Intelligent analysis method for oral medical image

Abstract

The invention relates to the technical field of image recognition, in particular to an intelligent oral medical image analysis method, which comprises the following steps: acquiring oral medical images of periodontitis in a historical period, and marking periodontitis-related key anatomical regions to obtain periodontitis sensitive parameters; calculating a change quantitative index of periodontitis sensitive parameters, analyzing a space gradient feature and a time gradient feature, and further determining a multi-dimensional gradient feature vector of each tooth; performing clustering analysis based on the multi-dimensional gradient feature vector to determine a typical mode feature vector; obtaining the oral medical image of the current period to determine the multi-dimensional gradient feature vector of each tooth; calculating the distance between the multi-dimensional gradient feature vector of the current period and the typical mode feature vector, and marking a potential hidden risk unit; acquiring dental plaque staining detection data of the current period; and marking a potential concealment risk correction unit based on the dental plaque staining detection data. The periodontitis latent development risk prediction precision of the individual is improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an intelligent analysis method for oral medical images. Background Technology

[0002] With the rapid development of oral imaging technology, especially the widespread adoption of cone-beam computed tomography (CBCT) and high-resolution digital X-rays, as well as the deep application of artificial intelligence in medical image analysis, significant progress has been made in periodontitis-assisted analysis based on oral imaging. Current technologies can automatically identify key anatomical structures such as teeth and alveolar bone through deep learning models, and even quantitatively measure and grade the degree of alveolar bone resorption, thus reducing the workload of dentists to some extent and providing more objective quantitative indicators than visual assessment. These advancements have made it possible to conduct refined and automated assessments of periodontal health using imaging data.

[0003] However, existing technologies primarily focus on analyzing single or localized images. This static assessment model fails to provide individualized predictions of the dynamic risks to tooth development in individuals with complete periodontal imaging records throughout their history. In particular, early, subtle changes in the biomechanical environment and biofilm activity are not effectively incorporated into the analytical framework. Therefore, there is an urgent need for an intelligent analytical method that can integrate multi-timepoint sequence imaging data with multimodal biological information, enabling a leap from static diagnosis to dynamic risk prediction.

[0004] Chinese Patent Publication No. CN120726118A discloses a system for locating and grading periodontal bone loss based on deep learning CT images. The system acquires panoramic and periapical radiographs of the periodontium, classifies and stores them, and annotates key anatomical points. The preprocessed image data is then input into a deep learning model for training and validation, resulting in a periodontal bone loss localization and detection model. Based on the key anatomical points located by the periodontal bone loss localization and detection model, a classification evaluation ratio is calculated. Based on the classification evaluation ratio, the periodontal bone loss is quantitatively classified and its site bias is determined.

[0005] Therefore, it is evident that the existing technology has the following problems: The periodontitis imaging analysis technology relies solely on static assessments based on single images, which cannot provide individualized and dynamic risk prediction. Furthermore, it fails to consider comprehensive analysis of micromechanical environmental changes that reflect early development risks, such as gaps between teeth and tooth displacement. Additionally, it does not incorporate real-time plaque activity data for cross-validation, resulting in low accuracy in predicting the latent development risk of periodontitis in individuals. Summary of the Invention

[0006] To address this, the present invention provides an intelligent analysis method for oral medical images to overcome the problems of existing periodontitis image analysis techniques that rely solely on static assessments based on single images, failing to provide individualized and dynamic risk prediction, and neglecting to consider comprehensive analysis of micromechanical environmental changes reflecting early development risks, such as gaps between teeth and tooth displacement. Furthermore, the lack of cross-validation using real-time plaque activity data results in low accuracy in predicting the latent development risk of periodontitis in individuals.

[0007] To achieve the above objectives, the present invention provides an intelligent analysis method for oral medical images, comprising: Acquire oral medical images of periodontitis at several time points within a historical period; Identify and label key anatomical regions related to periodontitis, including the interproximal contact area of ​​teeth, the gingival papilla area, and the three-dimensional centroid of teeth. Based on the key anatomical regions related to periodontitis, periodontitis sensitivity parameters of several teeth were obtained. These periodontitis sensitivity parameters include the three-dimensional volume of the gap between each tooth and adjacent teeth, as well as the precise positioning of each tooth. Quantitative indicators of changes in periodontitis sensitivity parameters of each tooth at adjacent time points; The spatial gradient characteristics corresponding to the changes in each tooth are analyzed based on the quantitative indicators. The temporal gradient features are determined based on the changing trend of the temporal sequence of the spatial gradient features of each tooth. The multidimensional gradient feature vector of each tooth is determined based on the spatial gradient features and the temporal gradient features of each tooth. Cluster analysis is performed based on the multidimensional gradient feature vectors of each tooth to determine the feature vectors of typical patterns. Obtain oral medical images at several time points in the current period to determine the multidimensional gradient feature vectors of each corresponding tooth; Calculate the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical pattern, and mark potential hidden risk units based on the distance change; Obtain dental plaque staining detection data corresponding to the current period, wherein the dental plaque staining detection data includes staining intensity value, staining distribution area and plaque biofilm maturity index; Based on the changes in dental plaque staining data of each tooth, the number of potential hidden risk units is adjusted to mark potential hidden risk correction units.

[0008] Furthermore, the quantitative indicators of change include the three-dimensional volume change rate of the gap between each tooth and adjacent teeth, as well as the minute displacement vector of each tooth.

[0009] Furthermore, the spatial gradient features include the difference in intensity of gradient changes and the synergy of gradient change directions; The gradient intensity difference is determined based on the ratio of the three-dimensional volume change rate of the gap corresponding to the first gradient region and the second gradient region of each tooth. The synergy of gradient change directions is determined based on the direction cosine of the minute displacement vectors corresponding to the first and second gradient regions of each tooth.

[0010] Furthermore, the temporal gradient features include the gradient intensity evolution slope and the gradient mode stability index; The gradient intensity evolution slope is determined based on the slope of the linear regression line corresponding to the time sequence of the spatial gradient features. The gradient pattern stability index is determined based on the similarity between spatial gradient feature patterns.

[0011] Furthermore, the process of determining the spatial gradient feature pattern includes: Several spatial gradient feature vectors are constructed based on the temporal sequence of the spatial gradient features of each tooth to determine several spatial gradient feature patterns.

[0012] Furthermore, the typical mode feature vector includes a typical deterioration mode feature vector and a typical stable mode feature vector; The typical deterioration mode feature vector is determined based on the cluster center vector of the multidimensional gradient feature vector corresponding to the first historical label. The typical stable pattern feature vector is determined based on the cluster center vector of the multidimensional gradient feature vector corresponding to the second historical label.

[0013] Furthermore, the process of marking potentially concealed risk units based on distance changes includes: If the first distance corresponding to each tooth is greater than the first preset distance threshold and the second distance is less than the second preset distance threshold, then the corresponding tooth is marked as a potential hidden risk unit.

[0014] Furthermore, the process of determining the first distance and the second distance includes: The first distance is determined based on the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical stable mode; The second distance is determined based on the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical deterioration mode.

[0015] Furthermore, determining whether to trigger a correction of the number of the potential hidden risks includes: Whether to trigger a correction of the number of potential hidden risks is determined based on the proportion of teeth in the first preset state of the current cycle, wherein the first preset state is that the proportion of dental plaque staining distribution area of ​​each tooth exceeds a preset proportion threshold.

[0016] Furthermore, the process of correcting the number of potential hidden risk units based on changes in plaque staining data for each tooth, and marking potential hidden risk correction units, includes: A comprehensive score for dental plaque staining detection is calculated based on the changes in the aforementioned dental plaque staining detection data. The plaque staining comprehensive score is compared with the preset first comprehensive score threshold and the second comprehensive score threshold to determine the potential hidden risk removal unit and the potential hidden risk addition unit. The potential hidden risk unit is modified based on the potential hidden risk removal unit and the potential hidden risk addition unit to mark the potential hidden risk modification unit.

[0017] Compared with existing technologies, the beneficial effects of this invention lie in its ability to achieve complete automated analysis from raw data to spatiotemporal features by acquiring historical periodontal sequence images and extracting periodontitis-sensitive parameters, quantitative indicators of change, spatiotemporal gradient features, and generating multidimensional feature vectors. This transforms complex oral images into calculable quantitative indicators characterizing the dynamic changes in the micromechanical environment of teeth. Furthermore, it establishes individualized typical risk patterns through cluster analysis based on historical data, and identifies teeth deviating from healthy patterns and approaching risk patterns by calculating distances in new monitoring cycles. Dental plaque staining, a data point directly reflecting the activity of pathogenic factors, is introduced to correct the initial identification results. The entire method, through multi-step, multi-modal data progressive analysis, achieves in-depth mining and early warning of the latent and progressive risks of periodontitis, improving the accuracy of predicting the latent development risk of periodontitis in individuals.

[0018] Furthermore, by defining the quantitative indicators of change as the three-dimensional volume change rate of the gap and the micro-displacement vector of the tooth, this invention precisely focuses on two micromechanical dimensions crucial to periodontal health. The gap volume change rate directly quantifies the process of loss of stability in the adjoint relationship between teeth. The micro-displacement vector of the tooth directly reflects the stability of the tooth within the alveolar socket, and its abnormal changes are sensitive indicators of periodontal supporting tissue damage and abnormal occlusal forces. By continuously monitoring these two indicators, this invention can capture early, subtle signals of mechanical environmental deterioration that are easily overlooked in single-time-point image analysis, thus laying a solid data foundation for achieving true early warning.

[0019] Furthermore, this invention defines spatial gradient features as the difference in intensity and the synergy of gradient directions, requiring calculations for the mesial and distal surfaces of each tooth separately. This allows for an independent and precise characterization of the spatial patterns of micromechanical changes on each proximal surface of the tooth. When constructing the comprehensive feature vector, these feature values ​​belonging to different proximal surfaces are included side-by-side, ensuring that the final vector representing the tooth's state comprehensively and completely reflects the mechanical environment changes in all adjacent regions. This improves the comprehensiveness of feature representation and the sensitivity of risk identification, providing a richer and more accurate data foundation for accurately assessing the overall risk status of teeth.

[0020] Furthermore, this invention introduces gradient intensity evolution slope and gradient pattern stability index as temporal gradient features, and requires the calculation of these dynamic indicators for each spatial feature sequence of each adjacent face, thereby achieving a refined temporal dimension analysis of risk evolution. This method can track the overall changing trend of a tooth. By capturing these finer-grained temporal dynamic characteristics, earlier and more accurate targeted warnings are achieved.

[0021] Furthermore, this invention clarifies that spatial gradient feature patterns are characterized by spatial gradient feature vectors and constructs their temporal sequences, providing a clear and operable data structure for subsequent temporal trend analysis and pattern stability calculations. Each feature vector is a snapshot of the spatial change pattern within a time interval, while the sequence is a film of these snapshots arranged chronologically. This ensures the logical rigor and computational feasibility of the transition from spatial analysis to temporal analysis, and is the key technical link that enables the entire method to process time-series data and achieve dynamic prediction.

[0022] Furthermore, this invention divides the sample set based on clear historical clinical outcomes and performs clustering to determine typical patterns. The multidimensional gradient feature vector upon which the clustering is based is a comprehensive vector containing all spatial and temporal feature information of the mesial and distal surfaces of the teeth. This invention establishes a highly individualized reference benchmark that reflects the overall condition of the teeth. The typical deterioration pattern feature vector originates from the comprehensive state characteristics of teeth that have actually deteriorated within the target individual's historical cycle before deterioration. This feature integrates abnormal information from the adjacent surfaces of the deteriorated teeth, thus possessing strong targeting for predicting similar comprehensive deviations in the overall condition of other teeth. The typical stable pattern feature vector represents the comprehensive baseline of stable dental health under the individual's physiological state. By comparing the characteristics of the latest monitoring cycle data with the characteristics of the individual's own historical data to determine the state of the new data, a more comprehensive assessment of the overall risk of a tooth can be made, making risk warnings more robust and reliable.

[0023] Furthermore, this invention employs a dual threshold judgment logic of first and second distances to mark risk units, and the distance calculation is based on a multi-dimensional gradient feature vector representing the overall state of the tooth. This invention achieves a comprehensive and robust risk quantification and identification mechanism. Calculating the distance to stable and deterioration modes essentially measures the similarity between the current overall state of the tooth and the individual's historical health baseline and risk precursors. Since the comprehensive vector encompasses information from all adjacent surfaces, this distance comparison can comprehensively weigh the condition of all areas on a tooth. Setting a threshold based on the statistical distribution of historical data, and only marking when both conditions are met simultaneously—a large difference from the healthy state and a high similarity to risk precursors—ensures that the marked teeth exhibit clear and comprehensive risk signs in their overall state, improving the specificity of the warning results.

[0024] Furthermore, this invention uses a macroscopic assessment based on the proportion of plaque exceeding the standard across all teeth to determine whether to trigger a correction process. This invention introduces an efficient cost control and logic optimization mechanism. If the patient's overall oral hygiene is good and the plaque load is low, the overall probability of plaque driving new risks as a major pathogenic factor in the current period is low. In this case, the analysis results based on imaging biomechanical characteristics can be considered less affected by plaque and relatively reliable. Therefore, there is no need to initiate complex tooth-by-tooth plaque data correction; the initial results can be output directly. This avoids unnecessary calculations in low-risk scenarios and improves system efficiency. Conversely, when the plaque load is high, a comprehensive correction is initiated to ensure that biological risk factors are fully considered, demonstrating the intelligence and practicality of the logic.

[0025] Furthermore, this invention achieves a deep and organic integration of radiographic and microbiological risk evidence by constructing a comprehensive score for dental plaque staining detection and setting high and low thresholds to specifically modify the risk unit list. The comprehensive score integrates information from three dimensions—staining intensity, distribution area, and maturity changes—in a weighted manner, forming a holistic quantification of local biological risks in teeth. By comparing thresholds, removal and addition operations are performed. This modification process relies on subjective judgment based on radiographic features. When radiographically suspicious areas lack plaque activity support, their risk level is cautiously reduced; simultaneously, potential risk points with extremely high plaque activity, but which may not yet have caused obvious radiographic changes due to being in a very early stage, are added. This cross-validation mechanism ensures that the final risk prediction results do not overly rely on a single modality while fully leveraging the advantages of different data sources, thereby improving the accuracy of predicting the risk of insidious periodontitis development in individuals. Attached Figure Description

[0026] Figure 1 This is a flowchart of the intelligent analysis method for oral medical images according to an embodiment of the present invention; Figure 2 A flowchart for determining spatial gradient features in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the determination of temporal gradient features in an embodiment of the present invention. Figure 4 A flowchart for determining typical pattern feature vectors in embodiments of the present invention. Detailed Implementation

[0027] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0028] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0030] Please see Figure 1 The diagram shows a flowchart of an intelligent analysis method for oral medical images according to an embodiment of the present invention. The embodiment of the present invention provides an intelligent analysis method for oral medical images, comprising: Step S1: Obtain oral medical images of several time points of periodontitis within a historical period; Step S2: Use a deep learning model to automatically identify and label key anatomical regions related to periodontitis. These key anatomical regions include the interproximal contact area of ​​teeth, the gingival papilla area, and the three-dimensional centroid of teeth. Step S3: Based on the key anatomical regions related to periodontitis, obtain periodontitis sensitivity parameters for several teeth. The periodontitis sensitivity parameters include the three-dimensional volume of the gap between each tooth and adjacent teeth, and the precise positioning of each tooth. Step S4: Calculate the quantitative indicators of the changes in periodontitis sensitivity parameters of each tooth at adjacent time points; Step S5: Analyze the spatial gradient characteristics corresponding to the changes in each tooth based on the quantitative indicators. Step S6: Determine the temporal gradient features based on the changing trend of the temporal sequence of the spatial gradient features of each tooth; Step S7: Determine the multidimensional gradient feature vector of each tooth based on the spatial gradient features and the temporal gradient features of each tooth. Step S8: Perform cluster analysis based on the multidimensional gradient feature vectors of each tooth to determine the typical pattern feature vectors. Step S9: Obtain oral medical images of several time points in the current period to determine the multidimensional gradient feature vector of each tooth. Step S10: Calculate the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the typical pattern feature vector, and mark potential hidden risk units based on the distance change. Step S11: Obtain the dental plaque staining detection data corresponding to the current cycle. The dental plaque staining detection data includes staining intensity value, staining distribution area, and plaque biofilm maturity index. Step S12: Based on the changes in plaque staining detection data of each tooth, the number of potential hidden risk units is corrected to mark potential hidden risk correction units.

[0031] In this embodiment, in step S1, for periodontitis patients requiring long-term follow-up management, oral medical images are collected at preset, regular time intervals, such as monthly, throughout a complete historical treatment and follow-up cycle, for example, a 24-month periodontal system treatment and maintenance period. These images should include image types that clearly show teeth, alveolar bone, and adjacent relationships, such as cone-beam computed tomography (CBCT) images or a series of panoramic X-rays, or oral dental images taken with a camera, as long as the images can cover all teeth, gingiva, and periodontium. All images must be collected using standardized positions and parameters, and the accurate correspondence and archiving of patient identification information and collection timestamps must be ensured to form a structured time-series image dataset.

[0032] In this embodiment, in step S2, a pre-trained deep learning segmentation model, such as an improved model based on U-Net++ or Mask R-CNN architecture, is deployed to process the aforementioned sequence of images. The training samples for this model consist of a dataset containing thousands of high-quality oral medical images, such as CBCT and panoramic films, each image of which has been pixel-level precisely annotated by oral medical experts. The annotations include the complete outline of the tooth crown and root, the location of the cementoenamel junction, the alveolar ridge crest, and the outline of the gingival papilla region between adjacent teeth. By learning from these annotated samples, the model can automatically identify the boundaries of teeth and surrounding soft and hard tissues in the images. For each input oral medical image to be identified, the trained model automatically outputs a binary mask of the following key anatomical regions: a complete three-dimensional surface model of the tooth, the contact area between the proximal surfaces of the teeth, and the outline of the triangular soft tissue region between adjacent teeth below the crown contact point. Simultaneously, based on the three-dimensional volume of each segmented tooth, the coordinates (x, y, z) of the three-dimensional centroid of each tooth are obtained by calculating its geometric center. This step transforms raw images into structured anatomical information, laying the foundation for subsequent quantitative analysis.

[0033] In this embodiment, in step S3, geometric calculations are performed in three-dimensional space based on the structured anatomical information output in step S2. For the three-dimensional volume of the gap between each tooth and its adjacent teeth, calculations are performed separately for the mesial and distal surfaces: First, based on the proximal contour mask and crown surface model of the adjacent teeth, three-dimensional models of the gap between the mesial and distal surfaces and the adjacent teeth are reconstructed, and their respective gap volumes are calculated. For precise positioning of each tooth, the three-dimensional centroid coordinates (x, y, z) of each tooth calculated in S2 are directly used. These parameters collectively constitute periodontitis-sensitive parameters reflecting the mechanical contact relationship between teeth and the spatial stability of teeth.

[0034] Specifically, in step S4, the change quantification indicators include the three-dimensional volume change rate of the gap between each tooth and adjacent teeth, as well as the minute displacement vector of each tooth.

[0035] In this embodiment, based on the three-dimensional volume of each interproximal gap and the precise positioning of the teeth obtained in S3 at each time point, the following calculations are performed for any two adjacent inspection time points T_i and T_{i+1} arranged in chronological order within the historical or current cycle: First, for the rate of change of the three-dimensional volume of each interproximal gap, the calculation formula is (V_{i+1}-V_i) / (T_{i+1}-T_i) for both the mesial and distal surfaces, where V_i and V_{i+1} represent the three-dimensional volume of the gap at time points T_i and T_{i+1}, respectively, and the unit of the result can be mm. 3 / day. This index quantifies the rate at which the interproximal gaps widen over time. Secondly, for the minute displacement vector of a tooth, the displacement vector ΔD is obtained by calculating the difference in the coordinates of the three-dimensional centroid of the same tooth at two time points T_i and T_{i+1}, i.e., ΔD = (x_{i+1} - x_i, y_{i+1} - y_i, z_{i+1} - z_i). This vector contains not only the magnitude of the displacement (module) but also the direction of movement in three-dimensional space. By performing this calculation on all adjacent time points throughout the entire cycle, a sequence of quantitative indicators for the changes corresponding to each time interval can be obtained.

[0036] Please see Figure 2 As shown, it is a flowchart for determining spatial gradient features according to an embodiment of the present invention.

[0037] Specifically, in step S5, the spatial gradient features include the difference in intensity of gradient changes and the synergy of gradient change directions; In step S51, the gradient intensity difference is determined based on the ratio of the three-dimensional volume change rate of the gap corresponding to the first gradient region and the second gradient region of each tooth. Step S52, the synergy of the gradient change method is determined based on the direction cosine of the micro displacement vector corresponding to the first gradient region and the second gradient region of each tooth.

[0038] In this embodiment, firstly, on the 3D image, for each proximal surface of the tooth, including the mesial and distal surfaces, the corresponding root and surrounding alveolar bone region is divided into two continuous gradient regions centered on the long axis of the tooth: a first gradient region, which is the coronal region close to the crown, for example, extending from the cementoenamel junction towards the root to half the root length; and a second gradient region, which is the apical region far from the crown. Then, for each time interval, the 3D volume change rate of the gap on that proximal surface is calculated, and its contribution or average change rate in the first and second gradient regions needs to be estimated separately. This can be achieved by analyzing the spatial location of the gap volume change within that time interval, or by modeling and assigning weights. Finally, the difference in change intensity between the gradients is calculated as the ratio of the average change rate of the second region to the average change rate of the first region. If the ratio is greater than 1, it indicates that the apical change is faster. To assess the synergy of gradient directions, calculations must be performed separately for the mesial and distal surfaces. First, based on the tooth displacement vector and the tooth geometry model, it's estimated whether the displacement primarily originates from crown forces or root resorption. This allows for the decomposition or association of minute displacement vectors to the first and second gradient regions. Then, the direction cosine of the displacement vectors in these two regions—the cosine of the angle between the two vectors—is calculated. A value closer to 1 indicates a more consistent direction of crown-root displacement and more synergistic mechanical behavior; a value closer to -1 indicates opposite directions, potentially suggesting complex lever or torsional effects.

[0039] Please see Figure 3 As shown, it is a flowchart for determining the time gradient characteristics in an embodiment of the present invention.

[0040] Specifically, in step S6, the temporal gradient features include the gradient intensity evolution slope and the gradient mode stability index. Step S61, the gradient evolution slope is determined based on the slope of the linear regression line corresponding to the time series of the spatial gradient features; Step S62, the gradient pattern stability index is determined based on the similarity between spatial gradient feature patterns.

[0041] Specifically, in step S62, the process of determining the spatial gradient feature pattern includes: Step S621: Construct several spatial gradient feature vectors based on the temporal sequence of the spatial gradient features of each tooth to determine several spatial gradient feature patterns.

[0042] In this embodiment, the process of determining the gradient intensity evolution slope is as follows: For each spatial gradient feature of each adjacent surface of each tooth, including the mesial and distal surfaces, such as the difference in gradient intensity between mesial surfaces, the values ​​calculated at each consecutive time interval throughout the entire historical period are arranged in chronological order to form an independent time series. A univariate linear regression analysis is performed on this series to fit a straight line. The slope of this straight line is determined as the gradient intensity evolution slope of the specific spatial feature of that particular adjacent surface. A positive slope indicates that the feature increases over time, while a negative slope indicates that it weakens. Determining the gradient pattern stability index requires first defining the spatial gradient feature pattern.

[0043] In this embodiment, at each time interval i, the values ​​corresponding to the spatial gradient features of all adjacent surfaces of a tooth calculated in step S5, such as the difference in gradient intensity between mesial surfaces = 1.5, the directional coherence between mesial surface gradients = 0.9, the difference in gradient intensity between distal surfaces = 1.2, and the directional coherence between distal surface gradients = 0.95, can be represented as a comprehensive vector, denoted as P_i. This vector P_i is the spatial gradient feature vector corresponding to that time point, which completely encodes the comprehensive spatial distribution pattern of the changes in the micromechanical environment of all adjacent surfaces of the tooth within that specific time interval. Therefore, for a period containing n time intervals, we can obtain n such spatial gradient feature vectors (P_1, P_2, ..., P_n), which constitute a sequence of several spatial gradient feature patterns. The subsequent calculation of the gradient pattern stability index is accomplished by comparing the similarity between these continuous vectors P_i and P_{i+1}.

[0044] The gradient pattern stability index measures the stability of a pattern. It requires calculating the cosine similarity between the spatial gradient feature vectors of two adjacent time intervals. Then, the average of these similarity values ​​over all adjacent time intervals is taken as the gradient pattern stability index. A higher index, closer to 1, indicates more similar and stable spatial variation patterns at different time points; a lower index indicates greater pattern fluctuations and instability.

[0045] In this embodiment, step S7 involves taking the temporal sequence of the spatial gradient feature vector constructed in step S6, calculating the mean, standard deviation, and maximum value for each term in the spatial gradient feature vector, and concatenating these values ​​with the corresponding temporal gradient features in a fixed order to create a multidimensional gradient feature vector for the tooth. This vector comprehensively represents all the key information regarding the spatial patterns and temporal trends of the micromechanical environment changes on all adjacent surfaces of the tooth throughout its entire historical period.

[0046] Please see Figure 4 As shown, it is a flowchart for determining the feature vector of a typical pattern in an embodiment of the present invention.

[0047] Specifically, in step S8, the typical mode feature vector includes a typical deterioration mode feature vector and a typical stable mode feature vector; In step S81, the typical deterioration mode feature vector is determined based on the cluster center vector of the multidimensional gradient feature vector corresponding to the first historical label. Step S82, the typical stable mode feature vector is determined based on the cluster center vector of the multidimensional gradient feature vector corresponding to the second historical label.

[0048] In this embodiment, based on the complete medical imaging records of the target individual throughout the entire historical period, teeth clearly marked as having newly developed periodontitis or showing clear progression of existing periodontitis at the end of the historical period are identified as the first historical label. Teeth marked as consistently healthy or with long-term stable lesions throughout the historical period are identified as the second historical label. A set of multidimensional gradient feature vectors corresponding to all teeth with the first historical label is obtained and clustered, with the cluster centers determined as the typical deterioration pattern feature vectors. A set of multidimensional gradient feature vectors corresponding to all teeth with the second historical label is obtained and clustered, with the cluster centers determined as the typical stability pattern feature vectors. These two vectors represent the two typical combined feature patterns of the individual's historical progression towards deterioration and stability, respectively.

[0049] In this embodiment, in step S9, when the same target individual enters a new, independent monitoring cycle, such as a health maintenance period after completing previous treatment and visits again, oral medical image sequences at three or more consecutive time points within the current cycle are collected. Steps S2 to S7 are then executed completely independently and repeatedly for this new sequence. That is, key anatomical regions are first identified, then newly generated periodontitis sensitivity parameters, change quantification indicators, spatial gradient features, and temporal gradient features within the current cycle are calculated, ultimately generating a completely new multidimensional gradient feature vector for each tooth within the current cycle, representing its change pattern in the latest time period.

[0050] Specifically, in step S10, the process of marking potential hidden risk units based on distance changes includes: Step S101: If the first distance corresponding to each tooth is greater than the first preset distance threshold and the second distance is less than the second preset distance threshold, then the corresponding tooth is marked as a potential hidden risk unit.

[0051] Specifically, in step S101, the process of determining the first distance and the second distance includes: Step S1011, the first distance is determined based on the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical stable mode; Step S1012, the second distance is determined based on the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical deterioration mode.

[0052] In this embodiment, for each multi-dimensional gradient feature vector newly generated for each tooth in the current period, the Euclidean distance between it and the typical stable pattern feature vector obtained from historical data is calculated as the first distance D_s; the Euclidean distance between it and the typical deterioration pattern feature vector is calculated as the second distance D_r. Two key thresholds need to be preset: the first preset distance threshold Th_s and the second preset distance threshold Th_r. In the historical period, calculate the distance distribution of the multi-dimensional gradient feature vectors corresponding to all teeth with the first historical label to the typical stable pattern feature vector, and determine the 90th percentile of this distance distribution as the first preset distance threshold Th_s; calculate the distance distribution of the multi-dimensional gradient feature vectors corresponding to all teeth with the second historical label to the typical deterioration pattern feature vector, and determine the 10th percentile of this distance distribution as the second preset distance threshold Th_r. For each tooth in the current period, if the conditions D_s>Th_s and D_r<Th_r are satisfied simultaneously, it means that the current comprehensive state feature of the tooth is sufficiently different from its own historical stable pattern and is sufficiently similar to the precursor pattern of deterioration. Therefore, it is marked as a potential hidden risk unit. This marking result indicates that there is a risk in the entire tooth, and the risk may come from abnormal changes in one or more adjacent surfaces.

[0053] Specifically, in step S11, at a time point that is the same as or as close as possible to the time when the oral medical image is collected in step S9, use a dental plaque staining agent to stain the tooth surface of the patient, and collect an image through a high-resolution color camera, and obtain three core data: the staining intensity value, the average value of the red channel intensity of all pixels in the stained area. The higher the staining intensity value, the deeper the plaque staining in this area, reflecting a larger plaque biomass; the stained distribution area, separate the area stained by the staining agent from the tooth image through an image segmentation algorithm, and calculate the actual pixel area of this area; the plaque biofilm maturity index, which is obtained by calculating the ratio (R / G) of the average intensity of the red channel to the average intensity of the green channel in the stained area image. As the plaque biofilm matures, it causes the color tone after staining to shift in a specific direction, and the R / G ratio can quantify this change in color characteristics. These data need to be accurately associated with the tooth site.

[0054] Specifically, in step S12, determining whether to trigger the process of correcting the number of the potential hidden risks includes: Step S121, determining whether to trigger the correction of the number of the potential hidden risks based on the proportion of the number of teeth in the first preset state in the current period, where the first preset state is that the proportion of the stained area distribution of dental plaque on each tooth exceeds a preset proportion threshold.

[0055] In this embodiment, after acquiring the plaque staining detection data for the current period, a full-mouth plaque load assessment is performed. For each examination time point in the current period, all teeth in the entire mouth are examined, and the percentage of plaque staining distribution area for each tooth (i.e., the percentage of the stained area of ​​the tooth to its crown surface area) is determined. Whether this percentage exceeds a preset threshold is checked; preferably, the threshold is set to 20%. If the percentage of plaque staining distribution area for a tooth is greater than the threshold, the oral cavity is in a first preset state. The number of teeth in this first preset state (i.e., exceeding the area threshold) at that time point is counted, and their proportion to the total number of teeth examined is calculated. Since there are multiple time points in the current period, the average of this proportion at each time point is taken. If the calculated full-mouth plaque excess proportion exceeds a preset trigger threshold, the risk unit list based on image feature marking needs to be corrected; otherwise, the plaque factor interference is considered small, and the initial marking result can be directly adopted. Preferably, the trigger threshold is set to 50%.

[0056] Specifically, in step S12, the process of correcting the number of potential hidden risk units based on changes in plaque staining data for each tooth, and marking potential hidden risk correction units, includes: Step S122: Calculate the comprehensive score for dental plaque staining detection based on the changes in the dental plaque staining detection data; Step S123: Based on the comprehensive score of dental plaque staining, compare it with the preset first comprehensive score threshold and the second comprehensive score threshold to determine the potential hidden risk removal unit and the potential hidden risk addition unit. Step S124: Based on the potential hidden risk removal unit and the potential hidden risk addition unit, the potential hidden risk unit is modified to mark the potential hidden risk modification unit.

[0057] In this embodiment, the comprehensive plaque staining detection score Score of each tooth is calculated. This score is a composite index that integrates multi-dimensional staining data, and the calculation formula is: Score = w1 * N(F_intensity) + w2 * N(A_coverage) + w3 * Trend_Index. Among them, N(F_intensity) is the normalized staining intensity value, N(A_coverage) is the normalized proportion of the stained area, and Trend_Index is the quantification value of the change trend of the plaque maturity index within the cycle. By performing linear regression on the sequence of plaque maturity index at several time points within the current cycle, the regression slope is taken as the value of Trend_Index. w1, w2, and w3 are preset weight coefficients. Preferably, they are set to 0.5, 0.3, and 0.2 respectively. Two thresholds are set: the first comprehensive score threshold Th_low and the second comprehensive score threshold Th_high. Preferably, the first comprehensive score threshold is set to the tenth percentile of the comprehensive plaque staining detection scores of all teeth marked with the second historical label in the historical cycle data, and the second comprehensive score threshold is set to the ninetieth percentile of the comprehensive plaque staining detection scores of all teeth marked with the first historical label in the historical cycle data. Then, a determination is made. For the teeth that have been marked as potential hidden risk units in step S10, if their Score < Th_low, it indicates that although there are risk signs in the imaging, the current plaque activity is extremely low, so they are determined as potential hidden risk removal units. For the teeth not marked in step S10, if their Score > Th_high, it indicates that their plaque activity is extremely high and they are clear biological risk factors, so they are determined as potential hidden risk new units. From the original risk unit list, all removal units are removed; at the same time, all new units are added to the list. The finally obtained updated list is the potential hidden risk correction unit. The potential hidden risk correction unit or the potential hidden risk unit determined not to be corrected in step S12 is used to generate the final target individual oral health risk warning report, which includes a list of high-risk tooth sites.

[0058] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

Claims

1. An oral medical image intelligent analysis method, characterized in that, include: Acquire oral medical images of periodontitis at several time points within a historical period; Identify and label key anatomical regions related to periodontitis, including the interproximal contact area of ​​teeth, the gingival papilla area, and the three-dimensional centroid of teeth. Based on the key anatomical regions related to periodontitis, periodontitis sensitivity parameters of several teeth were obtained. These periodontitis sensitivity parameters include the three-dimensional volume of the gap between each tooth and adjacent teeth, as well as the precise positioning of each tooth. Quantitative indicators of changes in periodontitis sensitivity parameters of each tooth at adjacent time points; The spatial gradient characteristics corresponding to the changes in each tooth are analyzed based on the quantitative indicators. The temporal gradient features are determined based on the changing trend of the temporal sequence of the spatial gradient features of each tooth. The multidimensional gradient feature vector of each tooth is determined based on the spatial gradient features and the temporal gradient features of each tooth. Cluster analysis is performed based on the multidimensional gradient feature vectors of each tooth to determine the feature vectors of typical patterns. Obtain oral medical images at several time points in the current period to determine the multidimensional gradient feature vectors of each corresponding tooth; Calculate the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical pattern, and mark potential hidden risk units based on the distance change; Obtain dental plaque staining detection data corresponding to the current period, wherein the dental plaque staining detection data includes staining intensity value, staining distribution area and plaque biofilm maturity index; Based on the changes in dental plaque staining data of each tooth, the number of potential hidden risk units is adjusted to mark potential hidden risk correction units.

2. The intelligent analysis method for oral medical images according to claim 1, characterized in that, The quantitative indicators of change include the three-dimensional volume change rate of the gap between each tooth and adjacent teeth, as well as the minute displacement vector of each tooth.

3. The intelligent analysis method for oral medical images according to claim 2, characterized in that, The spatial gradient characteristics include the difference in intensity of gradient changes and the synergy of gradient change directions. The gradient intensity difference is determined based on the ratio of the three-dimensional volume change rate of the gap corresponding to the first gradient region and the second gradient region of each tooth. The synergy of gradient change directions is determined based on the direction cosine of the minute displacement vectors corresponding to the first and second gradient regions of each tooth.

4. The intelligent analysis method for oral medical images according to claim 3, characterized in that, The temporal gradient features include the gradient intensity evolution slope and the gradient mode stability index. The gradient intensity evolution slope is determined based on the slope of the linear regression line corresponding to the time sequence of the spatial gradient features. The gradient pattern stability index is determined based on the similarity between spatial gradient feature patterns.

5. The intelligent analysis method for oral medical images according to claim 4, characterized in that, The process of determining the spatial gradient feature pattern includes: Several spatial gradient feature vectors are constructed based on the temporal sequence of the spatial gradient features of each tooth to determine several spatial gradient feature patterns.

6. The intelligent analysis method for oral medical images according to claim 1, characterized in that, The typical pattern feature vector includes the typical deterioration pattern feature vector and the typical stable pattern feature vector. The typical deterioration mode feature vector is determined based on the cluster center vector of the multidimensional gradient feature vector corresponding to the first historical label. The typical stable pattern feature vector is determined based on the cluster center vector of the multidimensional gradient feature vector corresponding to the second historical label.

7. The intelligent analysis method for oral medical images according to claim 6, characterized in that, The process of marking potentially hidden risk units based on distance changes includes: If the first distance corresponding to each tooth is greater than the first preset distance threshold and the second distance is less than the second preset distance threshold, then the corresponding tooth is marked as a potential hidden risk unit.

8. The intelligent analysis method for oral medical images according to claim 7, characterized in that, The process of determining the first distance and the second distance includes: The first distance is determined based on the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical stable mode; The second distance is determined based on the distance between the multidimensional gradient feature vector of each tooth in the current cycle and the feature vector of the typical deterioration mode.

9. The intelligent analysis method for oral medical images according to claim 8, characterized in that, The process of determining whether to trigger a correction to the number of the potential hidden risks includes: Whether to trigger a correction of the number of potential hidden risks is determined based on the proportion of teeth in the first preset state of the current cycle, wherein the first preset state is that the proportion of dental plaque staining distribution area of ​​each tooth exceeds a preset proportion threshold.

10. The intelligent analysis method for oral medical images according to claim 1, characterized in that, The process of revising the number of potential hidden risk units based on changes in plaque staining data for each tooth, and marking potential hidden risk revision units, includes: A comprehensive score for dental plaque staining detection is calculated based on the changes in the aforementioned dental plaque staining detection data. The plaque staining comprehensive score is compared with the preset first comprehensive score threshold and the second comprehensive score threshold to determine the potential hidden risk removal unit and the potential hidden risk addition unit. The potential hidden risk unit is modified based on the potential hidden risk removal unit and the potential hidden risk addition unit to mark the potential hidden risk modification unit.

Images