An artificial intelligence-based periodontal status evaluation method
By collecting multimodal tooth data and performing cross-modal registration and feature extraction, the problem of difficulty in capturing early periodontal lesions in existing technologies has been solved, achieving high sensitivity and accurate assessment of periodontal status.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SECOND AFFILIATED HOSPITAL OF COLLEGE OF MEDICINEOF XIAN JIAOTONG UNIV
- Filing Date
- 2025-09-23
- Publication Date
- 2026-06-02
AI Technical Summary
Existing periodontal condition analysis methods are unable to capture early lesions, especially minute enamel demineralization or mild periodontitis, and doctors have limited ability to observe microscopic pathological mechanisms, affecting the comprehensiveness and accuracy of the analysis results.
Multimodal tooth data, including tooth surface images and local oral vascular network images, are collected. Multidimensional features of blood vessels and macroscopic morphological features of teeth are extracted through cross-modal registration and feature extraction modules, and cross-modal fusion is performed to output periodontal status assessment results.
It improves the sensitivity and accuracy of early periodontal disease assessment, enabling early warning and providing a basis for intervention, thus filling the gaps in single-modal assessment.
Smart Images
Figure CN121213498B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical artificial intelligence technology, specifically relating to an artificial intelligence-based method for periodontal status assessment. Background Technology
[0002] Periodontal health is a crucial component of oral health, and its assessment relies on the quantitative analysis of the structural characteristics and pathological changes of hard tooth tissues such as enamel and dentin, soft tissues such as gingiva and tooth structure, and surrounding bone tissue. Traditional methods of periodontal status analysis primarily depend on the dentist's experience, combined with imaging techniques such as intraoral endoscopy, X-rays, and CBCT, to observe tooth surface morphology, alveolar bone density, and soft tissue inflammation. However, with the increasing demands for oral health management, the limitations of traditional methods are becoming increasingly apparent: on the one hand, early periodontal lesions, such as micro-demineralization of enamel or mild periodontitis, are not morphologically significant, making it difficult to detect subtle changes through visual inspection or conventional imaging alone; on the other hand, dentists have limited ability to observe microscopic pathological mechanisms such as microvascular abnormalities and the spread of inflammatory factors, affecting the comprehensiveness and accuracy of the analysis results.
[0003] In recent years, the rapid development of artificial intelligence technology has provided new ideas for periodontal condition analysis. Deep learning-based image classification models have been widely used in dental image analysis, improving the efficiency of periodontal condition analysis by automatically extracting and classifying image features. However, existing neural network models mostly focus on tooth surface morphology or macroscopic structures, such as cavity size and alveolar bone height, resulting in insufficient sensitivity to early lesions. Summary of the Invention
[0004] To address the aforementioned problems in the existing technology, this invention provides an artificial intelligence-based method for periodontal status assessment. The technical problem to be solved by this invention is achieved through the following technical solution:
[0005] This invention provides an artificial intelligence-based method for periodontal status assessment, comprising:
[0006] Collect multimodal dental data, which includes images of the tooth surface and images of the vascular network in the oral cavity;
[0007] The multimodal tooth data is preprocessed and cross-modal registration is performed using a registration module to obtain registered tooth surface images and vascular network images.
[0008] The first feature extraction module extracts multi-dimensional features of the registered vascular network image, including spatial morphological features, topological connectivity features, and hemodynamic features of the blood vessels.
[0009] The macroscopic morphological features and local lesion features of the registered tooth surface image are extracted using the second feature extraction module.
[0010] The multi-dimensional features, macroscopic morphological features, local lesion features, and clinical data are fused across modalities to obtain fused features;
[0011] The periodontal status assessment results are output based on the fusion features.
[0012] In one embodiment of the present invention, the tooth surface image includes: an intraoral endoscopic image and a cone-beam computed tomography (CBCT) image, and the vascular network image includes: capillary images of the gingival sulcus and alveolar bone surface acquired using an optical coherence tomography (OCTA) device, and three-dimensional images of the vascular network in the pulp and periapical region acquired using a CT device.
[0013] In one embodiment of the present invention, the step of preprocessing and cross-modal registration of the multimodal tooth data to obtain registered tooth surface images and vascular network images includes:
[0014] Using the oral endoscopy image as the reference image, and CBCT image, capillary image, and 3D vascular network image as the images to be registered;
[0015] Key points representing the same anatomical structure are labeled in the reference image and the image to be registered, respectively. The image to be registered is mapped to the coordinate system of the reference image through affine transformation to obtain a coarse registration image.
[0016] The reference image and the coarsely registered image are stitched together and then input into the registration module. The nonlinear deformation field is obtained using a pre-trained VoxelMorph network. ;
[0017] Based on the nonlinear deformation field The coarsely registered image is aligned to the reference image to obtain the finely registered image.
[0018] In one embodiment of the present invention, the step of extracting multi-dimensional features of the registered vascular network image through a first feature extraction module, wherein the multi-dimensional features include spatial morphological features, topological connectivity features, and hemodynamic features of blood vessels, includes:
[0019] Based on the target blood vessels to be focused on in the periodontal status assessment, the precisely registered vascular network image is input into the first feature extraction module, so that the first feature extraction module can extract the morphological features, topological connectivity features, and hemodynamic features of the vascular network; wherein, the target blood vessel is a soft tissue blood vessel or a hard tissue associated blood vessel.
[0020] In one embodiment of the present invention, the first feature extraction module includes a first branch, a second branch, and a third branch;
[0021] The first branch is used to extract the three-dimensional morphological features of blood vessels from the finely registered vascular network image using a 3D-ResNet50 backbone network, and to extract the center line of the blood vessels using a skeletonization algorithm to calculate the distribution of blood vessel diameter, branch point density and blood vessel curvature, thereby obtaining spatial morphological features.
[0022] The second branch is used to take the vascular intersections in the finely registered vascular network image as nodes. Using the blood vessel segment as the boundary The vascular network is modeled as an undirected graph, and the node features and adjacency matrix of the undirected graph are further input into a graph convolutional network to generate a topological feature vector, thereby obtaining topological connectivity features.
[0023] The third branch is used to acquire capillary images at preset time intervals and input the resulting dynamic sequence images into a 2D ConvLSTM network to obtain hemodynamic features including blood flow velocity maps and perfusion maps.
[0024] In one embodiment of the present invention, the macroscopic morphological features include: tooth contour and crown morphological parameters;
[0025] The steps for extracting the macroscopic morphological features and local lesion features of the registered tooth surface image through the second feature extraction module include:
[0026] The finely registered CBCT images are converted to grayscale to obtain grayscale images;
[0027] Tooth contours are extracted based on the grayscale image, and crown morphology parameters are calculated based on the tooth contours.
[0028] Adaptive threshold segmentation is performed on the grayscale image to extract the high-brightness abnormal regions;
[0029] The lesion region is obtained by performing region growing on the highlighted abnormal region using the gray-level similarity criterion;
[0030] Texture features are extracted within the lesion area to obtain the local features of the lesion.
[0031] In one embodiment of the present invention, the step of cross-modal fusion of the multi-dimensional features, the macroscopic morphological features, the local lesion features, and clinical data to obtain fused features includes:
[0032] The multi-dimensional features, macroscopic morphological features, local lesion features, and clinical data are respectively subjected to layer normalization and linear projection, mapped to the same dimension, and then weighted by a gated fusion unit to obtain fused features.
[0033] In one embodiment of the present invention, the step of outputting periodontal status assessment results based on the fusion features includes:
[0034] The fusion features are input into the fully connected layer to obtain the corresponding periodontal condition level; wherein...
[0035] The periodontal condition levels include: healthy, mild inflammation, moderate periodontitis, and moderate periodontitis.
[0036] In one embodiment of the present invention, after the step of extracting multi-dimensional features of the registered vascular network image through the first feature extraction module, the method further includes:
[0037] Based on the topological feature vector, the key parameters of the vascular network are determined; the key parameters include: the variance of the node degree distribution, which reflects the uniformity of vascular branching. And the average path length that reflects the efficiency of blood flow pathways. ;
[0038] When the variance of the node degree distribution The average path length is greater than the first preset threshold and When the value exceeds the second preset threshold, the bone resorption volume of the corresponding region is further calculated:
[0039] ;
[0040] In the formula, Indicates the calibration coefficient;
[0041] When bone resorption volume When the value exceeds the third preset threshold, the area is marked as a high-risk bone resorption area.
[0042] In one embodiment of the present invention, the average path length This is the average of the shortest path lengths between all distinct node pairs in the vascular network;
[0043] Calculate the variance of the node degree distribution using the following steps. :
[0044] Traverse all nodes in the vascular network, calculate the degree of each node, and count the number of nodes corresponding to each degree;
[0045] Calculate the probability of the number of nodes corresponding to each degree relative to the total number of nodes, and construct the degree distribution probability;
[0046] Based on the degree distribution probability, calculate the expected value of the degree of all nodes. ;
[0047] Compare the degree of each node with the expected value. The variance of the node degree distribution is obtained by multiplying the square of the difference by the sum of the corresponding probabilities. .
[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0049] (1) This invention provides a periodontal condition assessment method based on artificial intelligence. The collected multimodal tooth data includes not only tooth surface images but also images of local vascular networks in the oral cavity. Abnormalities in the morphology and dynamics of local vascular networks in the oral cavity are often early biomarkers of pathological processes such as inflammation and necrosis. This invention uses a first feature extraction module to extract the spatial morphological features, topological connectivity features, and hemodynamic features of blood vessels, which makes up for the blind spot of using only a single modality for periodontal condition assessment in the prior art. Thus, it can effectively utilize the microvascular network information closely related to the lesion, and can capture early inflammation based on blood flow abnormalities, thereby improving the sensitivity to periodontal diseases.
[0050] (2) In the process of periodontal status assessment, this invention integrates the spatial morphological features, topological connectivity features and hemodynamic features of blood vessels, as well as the macroscopic morphological features of teeth and the local features of lesions. This multimodal data collaborative analysis method is conducive to improving the accuracy of periodontal status assessment and early warning capability.
[0051] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0052] Figure 1 This is a flowchart of an artificial intelligence-based periodontal status assessment method provided in an embodiment of the present invention;
[0053] Figure 2 This is a schematic diagram of the periapical vascular network provided in an embodiment of the present invention. Detailed Implementation
[0054] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0055] Figure 1 This is a flowchart of an artificial intelligence-based periodontal status assessment method provided in an embodiment of the present invention. Figure 1 As shown, this embodiment of the invention provides an artificial intelligence-based periodontal status assessment method, including:
[0056] S1. Acquire multimodal tooth data, which includes tooth surface images and images of the local vascular network in the oral cavity.
[0057] It should be understood that the local vascular network of the oral cavity serves as a channel for tooth nutrition and metabolism. Abnormalities in its morphology and dynamic characteristics are early biomarkers for pathological processes such as inflammation and necrosis. For example, the reduction of gingival capillary branches in early periodontitis and the decrease in pulp blood perfusion during the progression of dental caries. In order to improve the sensitivity to early lesions during oral condition assessment, this embodiment not only acquires images of the tooth surface but also images of the local vascular network of the oral cavity.
[0058] Optionally, the tooth surface images include: intraoral endoscopic images and cone-beam computed tomography (CBCT) images, and the vascular network images include: capillary images of the gingival sulcus and alveolar bone surface acquired using optical coherence tomography (OCTA) equipment, and three-dimensional images of the vascular network in the pulp and periapical region acquired using CT equipment.
[0059] S2. Preprocess the multimodal tooth data and use the registration module to perform cross-modal registration to obtain the registered tooth surface image and vascular network image.
[0060] Specifically, the multimodal dental data is first preprocessed, including denoising. Then, using the intraoral endoscopic image as the reference image and CBCT images, capillary images, and 3D images of vascular networks as the images to be registered, the first stage performs coarse registration: key points representing the same anatomical structure, including cusps, gingival margins, and root apex points, are labeled in both the reference and the images to be registered. The coordinate system of the image to be registered is mapped to that of the reference image through affine transformation, resulting in a coarsely registered image. Next, the second stage performs fine registration: the reference image and the coarsely registered image are stitched together and input into the registration module. A pre-trained VoxelMorph network is used to obtain the nonlinear deformation field. Nonlinear deformation field It describes the displacement vector of each voxel, i.e., the mapping relationship from the coarse registration image to the reference image. Therefore, based on the nonlinear deformation field... It can align coarsely registered images to a reference image to obtain finely registered images, including finely registered CBCT images, capillary images, and 3D images of vascular networks.
[0061] S3. Extract multi-dimensional features of the registered vascular network image through the first feature extraction module. The multi-dimensional features include the spatial morphological features, topological connectivity features, and hemodynamic features of the blood vessels.
[0062] In step S3, the finely registered vascular network image is input into the first feature extraction module according to the target blood vessels that need to be focused on in the periodontal status assessment. The target blood vessels are soft tissue blood vessels or hard tissue associated blood vessels. It should be noted that the target blood vessel can be determined based on clinical data such as the user's gender, age, and symptoms. For example, if the user has gingivitis, their gingival capillaries may be dilated. Soft tissue blood vessels distributed in the soft tissue layers such as the gingiva, periodontal ligament, and oral mucosa can be selected as target blood vessels. Correspondingly, inputting the precisely registered capillary images of the gingival sulcus and alveolar bone surface into the first feature extraction module can more clearly display the tiny branches of the gingival capillary network. In another case, if the user has pulp necrosis, the blood flow signal of the pulp vessels will be lost. Inputting the precisely registered capillary images of the gingival sulcus and alveolar bone surface, as well as the precisely registered three-dimensional images of the vascular network in the pulp and periapical region into the first feature extraction module, will allow the precisely registered three-dimensional images of the vascular network in the pulp and periapical region to clearly display the capillaries within the pulp cavity and the periapical vascular network. These images are then used for subsequent extraction of hemodynamic features.
[0063] In this embodiment, the first feature extraction module includes a first branch, a second branch, and a third branch. The first branch is used to extract the three-dimensional morphological features of blood vessels from the finely registered vascular network image using a 3D-ResNet50 backbone network, and to extract the centerline of the blood vessels using a skeletonization algorithm to calculate the blood vessel diameter distribution, branch point density, and blood vessel tortuosity, thereby obtaining spatial morphological features. The second branch is used to use the blood vessel intersection points in the finely registered vascular network image as nodes. Using the blood vessel segment as the boundary The vascular network is modeled as an undirected graph. The node features and adjacency matrix of the undirected graph are then input into a graph convolutional network to generate topological feature vectors and obtain topological connectivity features. The third branch is used to collect capillary images at preset time intervals. The resulting dynamic sequence images are input into a 2D ConvLSTM network with three convolutional LSTM layers to obtain hemodynamic features including blood flow velocity and perfusion maps.
[0064] S4. Extract the macroscopic morphological features and local lesion features of the registered tooth surface image through the second feature extraction module.
[0065] In step S4, the macroscopic morphological features include: tooth outline and crown morphological parameters.
[0066] Optionally, the step of extracting the macroscopic morphological features and local lesion features of the registered tooth surface image through the second feature extraction module includes:
[0067] The finely registered CBCT images are converted to grayscale to obtain grayscale images;
[0068] Tooth contours are extracted from grayscale images, and crown morphology parameters are calculated based on the tooth contours.
[0069] Adaptive threshold segmentation is performed on grayscale images to extract highlighted abnormal regions;
[0070] The gray-scale similarity criterion is used to perform region growing on the highlighted abnormal regions to obtain the lesion regions;
[0071] Texture features are extracted within the lesion area to obtain the local features of the lesion.
[0072] S5. Multi-dimensional features, macroscopic morphological features, local lesion features and clinical data are fused across modalities to obtain fused features.
[0073] Multidimensional features, macroscopic morphological features, local lesion features, and clinical data are respectively subjected to layer normalization and linear projection, mapped to the same dimension, and then weighted by gating fusion units to obtain fused features.
[0074] S6. Output periodontal status assessment results based on fusion features.
[0075] Specifically, the fused features are input into the fully connected layer to obtain the corresponding periodontal status levels, including: healthy, mild inflammation, moderate periodontitis, and moderate periodontitis.
[0076] To better prevent and treat periodontal disease, after the step of extracting multi-dimensional features of the registered vascular network image through the first feature extraction module, the method may further include:
[0077] Step a: Determine the key parameters of the vascular network based on the topological feature vector; key parameters include: the variance of the node degree distribution reflecting the uniformity of vascular branching. And the average path length that reflects the efficiency of blood flow pathways. .
[0078] After modeling the vascular network as an undirected graph, nodes represent vascular bifurcation points or endpoints, edges represent vascular segments, and node degree refers to the number of edges directly connected to that node.
[0079] In this embodiment, the average path length This represents the average shortest path length between all distinct node pairs in the vascular network, while the variance of the node degree distribution is used to calculate the shortest path length. First, iterate through all nodes in the vascular network, calculate the degree of each node, and count the number of nodes corresponding to each degree. Then, calculate the probability of the number of nodes corresponding to each degree relative to the total number of nodes, and finally calculate the expected value of the degree of all nodes. ; Compare the degree of each node with the expected value The variance of the node degree distribution is obtained by multiplying the square of the difference by the sum of the corresponding probabilities. .
[0080] Figure 2 This is a schematic diagram of the periapical vascular network provided in an embodiment of the present invention, wherein the periapical vascular network includes 5 nodes: , , , , For example, , , , , , This represents the degree of a node, where the number of nodes has a degree of 1. ( The number of nodes with a degree of 2 ( The number of nodes with a degree of 3 ( If the probability of the number of nodes corresponding to each degree being equal to the total number of nodes is... , , Based on the degree distribution probability, the expected value of the degree of all nodes can be further calculated: Finally, the variance of the node degree distribution is obtained. .
[0081] Step b, when the variance of the node degree distribution Greater than the first preset threshold and average path length When the value exceeds the second preset threshold, the bone resorption volume of the corresponding region is further calculated:
[0082] ;
[0083] In the formula, This represents the calibration coefficient.
[0084] Step c, when bone resorption volume When the value exceeds the third preset threshold, the area is marked as a high-risk bone resorption area.
[0085] It should be understood that the variance of the node degree distribution A value greater than the first preset threshold indicates uneven vascular branching, such as the presence of a few high-degree pivotal nodes and a large number of low-degree terminal nodes, which may lead to insufficient local blood perfusion (e.g., in periapical vascular rupture areas). The average path length... If the blood flow path is greater than the second preset threshold, it indicates that the blood flow path is too long, which may be due to increased blood flow resistance caused by narrowing or tortuous blood vessels (such as microartery tortuosity in the early stage of periodontitis).
[0086] By combining the variance of node degree distribution With average path length These two characteristics, and further predict the risk of alveolar bone resorption, can provide a basis for early intervention in periodontitis during the assessment process, such as marking the periapical region as a high-risk area for bone resorption, so that subsequent root canal treatment or periodontal surgery can be performed.
[0087] As can be seen from the above embodiments, the beneficial effects of the present invention are as follows:
[0088] (1) This invention provides a periodontal condition assessment method based on artificial intelligence. The collected multimodal tooth data includes not only tooth surface images but also images of local vascular networks in the oral cavity. Abnormalities in the morphology and dynamics of local vascular networks in the oral cavity are often early biomarkers of pathological processes such as inflammation and necrosis. This invention uses a first feature extraction module to extract the spatial morphological features, topological connectivity features, and hemodynamic features of blood vessels, which makes up for the blind spot of using only a single modality for periodontal condition assessment in the prior art. Thus, it can effectively utilize the microvascular network information closely related to the lesion, and can capture early inflammation based on blood flow abnormalities, thereby improving the sensitivity to periodontal diseases.
[0089] (2) In the process of periodontal status assessment, this invention integrates the spatial morphological features, topological connectivity features and hemodynamic features of blood vessels, as well as the macroscopic morphological features of teeth and the local features of lesions. This multimodal data collaborative analysis method is conducive to improving the accuracy of periodontal status assessment and early warning capability.
[0090] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0091] The use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples" indicates that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0092] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A periodontal condition assessment method based on artificial intelligence, characterized in that, include: Collect multimodal dental data, which includes images of the tooth surface and images of the vascular network in the oral cavity; The tooth surface images include: oral endoscope images and cone-beam computed tomography (CBCT) images; the vascular network images include: capillary images of the gingival sulcus and alveolar bone surface acquired using optical coherence tomography (OCTA) equipment, and three-dimensional images of the vascular network in the pulp and periapical region acquired using CT equipment. The multimodal tooth data is preprocessed and cross-modal registration is performed using a registration module to obtain registered tooth surface images and vascular network images. The first feature extraction module extracts multi-dimensional features of the registered vascular network image, including spatial morphological features, topological connectivity features, and hemodynamic features of the blood vessels. The macroscopic morphological features and local lesion features of the registered tooth surface image are extracted using the second feature extraction module. The multi-dimensional features, macroscopic morphological features, local lesion features, and clinical data are fused across modalities to obtain fused features; The periodontal status assessment result is output based on the fusion features; wherein... The steps of preprocessing and cross-modal registration of the multimodal tooth data to obtain registered tooth surface images and vascular network images include: using the intraoral endoscopic image as the reference image, and using the CBCT image, capillary image, and 3D vascular network image as images to be registered; marking key points representing the same anatomical structure in the reference image and the images to be registered respectively; mapping the images to be registered to the coordinate system of the reference image through affine transformation to obtain a coarse registration image; stitching the reference image and the coarse registration image together and inputting them into the registration module; and using a pre-trained VoxelMorph network to obtain a nonlinear deformation field. Based on the aforementioned nonlinear deformation field The coarsely registered image is aligned to the reference image to obtain the finely registered image; The step of fusing the multi-dimensional features, macroscopic morphological features, local lesion features, and clinical data across modalities to obtain fused features includes: performing layer normalization and linear projection on the multi-dimensional features, macroscopic morphological features, local lesion features, and clinical data respectively, mapping them to the same dimension, and further weighting them through a gated fusion unit to obtain fused features.
2. The periodontal condition assessment method based on artificial intelligence according to claim 1, characterized in that, The process of extracting multi-dimensional features from the registered vascular network image using a first feature extraction module, wherein the multi-dimensional features include spatial morphological features, topological connectivity features, and hemodynamic features of blood vessels, includes the following steps: Based on the target blood vessels to be focused on in the periodontal status assessment, the precisely registered vascular network image is input into the first feature extraction module, so that the first feature extraction module can extract the morphological features, topological connectivity features, and hemodynamic features of the vascular network; wherein, the target blood vessel is a soft tissue blood vessel or a hard tissue associated blood vessel.
3. The periodontal condition assessment method based on artificial intelligence according to claim 2, characterized in that, The first feature extraction module includes a first branch, a second branch, and a third branch; The first branch is used to extract the three-dimensional morphological features of blood vessels from the finely registered vascular network image using a 3D-ResNet50 backbone network, and to extract the center line of the blood vessels using a skeletonization algorithm to calculate the distribution of blood vessel diameter, branch point density and blood vessel curvature, thereby obtaining spatial morphological features. The second branch is used to take the vascular intersections in the finely registered vascular network image as nodes. Using the blood vessel segment as the boundary The vascular network is modeled as an undirected graph, and the node features and adjacency matrix of the undirected graph are further input into a graph convolutional network to generate a topological feature vector, thereby obtaining topological connectivity features. The third branch is used to acquire capillary images at preset time intervals and input the resulting dynamic sequence images into a 2D ConvLSTM network to obtain hemodynamic features including blood flow velocity maps and perfusion maps.
4. The periodontal condition assessment method based on artificial intelligence according to claim 3, characterized in that, The macroscopic morphological features include: tooth outline and crown morphological parameters; The steps for extracting the macroscopic morphological features and local lesion features of the registered tooth surface image through the second feature extraction module include: The finely registered CBCT images are converted to grayscale to obtain grayscale images; Tooth contours are extracted based on the grayscale image, and crown morphology parameters are calculated based on the tooth contours. Adaptive threshold segmentation is performed on the grayscale image to extract the high-brightness abnormal regions; The lesion region is obtained by performing region growing on the highlighted abnormal region using the gray-level similarity criterion; Texture features are extracted within the lesion area to obtain the local features of the lesion.
5. The periodontal condition assessment method based on artificial intelligence according to claim 1, characterized in that, The steps for outputting periodontal status assessment results based on the fusion features include: The fusion features are input into the fully connected layer to obtain the corresponding periodontal condition level; wherein... The periodontal condition levels include: healthy, mild inflammation, moderate periodontitis, and moderate periodontitis.
6. The periodontal condition assessment method based on artificial intelligence according to claim 3, characterized in that, After the step of extracting multi-dimensional features of the registered vascular network image through the first feature extraction module, the following steps are also included: Based on the topological feature vector, the key parameters of the vascular network are determined; the key parameters include: the variance of the node degree distribution, which reflects the uniformity of vascular branching. And the average path length that reflects the efficiency of blood flow pathways. ; When the variance of the node degree distribution The average path length is greater than the first preset threshold and When the value exceeds the second preset threshold, the bone resorption volume of the corresponding region is further calculated: ； In the formula, Indicates the calibration coefficient; When bone resorption volume When the value exceeds the third preset threshold, the area is marked as a high-risk bone resorption area.
7. The periodontal condition assessment method based on artificial intelligence according to claim 6, characterized in that, The average path length This is the average of the shortest path lengths between all distinct node pairs in the vascular network; Calculate the variance of the node degree distribution using the following steps. : Traverse all nodes in the vascular network, calculate the degree of each node, and count the number of nodes corresponding to each degree; Calculate the probability of the number of nodes corresponding to each degree relative to the total number of nodes, and construct the degree distribution probability; Based on the degree distribution probability, calculate the expected value of the degree of all nodes. ; Compare the degree of each node with the expected value. The variance of the node degree distribution is obtained by multiplying the square of the difference by the sum of the corresponding probabilities. .