Intelligent periodontal disease classification method, device, medium, equipment and product

Abstract

The invention relates to the technical field of data processing, and particularly provides a periodontal disease intelligent classification method and device, a medium, equipment and a product, and the method can comprise the steps: obtaining image data and text data after preprocessing of user oral data; performing feature extraction on the image data and the text data to obtain an image feature vector and a text feature vector; fusing the image feature vector and the text feature vector to obtain a fused feature vector; based on a pre-trained classification model, classifying the fusion feature vector, and determining a classification result; wherein the classification result comprises a periodontal disease category grade to which the user belongs, and a periodontal feature matching degree and category grade confidence related to the periodontal disease category grade; the classification result is used for providing auxiliary suggestions for doctors. According to the embodiment of the invention, precise classification of periodontal diseases can be realized, and a reliable auxiliary reference basis is provided for doctors.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method, apparatus, medium, device, and product for intelligent classification of periodontal disease. Background Technology

[0003] Traditional periodontitis diagnosis primarily relies on visual inspection, probing, and X-ray examination by clinicians. Visual inspection mainly observes changes in the color, shape, and texture of the gingiva; probing is used to measure indicators such as periodontal pocket depth and attachment level; and X-ray examination helps dentists understand the extent of alveolar bone resorption. However, differences in clinical experience and techniques among dentists lead to a high degree of subjectivity in the final diagnosis of periodontitis. Moreover, providing diagnostic suggestions for periodontitis based solely on clinical experience cannot guarantee diagnostic accuracy.

[0004] Therefore, how to provide a technical solution for a highly accurate intelligent classification method for periodontal disease has become an urgent technical problem to be solved. Summary of the Invention

[0005] The purpose of some embodiments of this application is to provide a method, apparatus, medium, device and product for intelligent classification of periodontal disease. The technical solutions of the embodiments of this application can automatically classify periodontal disease with high accuracy, provide reliable auxiliary suggestions for doctors, and improve diagnostic efficiency.

[0006] In a first aspect, some embodiments of this application provide a method for intelligent classification of periodontal disease, comprising: acquiring image data and text data after preprocessing user oral data; extracting features from the image data and the text data respectively to obtain image feature vectors and text feature vectors; fusing the image feature vectors and the text feature vectors to obtain a fused feature vector; classifying the fused feature vector based on a pre-trained classification model to determine a classification result; wherein the classification result includes the periodontal disease category level to which the user belongs, as well as the periodontal feature matching degree and category level confidence related to the periodontal disease category level; the classification result is used to provide auxiliary suggestions for doctors.

[0007] Some embodiments of this application extract features from image data and text data separately, then fuse them to obtain a fused feature vector; finally, a classification model is used to classify the fused feature vector to obtain a classification result. Embodiments of this application can automatically classify periodontal disease with high accuracy, providing reliable auxiliary suggestions for doctors and improving diagnostic efficiency.

[0008] In some embodiments, obtaining the preprocessed image data and text data of the user's oral cavity data includes: performing image enhancement and noise reduction processing on the original image data in the user's oral cavity data to obtain the image data; and performing data cleaning and normalization processing on the original text data in the user's oral cavity data to obtain the text data.

[0009] Some embodiments of this application obtain image data and text data by performing corresponding preprocessing on the original image data and original text data, providing high-quality data support for subsequent data analysis.

[0010] In some embodiments, the step of extracting features from the image data and the text data to obtain image feature vectors and text feature vectors respectively includes: extracting features from the image data using a pre-trained image feature extraction model to obtain the image feature vector; and extracting features from the text data using a pre-trained text feature extraction model to obtain the text feature vector.

[0011] Some embodiments of this application use different models to extract features from image data and text data respectively, and obtain corresponding feature vectors, which can achieve efficient feature extraction processing.

[0012] In some embodiments, the step of using a pre-trained image feature extraction model to extract features from the image data and obtain the image feature vector includes: using the first layer structure of the image feature extraction model to extract periodontal region features from the image data; and using the second layer structure of the image feature extraction model to extract periodontal lesion features from the periodontal region features to obtain the image feature vector.

[0013] Some embodiments of this application extract features from image data using the first and second layer structures in the image feature extraction model to obtain image feature vectors, thereby achieving effective extraction of image features.

[0014] In some embodiments, the step of using a pre-trained text feature extraction model to extract features from the text data and obtain the text feature vector includes: using the text feature extraction model to process the structured text and unstructured text in the text data separately to obtain structured features and unstructured features; and concatenating the structured features and the unstructured features to obtain the text feature vector.

[0015] Some embodiments of this application process structured and unstructured text using a text feature extraction model to obtain text feature vectors, thereby achieving effective extraction of text features.

[0016] In some embodiments, fusing the image feature vector and the text feature vector to obtain a fused feature vector includes: aligning the dimensions of the image feature vector and the text feature vector, and obtaining the attention weights corresponding to the image feature vector and the text feature vector respectively; and performing a fusion calculation on the image feature vector and the text feature vector based on the attention weights to obtain the fused feature vector.

[0017] Some embodiments of this application fuse image feature vectors and text feature vectors to provide effective data for subsequent model processing, thereby achieving intelligent classification of periodontal disease.

[0018] In some embodiments, classifying the fused feature vector based on a pre-trained classification model and determining the classification result includes: inputting the fused feature vector into the classification model to obtain the number of votes for each category level; determining the function value for each category level based on a decision function; selecting the category level and associated information corresponding to the highest number of votes for each category level and the function value being greater than a preset value as the classification result; wherein the associated information includes the periodontal feature matching degree and the category level confidence degree.

[0019] Some embodiments of this application classify the fused feature vectors using a classification model to obtain classification results, thereby achieving automatic classification of periodontal disease categories with high accuracy.

[0020] In some embodiments, after outputting the classification results, the method further includes: labeling the image data and text data to generate a feature importance heatmap.

[0021] Some embodiments of this application achieve visual display by generating feature importance heatmaps for relevant personnel to view.

[0022] Secondly, some embodiments of this application provide an apparatus for intelligent classification of periodontal disease, comprising: an acquisition module for acquiring image data and text data after preprocessing of user oral data; an extraction module for extracting features from the image data and the text data respectively to obtain image feature vectors and text feature vectors; a fusion module for fusing the image feature vectors and the text feature vectors to obtain a fused feature vector; and a classification module for classifying the fused feature vector based on a pre-trained classification model to determine a classification result; wherein the classification result includes the periodontal disease category level to which the user belongs, as well as the periodontal feature matching degree and category level confidence related to the periodontal disease category level; and the classification result is used to provide auxiliary suggestions to doctors.

[0023] Thirdly, some embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the method described in any embodiment of the first aspect.

[0024] Fourthly, some embodiments of this application provide an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, can implement the method as described in any embodiment of the first aspect.

[0025] Fifthly, some embodiments of this application provide a computer program product, the computer program product including a computer program, wherein the computer program, when executed by a processor, can implement the method described in any embodiment of the first aspect. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of some embodiments of this application, the accompanying drawings used in some embodiments of this application will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A system diagram of intelligent periodontal disease classification provided for some embodiments of this application; Figure 2 One of the flowcharts for a method of intelligent classification of periodontal disease provided for some embodiments of this application; Figure 3 Flowchart 2 of the method for intelligent classification of periodontal disease provided for some embodiments of this application; Figure 4 An architecture diagram of intelligent periodontal disease classification provided for some embodiments of this application; Figure 5 Block diagrams of a device for intelligent classification of periodontal disease provided for some embodiments of this application; Figure 6 A schematic diagram of an electronic device provided for some embodiments of this application. Detailed Implementation

[0028] The technical solutions of some embodiments of this application will now be described with reference to the accompanying drawings.

[0029] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0030] In related technologies, the subjective nature of periodontal disease diagnosis is largely due to differences in clinical experience and techniques among dentists. For example, when measuring periodontal pocket depth, different dentists may use different probing pressures and angles, leading to varying measurement results and affecting the accuracy of the diagnosis. Furthermore, traditional examination methods struggle to comprehensively and accurately assess the severity and extent of periodontitis, easily missing early or occult periodontal lesions. In addition, traditional diagnostic methods lack standardized judgment procedures and unified quantitative indicators, making it difficult to guarantee the accuracy and reliability of diagnoses in complex cases. Moreover, with the introduction of new classification and diagnostic standards for periodontal disease, higher demands are placed on the precision and standardization of diagnosis, rendering traditional diagnostic methods inadequate for clinical and teaching needs.

[0031] In view of this, some embodiments of this application provide a method for intelligent classification of periodontal disease. This method can process and then fuse image data and text data from the user's oral cavity to obtain a fused feature vector. A trained classification model is then used to process the fused feature vector to obtain the classification result of periodontal disease. The embodiments of this application can quickly and accurately provide diagnostic classification results for periodontitis, providing clinicians with objective and accurate diagnostic reference suggestions, improving diagnostic efficiency and accuracy. It can also provide standardized and visualized teaching tools for oral medicine education, promoting the standardization and intelligentization of periodontal disease diagnosis, and has significant clinical and educational value.

[0032] The following is in conjunction with the appendix Figure 1 The overall structure of a system for intelligent classification of periodontal disease provided by some embodiments of this application is illustrated by way of example.

[0033] like Figure 1 As shown in the diagram, some embodiments of this application provide a system diagram for intelligent periodontal disease classification. This intelligent periodontal disease classification system may include a terminal 100 and a server 200. The terminal 100 can send raw image data and raw text data of the user's oral cavity to the server 200. The server 200 can perform a series of operations on the raw image data and raw text data, including preprocessing, feature extraction, and fusion, to obtain a fused feature vector. Finally, the fused feature vector is input into a pre-deployed classification model to output the user's periodontal disease classification result.

[0034] Understandably, the classification model is pre-trained and deployed on server 200, and its training process can also be executed by server 200.

[0035] In some embodiments of this application, terminal 100 can be a mobile terminal or a non-portable computer terminal; no specific limitation is made here. Alternatively, server 200 can be connected to an image acquisition device to directly obtain the user's raw image data from the image acquisition device. Raw text data can be obtained by the user inputting it on terminal 100. The specific choice can be made according to the actual application scenario.

[0036] The following is in conjunction with the appendix Figure 2 The present application provides an exemplary embodiment of the implementation process of intelligent periodontal disease classification performed by server 200.

[0037] Please see the appendix Figure 2 , Figure 2 A flowchart of a method for intelligent classification of periodontal disease is provided for some embodiments of this application. The method for intelligent classification of periodontal disease may include: S210: Acquire image and text data of the user's oral cavity after preprocessing.

[0038] For example, in a specific embodiment of this application, the server 200 can preprocess the user's oral cavity data to obtain image data and text data for subsequent classification. The image data may include CT scan data and / or image data captured by a medical camera; however, this embodiment does not specifically limit the scope of the data.

[0039] In some embodiments of this application, S210 may include: performing image enhancement and noise reduction processing on the original image data in the user's oral cavity data to obtain the image data; and performing data cleaning and normalization processing on the original text data in the user's oral cavity data to obtain the text data.

[0040] For example, in a specific embodiment of this application, the user can input their original image data on terminal 100. This original image data may include images acquired by a scanning device (such as a CT scanner), images captured by a camera, or both of the aforementioned types of data, such as X-ray films, periodontal probing data, and oral CT images. Simultaneously, the user can input their original text data on terminal 100, such as descriptions of the patient's symptoms (e.g., frequency of gingival bleeding, extent of gingival swelling, degree of halitosis, and tooth loosening), and examination results. Afterward, server 200, upon receiving the original image data, uses an image enhancement algorithm to adjust parameters such as brightness, contrast, and color balance in the original image data to enhance the characteristic information of periodontal tissues in the image, making it easier to identify and analyze. A noise reduction algorithm is used to remove noise interference from the image, improving image clarity and quality; the image data is obtained after these operations. For the original text data, natural language processing technology is used for data cleaning and normalization to remove invalid information and erroneous data, converting the text data into a format that the server can understand and process. Meanwhile, the terminal 100 or server 200 also has a data verification function, which can check the input raw image data and raw text data in real time. Once problems such as data format errors or values ​​exceeding the normal range are found, the user will be immediately prompted to make corrections to ensure the accuracy and completeness of the entered data.

[0041] S220, feature extraction is performed on the image data and the text data respectively to obtain image feature vectors and text feature vectors.

[0042] For example, in a specific embodiment of this application, the server 200 is provided with an intelligent analysis module, which can perform in-depth mining and analysis on image data and text data based on deep learning algorithms, extract key features, and obtain image feature vectors and text feature vectors.

[0043] In some embodiments of this application, S220 may include: S221, Use a pre-trained image feature extraction model to extract features from the image data and obtain the image feature vector.

[0044] For example, in a specific embodiment of this application, the basic network architecture of the image feature extraction model uses a lightweight improved ResNet-50 as the backbone network, and adds a Periodontal Region Attention Module (PRAM) after the traditional convolutional layers. By inputting preprocessed oral CT (0.5 mm slice thickness, after 3D reconstruction) and periapical radiographs (normalized to 512×512 pixels) (as a specific example of image data) into the image feature extraction model, the image feature vector is output.

[0045] In some embodiments of this application, S221 may include: extracting the image data using the first layer structure of the image feature extraction model to obtain periodontal region features; and extracting periodontal lesion features from the periodontal region features using the second layer structure of the image feature extraction model to obtain the image feature vector.

[0046] For example, in a specific embodiment of this application, the image feature extraction model extracts image data in two stages. The first stage's first layer structure includes three convolutional blocks and a PRAM. The three convolutional blocks extract periodontal edge and texture features from the image data, such as alveolar bone margins and grayscale changes in the periodontal ligament space. The PRAM's spatial attention mechanism automatically locates key regions such as the alveolar ridge crest, root bifurcation, and implant-osseointegration interface (as a specific example of periodontal region features). These key regions have a weight value more than 30% higher than non-target regions, serving as the localization standard. The second stage's second layer structure includes embedding a multi-scale receptive field module within the residual block, using 3×3, 5×5, and 7×7 convolutional kernels. These kernels extract periodontal lesion features of different sizes within the key regions (e.g., small-scale bone resorption <2mm, large-scale bone pockets >5mm, and peri-implant radiolucency areas), outputting a 1024-dimensional image feature vector containing quantitative features such as bone height, bone density, and defect type.

[0047] It should be understood that the aforementioned convolutional block (kernel) operations, through a "sliding window + weighted summation" approach, transform locally correlated features in oral imaging data (such as gray-level abrupt changes at bone edges and texture variations in periodontal pockets) into quantifiable feature maps. This is the foundation for convolutional networks to extract effective information from the original image. In the second-stage periodontal disease diagnosis module, customized convolutional kernel design (such as edge detection kernels) and multi-scale operations can accurately capture disease-related image features, providing reliable input for subsequent classification. Convolutional neural networks automatically learn local and global features in images through a combination of multiple convolutional and pooling layers. In the first-stage convolutional layer, convolutional operations are performed between the convolutional kernel and the image to extract features such as edges, textures, and shapes. In the second-stage pooling layer, the feature maps output by the convolutional layer are downsampled to reduce the amount of data while retaining important feature information. After multiple convolutional and pooling operations, a high-level feature representation of the image, i.e., the image feature vector, is obtained.

[0048] S222, Use a pre-trained text feature extraction model to extract features from the text data and obtain the text feature vector.

[0049] For example, in a specific embodiment of this application, the text feature extraction model can process the structured text and unstructured text in the text data separately, and finally output the text feature vector. The structured text in the text data may include probing depth (PD), clinical attachment level (CAL), bleeding index (BI), implant mobility, patient smoking history, etc., while the unstructured text is similar to the content of the doctor's complaint, such as "recurrent gingival swelling in the past 3 months".

[0050] In some embodiments of this application, S222 may include: processing the structured text and unstructured text in the text data separately using the text feature extraction model to obtain structured features and unstructured features; concatenating the structured features and the unstructured features to obtain the text feature vector.

[0051] For example, in a specific embodiment of this application, continuous values ​​such as PD (unit: mm) and CAL (unit: mm) are transformed into structured classification features (as a specific example of structured features) through the clinical indicator mapping layer in the text feature extraction model. For instance, PD ≥ 6 mm is mapped to "severe periodontal pocket," corresponding to the feature of "stage III periodontitis" in the international classification. The BERT fine-tuning model in the text feature extraction model is used to extract semantic features, focusing on identifying keywords such as "subosseous pocket," "peri-implant suppuration," and "insufficient keratinized gingiva," which are transformed into a 256-dimensional semantic vector (as a specific example of unstructured features). This vector is then concatenated with the structured classification features to form a 512-dimensional text feature vector.

[0052] S230, the image feature vector and the text feature vector are fused to obtain a fused feature vector.

[0053] For example, in a specific embodiment of this application, since image features and text features are heterogeneous, it is necessary to enhance the classification correlation through fusion; that is, to fuse image feature vectors and text feature vectors to obtain fused feature vectors for subsequent analysis.

[0054] In some embodiments of this application, S230 may include: aligning the dimensions of the image feature vector and the text feature vector, and obtaining the attention weights corresponding to the image feature vector and the text feature vector respectively; and performing a fusion calculation on the image feature vector and the text feature vector based on the attention weights to obtain the fused feature vector.

[0055] For example, in a specific embodiment of this application, a dynamic weight allocation mechanism is used to unify the dimensions of the 1024-dimensional image feature vector and the 512-dimensional text feature vector output by the CNN (e.g., mapping to 768 dimensions through a fully connected layer). Attention weights are assigned according to feature importance; for example, the weight ratio of "bone resorption depth" in the image to "PD value" in the text is set to 1.2:1, as they are more correlated in classification. For example, attention-based cross-fusion is used to calculate the cosine similarity between image features and text features, assigning higher attention weights to high-similarity feature pairs (e.g., "CT shows subosseous pocket" and "PD=8mm"). For example, the image attention weight is α_test and the text attention weight is β_test. Then, attention-weighted fusion is performed to obtain an initial fusion vector. A trained L1 regularization parameter λ=0.01 is used to remove redundant features from the initial fusion vector, retaining feature combinations strongly correlated with periodontal stage (e.g., "bone resorption > 5mm + PD ≥ 6mm"), and mapping them to a high-dimensional space through an RBF kernel to obtain the fusion feature vector.

[0056] S240, Based on a pre-trained classification model, the fused feature vector is classified to determine the classification result; wherein, the classification result includes the periodontal disease category level to which the user belongs, as well as the periodontal feature matching degree and category level confidence related to the periodontal disease category level; the classification result is used to provide auxiliary suggestions for doctors.

[0057] For example, in a specific embodiment of this application, the classification model is based on an enhanced SVM design, which overcomes the limitations of the fixed kernel function in traditional SVMs and dynamically adjusts the classification boundary in conjunction with the international classification standards for periodontal disease. By inputting the fused feature vector into the trained classification model, the classification result can be output. The output format of the classification result is as follows: Basic diagnosis: such as "stage III chronic periodontitis" (as a specific example of a periodontal disease classification level); Key evidence: such as, “Bone resorption depth of 4.2mm matches the Stage III template 92%, PD=6mm matches 88%” (as a specific example of periodontal feature matching); Classification confidence (as a specific example of category-level confidence): e.g., 92% (calculated based on feature matching degree). Classification confidence can range from 0 to 100, and a value greater than the set threshold of 70% is considered high confidence.

[0058] In some embodiments of this application, S240 may include: inputting the fused feature vector into the classification model to obtain the number of votes for each category level; determining the function value for each category level based on the decision function; selecting the category level and associated information corresponding to the highest number of votes for each category level and the function value being greater than a preset value as the classification result; wherein, the associated information includes the periodontal feature matching degree and the category level confidence degree.

[0059] For example, in a specific embodiment of this application, the fused feature vector is processed based on 28 binary SVM sub-models constructed on a "one-to-one" basis, and the number of votes obtained for each category is counted. The optimized fused feature vector V_fuse_test is compared with the phased feature template library constructed during the training of the classification model (the phased feature template library contains typical feature vectors of 12 international classifications in 4 phases and 3 levels) to obtain the function value of each category. Specifically, the function value is calculated using the SVM decision function, and the calculation formula is: f_k = sign(w_k·K(V_fuse_test) + b_k) (w_k and b_k are the trained hyperplane parameters). In this way, the function value between the fused feature vector and each of the typical feature vectors of the 12 international classifications can be obtained. The category with the highest number of votes and the decision function value f_k>0 (as a specific example of the preset value) is selected as the preliminary result. If there is a tie (such as the difference in the number of votes between the two categories ≤ 1), the minimum Euclidean distance d_k is used for further determination. Specifically, the Euclidean distance between V_fuse_test and the typical feature vectors of the eight international classifications is calculated, i.e., Euclidean distance d_k = ||V_fuse_test - Template_k|| (k=1...12, corresponding to the 12 classifications). The category with the smallest Euclidean distance is selected as the classification result of periodontal disease (as a specific example of periodontal disease classification level).

[0060] The process of obtaining the classification model is briefly described below.

[0061] After obtaining the fused feature vector samples for training, L1 regularized PCA is applied to the 768-dimensional fused feature vector samples to retain 95% of the information while removing redundant features (such as tooth morphology features irrelevant to classification), reducing the dimensionality to 256 dimensions and improving the SVM computational efficiency. The SVM kernel function is adaptively selected. For example, the RBF kernel function (suitable for nonlinear boundaries) is used for continuous spectrum classification of "periodontal health-gingivitis-periodontitis"; for discrete classification of peri-implant diseases (such as "health-mucositis-periodontitis"), a multinomial kernel function is automatically switched (because there are explicit thresholds in the classification criteria, such as "peri-implant bone resorption > 2 mm" for periodontitis). SVM parameters are dynamically optimized: the penalty coefficient C and kernel parameter γ of the SVM are adjusted in real time using a Bayesian optimization algorithm, and the parameters are updated every 50 iterations in the training set to ensure that the classification boundary conforms to international standards (e.g., strongly correlated with the classification threshold of "periodontitis stage IV" and "bone resorption ≥ 5 mm with tooth loosening"). By using a classification result calibration module, the SVM model output is transformed into diagnostic results that conform to international standards, enhancing its clinical applicability. Specifically, the classification probabilities of the SVM output (e.g., 85% probability of "Stage III Periodontitis") are mapped to quantitative indicators of international classification (e.g., 3-4mm bone resorption with mild tooth mobility) to generate a "feature-diagnosis" correspondence table (i.e., a staging feature template library). A clinical rule base is introduced to perform secondary validation on the model output; for example, if the model diagnoses "peri-implant mucositis," and the image shows "1mm bone resorption" with no "suppuration" in the text, it is considered valid; if "3mm bone resorption" is present, a warning is triggered, indicating a possible misdiagnosis as "periodontitis." Through the above training and validation, a classification model for practical application is obtained.

[0062] In some embodiments of this application, after performing S240, the method for intelligent classification of periodontal disease further includes: labeling the image data and text data to generate a feature importance heatmap.

[0063] For example, in a specific embodiment of this application, Grad-CAM++ is used to visualize the CNN region of interest, marking the regions in the image data that contribute the most to the classification results (such as the edge of bone resorption, low-density areas around implants), generating a feature importance heatmap, and displaying the weight ratio of text features (such as "smoking history") and image features (such as "bone density") in the classification results in the feature importance heatmap, to help doctors understand the model's decision.

[0064] It should be noted that the image feature extraction model and text feature extraction model used in the above implementation process are all trained. The training process is the same as the processing process described above, and will not be repeated here.

[0065] The following is in conjunction with the appendix Figure 3The present application provides an exemplary description of the specific process of intelligent periodontal disease classification provided by some embodiments.

[0066] Please see the appendix Figure 3 , Figure 3 A flowchart of a method for intelligent classification of periodontal disease provided for some embodiments of this application.

[0067] The above process is illustrated below by example.

[0068] S310 performs image enhancement and noise reduction processing on the raw image data in the user's oral cavity data to obtain image data.

[0069] For example, for the input patient's (i.e., user's) oral CT (0.5mm slice thickness) and periapical radiograph (i.e., raw image data), the grayscale contrast of alveolar bone and periodontal ligament is first enhanced by an adaptive histogram equalization algorithm (e.g., brightness adjustment range ±20%, contrast gain coefficient 1.5). Then, a nonlocal mean denoising algorithm (window size 7×7, similarity threshold 0.1) is used to remove CT artifacts and periapical radiograph noise. Finally, it is normalized to a single-channel grayscale image of 512×512 pixels (as a specific example of image data).

[0070] S320 performs data cleaning and normalization on the raw text data in the user's oral cavity data to obtain the text data.

[0071] For example, for patient clinical data, such as "probing depth 4mm, gingival bleeding (+)" and "bone resorption of about 2mm in the past year", structured indicators are extracted using natural language processing tools (such as the spaCy word segmentation model): continuous values ​​(PD, CAL, etc.) are retained to one decimal place; categorical values ​​(such as bleeding index BI) are mapped to quantitative labels (0 = no bleeding, 1 = mild, 2 = severe); unstructured descriptions, such as "peri-implant suppuration", are converted into a 256-dimensional semantic vector; finally, the structured indicators are concatenated to form 512-dimensional initial text data.

[0072] S330 uses a pre-trained image feature extraction model to extract features from image data and obtain image feature vectors.

[0073] For example, low-level features (edges, textures) are extracted by three convolutional blocks of the improved ResNet-50 backbone network (i.e., image feature extraction model). Key regions such as alveolar crest (weight increased by 30%) and implant-bone interface are automatically located through the PRAM module to generate a preliminary feature map of 256×256×256. High-level features are extracted by multi-scale residual blocks (3×3 / 5×5 / 7×7 convolutional kernels) to output a 1024-dimensional image feature vector V_img_test (containing quantitative parameters such as bone resorption depth and bone density standard deviation).

[0074] S340 uses a pre-trained text feature extraction model to extract features from text data and obtain text feature vectors.

[0075] For example, using the same clinical indicator mapping layer and BERT fine-tuning model (i.e., text feature extraction model) as the training phase, the preprocessed text data is transformed into a 512-dimensional text feature vector V_txt_test (containing hierarchical features of PD and CAL and semantic features).

[0076] S350 fuses image feature vectors and text feature vectors to obtain a fused feature vector.

[0077] For example, based on the mean μ and standard deviation σ of the pre-constructed training set, V_img_test and V_txt_test are Z-score standardized. Then, the image feature vector is reduced to 512 dimensions (V_img_512_test) using a pre-trained 1×1 convolutional kernel W1 and a fully connected layer W2, while the text features remain 512 dimensions (V_txt_512_test). The cosine similarity matrix template optimized during the training phase is called to calculate the correlation Sim_test between image features and text features, generating image attention weights α_test and text attention weights β_test. Finally, a weighted sum is performed, and redundant features are removed using a pre-trained L1 regularization parameter λ=0.01, retaining feature combinations strongly correlated with periodontal stage (such as "bone resorption > 5mm + PD ≥ 6mm"), and mapped to a high-dimensional space using an RBF kernel to obtain the fused feature vector.

[0078] S360, based on a pre-trained classification model, classifies fused feature vectors and determines the classification result.

[0079] It is understood that the specific implementation process of S310~S360 can refer to the method embodiment provided above. To avoid repetition, detailed descriptions are omitted here. In addition, S310~320 and 330~340 can be executed synchronously or separately, and the execution order is not limited to this.

[0080] In addition, some embodiments of this application also provide, as Figure 4 The diagram shown illustrates the architecture of the intelligent periodontal disease classification system. This architecture includes a user input module, an intelligent analysis module, a diagnostic output module, a teaching extension module, and a clinical assistance module. Each module operates independently yet collaboratively to ensure efficient and stable software operation. The functions of each module are illustrated below.

[0081] The user input module serves as the entry point for software interaction between the terminal 100 and the user. It features a simple and intuitive interface, supports multiple data entry methods, and caters to different user habits. Users can manually input detailed patient symptoms, such as frequency of gingival bleeding, extent of gingival swelling, degree of halitosis, and tooth mobility. They can also directly import digital examination results, including X-ray images, periodontal probing data, and oral CT scan data. Furthermore, this module has a data verification function, capable of real-time checking of input data. If errors in data format or values ​​exceeding normal ranges are detected, the module immediately prompts the user for correction, ensuring the accuracy and completeness of the entered data. The intelligent analysis module, the core of the server, integrates advanced artificial intelligence algorithms and big data analytics. This module first preprocesses the input patient oral data, optimizing the image data using image enhancement and noise reduction techniques to improve image clarity and quality. It then processes the text data using data cleaning and normalization methods to meet analytical requirements. Next, based on deep learning algorithms, it performs in-depth mining and analysis of the preprocessed data, automatically identifying and extracting key features of periodontal tissues, such as changes in alveolar bone density, periodontal pocket depth distribution, and the degree of gingival recession. Finally, it compares the extracted features with a diagnostic model generated from big data analysis to comprehensively assess the patient's periodontal health. The diagnostic output module presents the results from the intelligent analysis module to users in a clear and easy-to-understand manner. For clinicians, the output diagnostic report is detailed and professional, including the staging and grading results of periodontitis, such as mild periodontitis (Stage I, Grade A), moderate periodontitis (Stage II, Grade B), and severe periodontitis (Stage III or IV, Grade C). It also provides specific diagnostic criteria and detailed condition analysis, such as "Based on the alveolar bone resorption reaching 2 / 3 and the periodontal pocket depth generally greater than 6mm, it is judged to be severe periodontitis (Stage III, Grade C)," providing strong support for doctors to formulate treatment plans. For teaching scenarios, the output content focuses more on knowledge explanation and case analysis, explaining the standards and basis for periodontitis staging and grading in a vivid and engaging way, and guiding students to understand and master relevant knowledge through specific cases. The teaching extension module provides rich teaching resources and diverse teaching tools for oral medicine education. Regarding teaching resources, the module includes numerous periodontal disease teaching cases. Each case contains detailed patient information, diagnostic process, treatment plan, and treatment effect follow-up, facilitating case-based teaching and allowing students to deepen their understanding of periodontal disease through real-world examples. It also provides professional periodontal disease teaching videos covering all aspects of periodontal disease, including etiology, pathology, diagnostic methods, and treatment techniques, explained and demonstrated by experienced oral medicine experts, providing students with intuitive and comprehensive learning materials. In terms of teaching tools, the module supports interactive classroom teaching, such as posing questions, organizing discussions, and conducting group learning, thereby enhancing students' learning enthusiasm and participation. The clinical support module closely integrates with actual clinical needs, providing doctors with comprehensive support. On one hand, it recommends personalized treatment plans based on diagnostic results, referencing authoritative domestic and international periodontal disease treatment guidelines and clinical experience, and considering factors such as the patient's specific condition, physical status, age, and lifestyle habits. It provides detailed treatment suggestions, including the selection of treatment methods (such as supragingival scaling, subgingival scaling, periodontal surgery), recommendations for medications (such as the type and dosage of antibiotics and anti-inflammatory drugs), and planning of the treatment cycle. On the other hand, this module also has patient management functions, comprehensively recording and managing patients' basic information, diagnostic records, treatment processes, and follow-up information. This allows doctors to easily access and track changes in the patient's condition, providing a reference for subsequent treatment. Furthermore, it automatically reminds doctors to conduct follow-up and re-examinations based on the patient's treatment progress and recovery status, ensuring that patients receive timely and effective treatment.

[0082] Please refer to Figure 5 , Figure 5 The diagram shows a block diagram of a device for intelligent periodontal disease classification provided in some embodiments of this application. It should be understood that this device corresponds to the method embodiments described above and is capable of performing the various steps involved in the method embodiments. The specific functions of this device can be found in the description above; detailed descriptions are omitted here to avoid repetition.

[0083] Figure 5The device for intelligent periodontal disease classification includes at least one software functional module that can be stored in a memory or embedded in the device in the form of software or firmware. The device includes: an acquisition module 510 for acquiring preprocessed image data and text data of the user's oral cavity; an extraction module 520 for extracting features from the image data and text data respectively to obtain image feature vectors and text feature vectors; a fusion module 530 for fusing the image feature vectors and text feature vectors to obtain a fused feature vector; and a classification module 540 for classifying the fused feature vector based on a pre-trained classification model to determine the classification result. The classification result includes the user's periodontal disease category level, the periodontal feature matching degree related to the periodontal disease category level, and the category level confidence. The classification result is used to provide auxiliary suggestions to doctors.

[0084] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the aforementioned method, and will not be elaborated further here.

[0085] Some embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can perform the operation of any of the methods corresponding to the methods provided in the above embodiments.

[0086] Some embodiments of this application also provide a computer program product, which includes a computer program, wherein when the computer program is executed by a processor, it can implement the operation of any of the methods corresponding to the above embodiments provided in the above embodiments.

[0087] like Figure 6 As shown, some embodiments of this application provide an electronic device 600, which includes a memory 610, a processor 620, and a computer program stored in the memory 610 and executable on the processor 620. When the processor 620 reads the program from the memory 610 via a bus 630 and executes the program, it can implement the methods of any of the above embodiments.

[0088] Processor 620 can process digital signals and can include various computing architectures. For example, it can be a complex instruction set computer architecture, a reduced instruction set computer architecture, or an architecture that implements multiple instruction set combinations. In some examples, processor 620 can be a microprocessor.

[0089] The memory 610 can be used to store instructions executed by the processor 620 or data related to the execution of instructions. These instructions and / or data may include code for implementing some or all of the functions of one or more modules described in the embodiments of this application. The processor 620 of this disclosure embodiment can be used to execute the instructions in the memory 610 to implement the methods shown above. The memory 610 includes dynamic random access memory, static random access memory, flash memory, optical memory, or other memories well known to those skilled in the art.

[0090] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for intelligent classification of periodontal disease, characterized in that, include: Acquire preprocessed image and text data of the user's oral cavity; Feature extraction is performed on the image data and the text data respectively to obtain image feature vectors and text feature vectors; The image feature vector and the text feature vector are fused to obtain a fused feature vector; Based on a pre-trained classification model, the fused feature vector is classified to determine the classification result; wherein, the classification result includes the periodontal disease category level to which the user belongs, as well as the periodontal feature matching degree and category level confidence related to the periodontal disease category level; the classification result is used to provide auxiliary suggestions for doctors.

2. The method as described in claim 1, characterized in that, The acquisition of image data and text data after preprocessing of user oral cavity data includes: The original image data in the user's oral cavity data is subjected to image enhancement and noise reduction processing to obtain the image data; The original text data in the user's oral cavity data is cleaned and normalized to obtain the text data.

3. The method as described in claim 1 or 2, characterized in that, The step of extracting features from the image data and the text data respectively to obtain image feature vectors and text feature vectors includes: The image data is used to extract features using a pre-trained image feature extraction model to obtain the image feature vector; The text data is used to extract features using a pre-trained text feature extraction model to obtain the text feature vector.

4. The method as described in claim 3, characterized in that, The step of using a pre-trained image feature extraction model to extract features from the image data and obtain the image feature vector includes: The image data is extracted using the first layer structure of the image feature extraction model to obtain periodontal region features; The image feature vector is obtained by extracting periodontal lesion features from the periodontal region features using the second layer structure of the image feature extraction model.

5. The method as described in claim 3, characterized in that, The step of extracting features from the text data using a pre-trained text feature extraction model to obtain the text feature vector includes: The text feature extraction model is used to process the structured and unstructured text in the text data separately to obtain structured and unstructured features. The structured features and the unstructured features are concatenated to obtain the text feature vector.

6. The method as described in claim 1 or 2, characterized in that, The step of fusing the image feature vector and the text feature vector to obtain a fused feature vector includes: Align the dimensions of the image feature vector and the text feature vector, and obtain the attention weights corresponding to the image feature vector and the text feature vector respectively; Based on the attention weights, the image feature vector and the text feature vector are fused to obtain the fused feature vector.

7. The method as described in claim 1 or 2, characterized in that, The pre-trained classification model classifies the fused feature vectors and determines the classification result, including: The fused feature vector is input into the classification model to obtain the number of votes for each category level; the function value for each category level is determined based on the decision function. The category level with the highest number of votes in each category level and whose function value is greater than a preset value, along with the associated information, is selected as the classification result; wherein, the associated information includes the periodontal feature matching degree and the category level confidence degree.

8. The method as described in claim 1 or 2, characterized in that, After outputting the classification result, the method further includes: The image and text data are labeled to generate a feature importance heatmap.

9. A device for intelligent classification of periodontal disease, characterized in that, include: The acquisition module is used to acquire the image data and text data of the user's oral cavity after preprocessing; The extraction module is used to extract features from the image data and the text data respectively, and obtain image feature vectors and text feature vectors; The fusion module is used to fuse the image feature vector and the text feature vector to obtain a fused feature vector; A classification module is used to classify the fused feature vector based on a pre-trained classification model and determine the classification result; wherein, the classification result includes the periodontal disease category level to which the user belongs, as well as the periodontal feature matching degree and category level confidence related to the periodontal disease category level; the classification result is used to provide auxiliary suggestions for doctors.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, performs the method as described in any one of claims 1-8.

11. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program is executed by the processor to perform the method as claimed in any one of claims 1-8.

12. A computer program product, characterized in that, The computer program product includes a computer program, wherein the computer program is executed by a processor to perform the method as described in any one of claims 1-8.

Images