterize the severity of enamel density loss and porosity increase.

[0189] Based on tooth region feature data, feature components corresponding to the dentin structure region are separated to obtain the dentin permeability feature vector; the dentin permeability feature vector is used to help determine whether caries has penetrated the enamel layer and affected the dentin;

[0190] Based on the enamel demineralization feature vector and the dentin permeability feature vector, the preliminary probability of dental caries is obtained using the following formula:

[0191]

[0192] in, This is a preliminary probability of dental caries. This is the characteristic vector of glaze demineralization. This is the dentin permeability feature vector. This represents a vector concatenation operation. This is the weight matrix. For bias terms, Use the Sigmoid activation function;

[0193] Based on the preliminary probability of dental caries, attention weights are assigned to the tooth region feature data to obtain weighted region feature data.

[0194] Based on weighted regional feature data, the caries depth level is obtained;

[0195] Based on the preliminary probability of dental caries and the degree of caries, the following formula can be used to obtain the probability of dental caries risk:

[0196]

[0197] in, This represents the probability of dental caries risk. This is a preliminary probability of dental caries. The degree of caries is graded. To balance the weighting coefficients of the initial probability of dental caries, To balance the weighting coefficients of the impact of caries depth grade, and , This represents the maximum value for the caries depth grade.

[0198] Furthermore, the dental health analysis module 205 is also used for:

[0199] Alveolar bone height features were extracted from the tooth region feature data to obtain the alveolar bone height feature vector from the alveolar ridge crest to the cementoenamel junction; the alveolar bone height feature vector is used to characterize the severity of alveolar bone resorption.

[0200] Gingival margin morphology analysis was performed on tooth region feature data to obtain gingival morphology feature vectors; these vectors were used to characterize the inflammatory state of gingival tissue.

[0201] Based on the alveolar bone height feature vector and the gingival morphology feature vector, the preliminary periodontitis risk index is obtained using the following formula:

[0202]

[0203] in, This serves as a preliminary risk index for periodontitis. For feature dimension, Indicates the first Alveolar bone height characteristics in several dimensions Indicates the first Gingival morphological features in 3 dimensions For the first The weights of alveolar bone height features in each dimension. For the first Weights of gingival morphological features in each dimension;

[0204] Based on the preliminary periodontitis risk index, and combined with the periodontal pocket depth features extracted from the feature data of adjacent teeth, the corrected risk value is obtained using the following formula:

[0205]

[0206] in, This is the corrected risk value. This serves as a preliminary risk index for periodontitis. The risk correction factor for periodontal pocket depth. It is a feature of the actual periodontal pocket depth. The threshold for normal periodontal pocket depth;

[0207] The corrected risk values ​​were standardized to obtain a periodontitis risk score.

[0208] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the image recognition-based dental health screening method as described above.

[0209] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0210] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0211] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for preliminary screening of dental health based on image recognition, characterized in that, The method includes: Based on the acquired multi-source image data of teeth, image preprocessing is performed on the multi-source image data of teeth to obtain preprocessed tooth image data; the multi-source image data of teeth includes visible light tooth surface images, X-ray images of tooth roots and jawbones, and near-infrared caries detection images; Multi-scale tooth features are extracted from the preprocessed tooth image data to obtain tooth feature map data; the tooth feature map data is used to characterize the morphological and structural properties of teeth under different imaging modalities. The tooth feature map data is subjected to multi-feature fusion processing to obtain fused tooth feature data; Based on the fused tooth feature data, and combined with a preset tooth anatomy template, feature regions are divided and mapped to obtain tooth region feature data; the tooth region feature data is used to characterize the features of each individual tooth or a specific tooth surface region. Health status analysis is performed on the tooth region feature data to obtain tooth health status data; the tooth health status data is used to characterize the potential risks of teeth in terms of caries, periodontitis and malocclusion. Based on the dental health data, dental health screening results are generated.

2. The method according to claim 1, characterized in that, The step of extracting multi-scale tooth features from the preprocessed tooth image data to obtain tooth feature map data includes: The preprocessed tooth image data is input into a preset convolutional neural network; The first feature map is obtained by extracting local texture features of the teeth through the first convolutional kernel of the convolutional neural network. The morphological features of the teeth are extracted through the intermediate convolutional kernels of the convolutional neural network to obtain the second feature map; The global semantic features of the teeth are extracted using the deep convolutional kernels of the convolutional neural network to obtain the third feature map; The tooth feature map data is obtained based on the first feature map, the second feature map, and the third feature map.

3. The method according to claim 2, characterized in that, The process of performing multi-feature fusion processing on the tooth feature map data to obtain fused tooth feature data includes: Based on the preset multi-scale feature pyramid, the first feature map, the second feature map, and the third feature map are fused to obtain a multi-scale feature set; Cross-scale feature enhancement is performed on the multi-scale feature set to obtain the fused tooth feature data.

4. The method according to claim 1, characterized in that, Based on the fused tooth feature data, and combined with a preset tooth anatomical structure template, feature regions are divided and mapped to obtain tooth region feature data, including: The fused tooth feature data is aligned with a preset tooth anatomy template to obtain an aligned feature map; Based on the aligned feature map, each independent tooth region is divided to obtain a tooth region mask. The features within the tooth region mask are statistically aggregated to obtain the feature vector of each tooth region. The feature vectors of each tooth region are mapped to a preset tooth surface region template to obtain the tooth surface region features; By integrating the feature vectors of each tooth region with the features of the tooth surface region, tooth region feature data is obtained.

5. The method according to claim 1, characterized in that, The health status analysis of the tooth region feature data yields tooth health status data, including: The dental region feature data is used to classify caries risk to obtain caries risk probability; Periodontitis risk is predicted based on the tooth region feature data to obtain a periodontitis risk score; The malocclusion was detected by analyzing the tooth region feature data to determine the degree of malocclusion. The caries risk probability, the periodontitis risk score, and the degree of malocclusion are weighted and fused to obtain a comprehensive health score. The dental health status data is generated based on the comprehensive health score.

6. The method according to claim 5, characterized in that, The process of classifying the dental region feature data to obtain the dental caries risk probability includes: The enamel microstructure change features are extracted from the tooth region feature data to obtain the enamel demineralization feature vector; the enamel demineralization feature vector is used to quantitatively characterize the severity of enamel density loss and porosity increase; Based on the tooth region feature data, feature components corresponding to the dentin structure region are separated to obtain the dentin permeability feature vector; the dentin permeability feature vector is used to help determine whether caries has penetrated the enamel layer and affected the dentin; Based on the enamel demineralization feature vector and the dentin permeability feature vector, the preliminary probability of dental caries is obtained using the following formula: in, This is a preliminary probability of dental caries. This is the demineralization feature vector of the glaze. This is the dentin permeability feature vector. This represents a vector concatenation operation. This is the weight matrix. For bias terms, Use the Sigmoid activation function; Based on the preliminary probability of dental caries, attention weights are assigned to the tooth region feature data to obtain weighted region feature data; Based on the weighted regional feature data, the caries depth level is obtained; Based on the preliminary probability of dental caries and the degree of dental caries, the following formula is used to obtain the probability of dental caries risk: in, This represents the probability of dental caries risk. This is a preliminary probability of dental caries. The degree of caries is graded. To balance the weighting coefficients of the initial probability of dental caries, To balance the weighting coefficients of the impact of caries depth grade, and , This represents the maximum value for the caries depth grade.

7. The method according to claim 5, characterized in that, The process of predicting periodontitis risk from the tooth region feature data to obtain a periodontitis risk score includes: Alveolar bone height features are extracted from the tooth region feature data to obtain an alveolar bone height feature vector from the alveolar ridge crest to the cementoenamel junction; the alveolar bone height feature vector is used to characterize the severity of alveolar bone resorption. Gingival margin morphology analysis is performed on the tooth region feature data to obtain gingival morphology feature vectors; these gingival morphology feature vectors are used to characterize the inflammatory state of the gingival tissue. Based on the alveolar bone height feature vector and the gingival morphology feature vector, the preliminary periodontitis risk index is obtained using the following formula: in, This serves as a preliminary risk index for periodontitis. For feature dimension, Indicates the first Alveolar bone height characteristics in several dimensions Indicates the first Gingival morphological features in 3 dimensions For the first The weights of alveolar bone height features in each dimension. For the first Weights of gingival morphological features in each dimension; Based on the preliminary periodontitis risk index, and combined with the periodontal pocket depth features extracted from the adjacent tooth region feature data, the corrected risk value is obtained using the following formula: in, This is the corrected risk value. This serves as a preliminary risk index for periodontitis. The risk correction factor for periodontal pocket depth. It is a feature of the actual periodontal pocket depth. The threshold for normal periodontal pocket depth; The corrected risk value is standardized to obtain the periodontitis risk score.

8. A dental health screening device based on image recognition, characterized in that, The device includes: The data acquisition and preprocessing module is used to perform image preprocessing on the acquired multi-source image data of teeth to obtain preprocessed tooth image data; the multi-source image data of teeth includes visible light tooth surface images, X-ray images of tooth roots and jawbones, and near-infrared caries detection images; The tooth feature extraction module is used to extract multi-scale tooth features from the preprocessed tooth image data to obtain tooth feature map data; the tooth feature map data is used to characterize the morphological and structural properties of teeth under different imaging modalities. The tooth feature fusion module is used to perform multi-feature fusion processing on the tooth feature map data to obtain fused tooth feature data. The tooth region segmentation module is used to segment and map feature regions based on the fused tooth feature data and a preset tooth anatomy template to obtain tooth region feature data; the tooth region feature data is used to characterize the features of each individual tooth or a specific tooth surface region. The dental health analysis module is used to analyze the health status of the tooth region feature data to obtain dental health status data; the dental health status data is used to characterize the potential risks of teeth in terms of caries, periodontitis and malocclusion. The teeth screening results module is used to generate teeth health screening results based on the teeth health status data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images