Visual recognition method and system for dental cracks

Abstract

The application provides a tooth crack visual recognition method and system, which comprises the following steps: collecting multi-channel visual images of a target tooth; preprocessing the multi-channel visual images to obtain preprocessed multi-channel visual images; extracting multi-scale visual features of the target tooth from the preprocessed multi-channel visual images, performing tooth texture feature screening on the multi-scale visual features, and obtaining a texture feature subset of the target tooth; obtaining historical tooth crack recognition results; performing tooth crack focusing processing on the texture feature subset based on tooth crack recognition features in the historical tooth crack recognition results to obtain a focused feature subset of the target tooth crack; and performing crack recognition on shallow micro-cracks and hidden cracks of the target tooth according to the focused feature subset to output a recognition result of whether the target tooth has cracks.

Description

Technical Field

[0001] This application relates to the field of visual recognition technology, and more specifically, to a method and system for visual recognition of tooth cracks. Background Technology

[0002] Visual recognition is a core technology in the field of computer vision, aiming to give machines the ability to perceive, understand and analyze visual information like humans. It extracts features from images or videos through algorithmic models (especially deep learning and convolutional neural networks) to achieve tasks such as object detection, classification, segmentation and scene understanding.

[0003] In the field of oral clinical diagnosis, tooth cracks are a significant cause of tooth sensitivity, pulpitis, and even tooth fracture. Timely and accurate identification of such cracks is crucial for early intervention in oral diseases. Traditional methods for detecting tooth cracks mainly rely on clinical visual observation and probe examination. However, due to the concealed nature of crack morphology and interference from tooth surface structures, it is difficult to capture subtle grayscale changes in shallow microcracks and texture distortion characteristics of hidden cracks, resulting in a high rate of missed detections. With the application of computer vision technology in oral medicine, some tooth crack identification methods based on single-channel images have been... While proposed, such methods have several limitations: First, the feature extraction dimension is singular, lacking multi-scale feature layering of macroscopic contours, mesoscopic textures, and microscopic grayscale fluctuations, making it difficult to distinguish crack features from interfering features such as physiological tooth depressions and mild enamel wear; second, the lack of focused processing based on crack feature data from historical clinically confirmed cases results in insufficient recognition of crack features by the identification model, with false positive and false negative rates failing to meet clinical application requirements. Therefore, how to reduce the impact of tooth texture features on the identification of tooth cracks has become a problem faced by the industry. Summary of the Invention

[0004] This application provides a method and system for visual recognition of tooth cracks, which can reduce the influence of tooth texture features on the recognition of tooth cracks.

[0005] In a first aspect, this application provides a method for visually recognizing tooth cracks, comprising the following steps: Acquire multi-channel visual images of the entire target tooth structure; The multi-channel visual image is preprocessed to obtain a preprocessed multi-channel visual image; Multi-scale visual features of the target tooth are extracted from the preprocessed multi-channel visual image, and tooth texture features are filtered from the multi-scale visual features to obtain a subset of texture features of the target tooth. Obtain historical tooth crack identification results, and perform tooth crack focusing processing on the texture feature subset based on the tooth crack identification features in the historical tooth crack identification results to obtain the focused feature subset of the target tooth crack; Based on the focused feature subset, crack identification is performed on the shallow microcracks and hidden cracks of the target tooth to output the identification result of whether the target tooth has cracks.

[0006] In some embodiments, preprocessing the multi-channel visual image to obtain a preprocessed multi-channel visual image specifically includes: The multi-channel visual image is subjected to de-reflection processing to obtain a de-reflection processed multi-channel visual image; The de-reflection processed multi-channel visual image is enhanced to obtain an enhanced multi-channel visual image. The enhanced multi-channel visual image is then denoised to obtain a preprocessed multi-channel visual image.

[0007] In some embodiments, extracting multi-scale visual features of the target tooth from the preprocessed multi-channel visual image specifically includes: Multi-scale feature extraction is performed on the preprocessed multi-channel visual image to obtain image features at each scale; The multi-scale visual features of the target tooth are determined based on image features at various scales.

[0008] In some embodiments, filtering the tooth texture features of the multi-scale visual features to obtain a subset of texture features of the target tooth specifically includes: Obtain a texture reference for normal teeth; Based on the texture reference benchmark, the texture features of the multi-scale visual features are compared and filtered to obtain the texture features at each scale. By integrating the texture features at various scales, a subset of texture features of the target tooth is obtained.

[0009] In some embodiments, the historical tooth crack identification results include clinically validated case basic information, multi-scale abnormal texture feature parameters of the corresponding case, correlation coefficient between features and cracks, and the final clinical crack determination result.

[0010] In some embodiments, the tooth crack identification features in the historical tooth crack identification results are used to perform tooth crack focusing processing on the texture feature subset to obtain the focused feature subset of the target tooth crack, specifically including: Determine the tooth crack identification features of the historical tooth crack identification results; The tooth crack identification features are compared one by one with the texture feature subset to filter out the relevant texture features that focus on tooth cracks. The focused feature subset of the target tooth crack is determined by identifying all relevant texture features.

[0011] In some embodiments, crack identification of shallow microcracks and occult cracks in the target tooth based on the focused feature subset, and outputting the identification result of whether the target tooth has cracks, specifically includes: Determine the crack identification rules for shallow microcracks and hidden cracks in the target tooth structure; Based on the crack determination rule, feature recognition is performed on the focused feature subset to obtain feature recognition scores at each scale; The presence or absence of cracks in the target tooth is determined by feature recognition scores at various scales.

[0012] Secondly, this application provides a visual recognition system for tooth cracks, comprising: The acquisition module is used to acquire multi-channel visual images of the entire target tooth structure; The processing module is used to preprocess the multi-channel visual image to obtain a preprocessed multi-channel visual image. The processing module is further configured to extract multi-scale visual features of the target tooth from the preprocessed multi-channel visual image, and to perform tooth texture feature filtering on the multi-scale visual features to obtain a subset of texture features of the target tooth. The processing module is also used to obtain historical tooth crack recognition results, and to perform tooth crack focusing processing on the texture feature subset based on the tooth crack recognition features in the historical tooth crack recognition results to obtain the focused feature subset of the target tooth crack. The execution module is used to identify shallow microcracks and hidden cracks in the target tooth based on the focused feature subset, and to output the identification result of whether the target tooth has cracks.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described visual recognition method for tooth cracks.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for visual recognition of tooth cracks.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The method and system for visual recognition of tooth cracks provided in this application first acquire a multi-channel visual image of the entire target tooth; preprocess the multi-channel visual image to obtain a preprocessed multi-channel visual image; extract multi-scale visual features of the target tooth from the preprocessed multi-channel visual image, and filter the multi-scale visual features for tooth texture features to obtain a subset of texture features of the target tooth; obtain historical tooth crack recognition results, and perform tooth crack focusing processing on the texture feature subset based on the tooth crack recognition features in the historical tooth crack recognition results to obtain a focused feature subset of the target tooth crack; and perform crack recognition on shallow microcracks and hidden cracks of the target tooth according to the focused feature subset to output the recognition result of whether the target tooth has cracks.

[0016] Therefore, in the process of visual recognition of tooth cracks, this application firstly enriches the tooth characterization information from different physical property levels by acquiring multi-channel visual images; secondly, the preprocessing step reduces interference introduced by inconsistent imaging conditions through operations such as denoising, enhancement, and standardization, avoiding misjudging image noise as texture or crack; then, multi-scale visual feature extraction and texture feature screening are performed, with multi-scale features able to simultaneously capture the morphological differences between macroscopic textures and microscopic cracks; and actively screening texture feature subsets can directly exclude a large number of normal physiological textures of the tooth that are unrelated to cracks, significantly reducing the weight of texture features in subsequent recognition; furthermore, crack focusing processing based on historical recognition results suppresses the response of general tooth textures, enhancing the saliency and separability of crack features; finally, shallow microcracks and hidden cracks are identified, and the crack-sensitive feature set, which is selected through layers of screening and focusing, rather than ordinary tooth texture features, can effectively avoid misjudging complex tooth textures as cracks when outputting recognition results, thus improving the accuracy and reliability of recognition. By adopting the above method, the influence of tooth texture features on the identification of tooth cracks can be reduced. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a visual recognition method for tooth cracks according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the determination of a subset of texture features according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of a subset of focused features according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a visual recognition system for tooth cracks according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a visual recognition method for tooth cracks according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a visual recognition method for tooth cracks according to some embodiments of this application. The visual recognition method for tooth cracks mainly includes the following steps: In step 101, multi-channel visual images of the entire target tooth are acquired.

[0020] In practice, a multi-view camera module for oral cavity use is equipped with 3-5 distributed miniature high-definition cameras. With the assistance of low-brightness cold light source to avoid interference from saliva reflection, the camera module is inserted into the oral cavity and aimed at the target tooth. The angle of each camera is adjusted by a robotic arm or by manual operation to focus on key areas such as the crown surface, enamel margin, interproximal surfaces, occlusal surfaces, and neck of the tooth. Visual images of 1080P or higher resolution and 30 frames per second are acquired simultaneously from each view. All the images from the view together form a multi-channel visual image covering the entire area of ​​the inner and outer surfaces of the target tooth, the occlusal surface, and the interproximal spaces, ensuring that no high-incidence or hidden areas of cracks are missed.

[0021] It should be noted that the multi-channel visual images of this application represent a set of visual images of different key areas of the target tooth, with each channel corresponding to high-definition image data of a specific acquisition viewpoint or area; reflecting the complete visual information of the entire target tooth, covering the original visual signals of weak features such as shallow microcracks.

[0022] In step 102, the multi-channel visual image is preprocessed to obtain a preprocessed multi-channel visual image.

[0023] In some embodiments, preprocessing the multi-channel visual image to obtain a preprocessed multi-channel visual image can be achieved by the following steps: The multi-channel visual image is subjected to de-reflection processing to obtain a de-reflection processed multi-channel visual image; The de-reflection processed multi-channel visual image is enhanced to obtain an enhanced multi-channel visual image. The enhanced multi-channel visual image is then denoised to obtain a preprocessed multi-channel visual image.

[0024] In practice, the process begins with deglare processing. Based on the image grayscale histogram, the pixel brightness distribution of each channel is analyzed to identify highlight reflective areas with grayscale values ​​higher than 240. In oral scenes, reflective areas formed by saliva and direct cold light are also identified. An adaptive threshold segmentation algorithm separates these reflective areas from normal tooth tissue. Local exposure compensation is then applied to the segmented reflective areas, linearly reducing their brightness based on the average brightness of the surrounding normal tissue while maintaining the pixel values ​​of non-reflective areas. This ensures that reflective interference in each channel is suppressed, resulting in a deglare-treated multi-channel visual image. Next, enhancement processing is performed using a contrast-limited adaptive histogram equalization algorithm. Each channel image is divided into 8×8 pixel local blocks, and a histogram is constructed and equalized for each local block, while the contrast gain is limited to a certain value. Within 2.0, to avoid over-enhancement leading to noise amplification, the horizontal and vertical edge gradients of the image are calculated using the Sobel operator. The edge gradient information is then superimposed on the equalized image with a weight of 1:4 to sharpen and enhance the details of the crack edges, resulting in an enhanced multi-channel visual image that highlights the crack features. Finally, denoising is performed. First, a Gaussian low-pass filter with a 3×3 convolution kernel is used to smooth the enhanced image with a standard deviation of 0.8 to 1.2, filtering out Gaussian noise introduced by the sensor. Then, a median filter with a 5×5 window is used for secondary processing to replace isolated salt-and-pepper noise points in the image. During the processing, the relative positions of the crack edge pixels within the window remain unchanged. Finally, a pre-processed multi-channel visual image with reduced reflection interference, highlighted crack details, and sufficient noise suppression is obtained. Other methods can be used in other embodiments, which are not limited here.

[0025] It should be noted that the preprocessed multi-channel visual images in this application reflect complete visual information of the target tooth with no interference and prominent crack details.

[0026] In step 103, multi-scale visual features of the target tooth are extracted from the preprocessed multi-channel visual image, and tooth texture features are filtered from the multi-scale visual features to obtain a subset of texture features of the target tooth.

[0027] In some embodiments, extracting multi-scale visual features of the target tooth from the preprocessed multi-channel visual image can be achieved using the following steps: Multi-scale feature extraction is performed on the preprocessed multi-channel visual image to obtain image features at each scale; The multi-scale visual features of the target tooth are determined based on image features at various scales.

[0028] In specific implementation, multi-scale feature extraction is performed on the preprocessed multi-channel visual image to obtain image features at each scale. This can be achieved in the following way: Based on the consistent brightness and prominent crack details of the preprocessed multi-channel visual image, features are extracted sequentially at three scales: macro, meso, and micro. At the macro scale, the Canny edge detection algorithm is used. First, the gray-level mean of each channel image is calculated, and an adaptive high threshold is set to 1.8–2.2 times the gray-level mean, and a low threshold is set to 0.4–0.6 times the high threshold. This detects the overall contour edge of the tooth. Then, a polygon fitting algorithm is used to fit the contour curve, extracting features such as the continuity of the contour, curvature abrupt change points, and contour fracture length, reflecting whether there are obvious fracture-like cracks in the overall tooth structure. At the meso scale, a 5×5 window gray-level co-occurrence matrix algorithm is used, with features extracted at four latitudes: 0°, 45°, 90°, and 135°. Five texture parameters—contrast, correlation, energy, entropy, and inverse moment—are calculated in the canonical direction to capture the natural texture and distribution uniformity of the enamel surface, and to identify texture distortion, interruption, or abnormal interlacing and crack-related features. At the microscale, a 3×3 window local variance calculation combined with Laplacian edge enhancement technology is used. First, the gray-level variance of pixels within each window is calculated, and regions with variances greater than a set threshold are selected as potential microcrack regions. The threshold is based on 1.5 to 2.0 times the overall gray-level variance of the image. Then, the Laplacian operator is used to enhance the edges of these regions, extracting the amplitude and distribution range of subtle gray-level fluctuations and the continuity of minute edges, capturing the weak visual signals corresponding to shallow and hidden microcracks. Finally, image features at each scale are obtained. Other methods can be used in other embodiments, which are not limited here.

[0029] In specific implementation, the multi-scale visual features of the target tooth can be determined based on the image features at various scales in the following way: a feature fusion strategy is adopted, firstly integrating the image features at various scales, taking the mean of the feature parameters in the image features at various scales to retain the common features of the whole domain, taking the extreme values ​​of the feature parameters of the same scale in each channel to retain the key difference features, and then stitching the integrated macroscopic contour features, mesoscopic texture features, and microscopic fine features in sequence according to the feature dimensions to form the multi-scale visual features of the target tooth that cover the three levels of the overall tooth structure, surface texture, and fine defects, and integrates the whole domain information of all channels. Other methods can also be used in other embodiments, which are not limited here.

[0030] It should be noted that the image features in this application represent the basic quantitative information of the visual attributes of the tooth, reflecting the intuitive visual differences between the surface and internal structure of the tooth; the multi-scale visual features represent the visual feature set of the target tooth, which is classified and integrated according to the "macro, meso, and micro" levels. Among them, the macro-scale features correspond to the overall structural attributes of the tooth, such as the continuity of the overall contour, the curvature change point, and the length of the contour break, reflecting whether there are obvious fracture-like cracks; the meso-scale features correspond to the surface texture attributes such as the direction of the natural texture of the enamel, the uniformity of distribution, and the degree of texture distortion, reflecting whether there are texture interruptions or abnormal intersections related to cracks; the micro-scale features correspond to the fine visual attributes such as the pixel-level fine gray-scale fluctuation range, the small edge continuity, and the abnormal areas of local variance, reflecting whether there are shallow microcracks, hidden cracks on the interproximal surfaces of interdental spaces and the grooves of the occlusal surfaces, clearly presenting the natural visual characteristics of normal tooth tissue, while highlighting the abnormal visual signals corresponding to different types of cracks.

[0031] In some embodiments, reference Figure 2 As shown, this figure is an exemplary flowchart for determining a subset of texture features in some embodiments of this application. In this embodiment, the tooth texture feature selection of the multi-scale visual features to obtain a subset of texture features of the target tooth can be achieved by the following steps: In step 1031, a texture reference datum for normal tooth structure is obtained; In step 1032, the texture features of the multi-scale visual features are compared and filtered based on the texture reference benchmark to obtain the texture features at each scale. In step 1033, the texture features at each scale are integrated to obtain a subset of texture features of the target tooth.

[0032] In practice, firstly, over 1000 multi-channel visual images of normal teeth without cracks are collected. After preprocessing including de-reflection, enhancement, and denoising, and multi-scale feature extraction at macro, meso, and micro levels, sufficient multi-scale visual feature samples of normal teeth are obtained. A texture reference baseline is established using statistical analysis methods. At the macro scale, polygon fitting is performed on the contour of the normal tooth. The normal threshold (mean + 3 standard deviation) is determined by statistically analyzing the mean and three times the standard deviation of the contour fracture length. The density threshold is then calculated as the ratio of the number of curvature abrupt change points to the total contour length. At the meso scale, a 5×5 window gray-level co-occurrence matrix with a step size of 1 pixel is used to calculate the contrast and entropy of normal teeth in four directions: 0°, 45°, 90°, and 135°. Five types of parameters are used: contrast, correlation, energy, and inverse moment. The mean ± 3 standard deviations of each parameter are used to construct a normal parameter range, i.e., the range from mean - 3 standard deviations to mean + 3 standard deviations. For example, contrast is 0.2~0.8, entropy is 1.0~2.5, correlation is 0.6~0.9, energy is 0.1~0.3, and inverse moment is 0.4~0.7. At the microscale, the local gray-level variance of the normal tooth image is calculated using a 3×3 window that slides pixel by pixel. The mean and 3 standard deviations of all window variances are statistically analyzed to determine the upper limit of the normal gray-level fluctuation variance, i.e., mean + 3 standard deviations. Other implementations can be used in other embodiments, which are not limited here.

[0033] Furthermore, in specific implementation, the texture feature comparison and screening of the multi-scale visual features based on the texture reference benchmark to obtain the texture features at each scale can be achieved in the following way: For each scale feature of the multi-scale visual features, the macro-scale uses a contour continuity verification algorithm. First, polygon fitting is performed on the contour of the target tooth. Then, the fracture length and curvature change point density of the fitted contour are compared with the benchmark threshold in the texture reference benchmark one by one, and abnormal contour texture features with fracture length > benchmark threshold and curvature change point density > benchmark threshold are retained. The meso-scale uses a texture parameter threshold comparison method to calculate the meso-scale texture of the target tooth window by window. The five parameters of the gray-level co-occurrence matrix are used to filter out texture distortion, interruption, or abnormal interlacing features where any parameter exceeds the normal range in the texture reference benchmark. At the microscale, a local variance anomaly detection algorithm is used to calculate the local gray-level variance of the target tooth image by sliding a 3×3 window pixel by pixel. Small gray-level fluctuation regions with variances greater than the upper limit of the normal gray-level fluctuation variance in the texture reference benchmark are retained, along with the small edge features in these regions enhanced by the Laplacian operator. This yields abnormal texture features that differ from normal tooth texture at each scale. Finally, texture features at each scale are obtained. Other methods can be used in other embodiments, which are not limited here.

[0034] In addition, in specific implementation, the texture features of various scales are integrated to obtain a subset of texture features of the target tooth body. This can be achieved by the following method: using feature dimension splicing and redundancy removal, the abnormal texture features selected at the macro, meso, and micro scales are first spliced ​​in the order of "macro contour features, meso texture features, and micro fine features". Then, the cosine similarity between feature parameters is calculated and the feature parameters with a similarity ≥ 0.95 are judged as duplicate features to remove duplicate and redundant feature parameters. Finally, a subset of texture features of the target tooth body that focuses on crack-related abnormal textures and has no redundant information is formed. Other methods can also be used in other embodiments, which are not limited here.

[0035] It should be noted that the texture reference benchmark in this application represents the set of macroscopic contour parameter thresholds, mesoscopic gray-level co-occurrence matrix parameter ranges, and microscopic local gray-level variance upper limits for a crack-free normal tooth sample. It reflects the normal value range of texture features of crack-free normal teeth at different scales and is a quantitative judgment standard for distinguishing normal physiological textures from crack-related abnormal textures. Texture features represent multi-scale quantitative visual information of the target tooth's macroscopic contour continuity, mesoscopic texture distribution uniformity, and microscopic gray-level fluctuation amplitude. It reflects all texture morphology and structural attributes of the target tooth surface from the overall to the microscopic level. The texture feature subset represents the set of texture features of the target tooth's macroscopic contour continuity, mesoscopic texture distribution uniformity, and microscopic gray-level fluctuation amplitude. It reflects visual signals on the target tooth that are highly correlated with cracks and can be used to analyze cracks in the target tooth.

[0036] In step 104, historical tooth crack identification results are obtained, and tooth crack focusing processing is performed on the texture feature subset based on the tooth crack identification features in the historical tooth crack identification results to obtain the focused feature subset of the target tooth crack.

[0037] It should be noted that the historical tooth crack identification results in this application represent a dataset generated after various clinically diagnosed tooth crack cases have undergone a process consistent with the target tooth, including multi-channel visual image acquisition, preprocessing, multi-scale visual feature extraction, and texture feature screening. This dataset reflects the characteristic patterns of different types of tooth cracks and the quantitative correlation between features and cracks. The historical tooth crack identification results include clinically validated basic case information, namely tooth location, crack type, crack depth and length, multi-scale abnormal texture feature parameters of the corresponding cases, namely macroscopic contour fracture length, mesoscopic gray-level co-occurrence matrix abnormal parameters, microscopic local gray-level variance constants, the correlation coefficient between features and cracks, namely the Pearson correlation coefficient, and the final clinical crack determination result, namely whether a crack exists and its grade.

[0038] In some embodiments, reference Figure 3As shown, this figure is an exemplary flowchart for determining a subset of focused features in some embodiments of this application. In this embodiment, the subset of texture features is processed to focus on tooth cracks based on the tooth crack identification features in the historical tooth crack identification results to obtain the subset of focused features of the target tooth crack. This can be achieved by the following steps: In step 1041, the tooth crack identification features of the historical tooth crack identification results are determined; In step 1042, the tooth crack identification features are compared one by one with the texture feature subset to filter out each relevant texture feature that focuses on the tooth crack; In step 1043, the focused feature subset of the target tooth crack is determined by using all relevant texture features.

[0039] In specific implementation, the tooth crack identification features of the historical tooth crack identification results can be determined in the following way: Extract the feature representations corresponding to different types of cracks from the historical tooth crack identification results, including the macroscopic contour fracture length and curvature abrupt change point distribution characteristics corresponding to obvious fracture cracks, the mesoscopic texture distortion parameters and texture interruption location characteristics corresponding to hidden cracks, and the microscopic local gray-scale variance abnormality amplitude and minute edge continuity characteristics corresponding to shallow microcracks. Then, calculate the correlation between each feature and the corresponding crack type through Pearson correlation verification, and eliminate redundant features with low correlation. That is, take the 20th percentile of all correlation coefficients as the judgment threshold. This threshold represents that only 20% of the features in all statistical features have a lower correlation with the crack than this value. Features below this threshold are considered redundant features with low correlation, and features with strong correlation with the crack are retained. Finally, classify and organize the filtered features according to crack type, clarify the parameter range, manifestation form, and correspondence with the crack for each type of feature, thereby determining the tooth crack identification features in the historical tooth crack identification results. Other methods can also be used in other embodiments, which are not limited here.

[0040] In addition, in specific implementation, the tooth crack identification features are compared one by one with the texture feature subset to select the relevant texture features focusing on tooth cracks. This can be achieved in the following way: First, the tooth crack identification features are classified and organized according to three scales: macroscopic, mesoscopic, and microscopic, forming a multi-scale crack feature benchmark set. At the same time, the texture feature subset of the target tooth is also classified according to the same three scales to ensure that the scale division of the two types of features is completely consistent. Then, according to the principle of comparing one by one at the same scale, each feature in the target tooth texture feature subset is matched with all crack identification features of the same scale in the benchmark set. The cosine similarity algorithm is used to calculate the similarity value of each set of matched features, and the highest similarity value after matching each target feature with the crack identification feature of the same scale is recorded. Then, the calculated highest similarity value is compared with a pre-determined similarity judgment standard. The pre-determined similarity judgment standard is determined by collecting clinically diagnosed tooth crack features. For both sample and non-crack feature samples, the cosine similarity algorithm is used to calculate the similarity values ​​between the two types of samples and the features in the crack typical feature benchmark library. The lower limit of the distribution of crack feature similarity and the upper limit of the distribution of non-crack feature similarity are statistically analyzed, and the median value of the two is taken as the initial judgment threshold. Then, a sufficient number of clinical cases covering different crack types and non-crack interference factors are used for back-checking and verification. The threshold is adjusted until the accuracy of crack feature recognition and the false detection rate of non-crack feature both meet the preset clinical standards. Finally, a stable and reproducible similarity judgment standard is determined to screen out target features with similarity values ​​higher than the standard. These features are the texture features that are highly related to tooth cracks. At the same time, target features with similarity values ​​lower than the standard are removed. These features are mostly non-crack related texture features such as physiological tooth depressions and mild enamel wear. Finally, all the screened high-matching features are summarized to obtain various relevant texture features focused on tooth cracks. Other methods can be used in other embodiments, which are not limited here.

[0041] In addition, in specific implementation, the determination of the focused feature subset of the target tooth crack through all relevant texture features can be achieved in the following way: all relevant texture features are classified and organized according to three scales: macro, meso, and micro, ensuring that the hierarchical division of features is completely consistent with the logic of the previous multi-scale feature extraction; then, weights are assigned based on the matching similarity between each feature and the typical crack features, with features with higher matching degrees being assigned higher weight coefficients, thereby strengthening the recognition of highly correlated features; next, the well-known cosine similarity algorithm is used to calculate the similarity between features within the same scale, and features with similarity reaching a preset high value are judged as duplicate and redundant features and removed to avoid the duplication of feature information; finally, the high-weight crack-related features of each scale after weight allocation and redundancy removal are systematically integrated in the order from macro to micro to form a focused feature subset that is highly focused, free of redundant interference, and can accurately characterize the target tooth crack. Other methods can also be used in other embodiments, which are not limited here.

[0042] It should be noted that the tooth crack identification features in this application represent a set of quantitative features extracted from the clinically diagnosed historical tooth crack identification results, reflecting the visual signal patterns of different types of tooth cracks; the related texture features represent features in the target tooth texture feature subset that have a similarity to the tooth crack identification features that meet the preset judgment criteria, reflecting the texture signals of suspected cracks on the target tooth; the focused feature subset reflects the focused visual features on the target tooth that are highly correlated with cracks, which can be used to provide core basis for the identification of tooth cracks.

[0043] In step 105, shallow microcracks and hidden cracks of the target tooth are identified based on the focused feature subset, so as to output the identification result of whether there are cracks in the target tooth.

[0044] In some embodiments, crack identification of shallow microcracks and occult cracks in the target tooth based on the focused feature subset, and outputting the identification result of whether the target tooth has cracks, can be achieved by the following steps: Determine the crack identification rules for shallow microcracks and hidden cracks in the target tooth structure; Based on the crack determination rule, feature recognition is performed on the focused feature subset to obtain feature recognition scores at each scale; The presence or absence of cracks in the target tooth is determined by feature recognition scores at various scales.

[0045] In practice, the crack determination rules for identifying superficial microcracks and occult cracks in the target tooth can be implemented as follows: First, collect historical cases of clinically diagnosed superficial microcracks and occult cracks, extract the corresponding historical focused feature subsets of these cases, and use statistical analysis methods to classify and statistically analyze the features at each scale in the historical focused feature subsets to clarify the correlation between different scale features and the two types of cracks. Based on this, the crack determination rules are determined. Specifically, the microscale rule focuses on superficial microcracks, clarifying the abnormal local grayscale fluctuations and the normality or abnormality of the continuity of minute edges associated with superficial microcracks at this scale. The rules are defined as follows: The mesoscale rules focus on hidden cracks, clarifying the degree of texture distortion and the characteristic range of texture interruption locations associated with hidden cracks at that scale; the macroscale rules serve as an auxiliary judgment basis, clarifying the abnormal judgment criteria for the distribution of contour curvature abrupt change points, and assigning corresponding weights based on the influence of each scale feature on crack identification. The microscale is given the highest weight because it is directly related to the core features of shallow microcracks, the mesoscale is given the second highest weight because it corresponds to the key features of hidden cracks, and the macroscale is given the lowest weight because it only plays an auxiliary role; thus, the crack judgment rules for shallow microcracks and hidden cracks in the target tooth are obtained.

[0046] Furthermore, in specific implementation, feature recognition of the focused feature subset based on the crack judgment rules to obtain feature recognition scores at each scale can be achieved in the following way: A weighted feature matching algorithm is used to compare the microscopic, mesoscopic, and macroscopic features in the target tooth's focused feature subset with the corresponding crack judgment rules item by item. The abnormal features at each scale in the crack judgment rules are decomposed into clear quantitative indicators and morphological representations. For example, the microscopic scale corresponds to the local grayscale fluctuation range and the integrity of the continuity of small edges in shallow microcracks; the mesoscopic scale corresponds to the degree of texture distortion and the location distribution of texture interruptions in hidden cracks; and the macroscopic scale corresponds to the distribution density of contour curvature abrupt change points in the auxiliary judgment. Simultaneously, a unified full score benchmark is set for each feature matching. Then, the features to be matched in the target tooth's focused feature subset are compared item by item with the corresponding abnormal features in the rules. Yes, if all quantitative indicators and morphological representations of the feature to be matched are completely consistent with the description of the abnormal features in the rules, without any deviation, then the feature is directly given a full score. If only some quantitative indicators or morphological representations of the feature to be matched meet the requirements of the abnormal features in the rules, such as the length of the texture interruption not reaching the completely abnormal standard defined by the rules but exceeding the normal benchmark, or the amplitude of local grayscale fluctuations being in the middle range between normal and completely abnormal, then a known proportional conversion method is used. Based on the proportion of the number of indicators that meet the requirements of the rules to the total number of indicators of the feature, or the proportion of the degree of performance of the conforming part to the degree of performance of the completely abnormal part, this proportion is used as the score of the feature. This ensures that the score result can accurately reflect the actual matching degree between the feature and the abnormal judgment rule. After completing the comparison of all features, the feature recognition scores at the micro, meso, and macro scales are summarized and statistically analyzed respectively.

[0047] In addition, in specific implementation, the determination of whether a target tooth has a crack by using feature recognition scores at various scales can be achieved as follows: The weight allocation is determined based on the degree of correlation between features at each scale and shallow microcracks and occult cracks. Microscale features directly correspond to the core visual signals of shallow microcracks and are given the highest weight; mesoscale features correspond to the key manifestations of occult cracks and are given the second highest weight; macroscale features are only used as auxiliary criteria and are given the lowest weight. Then, the feature recognition scores at the micro, meso, and macroscale scales are multiplied by their respective weights, and the three products are summed sequentially to obtain the comprehensive score of the target tooth. Determining the comprehensive score threshold requires collecting a sufficient number of clinical cases covering different tooth positions, age groups, and tooth conditions, including cases of shallow microcracks, occult cracks, and normal teeth without cracks diagnosed by professional instruments. The focused feature subsets of these cases are substituted into the above scoring calculation process, and the comprehensive score distribution of all cracked cases is statistically analyzed. The upper limit of the comprehensive score distribution of the lower limit and the upper limit of the crack-free cases are taken as the midpoint as the initial threshold. Then, the initial threshold is substituted into all clinical cases for back-testing and verification. The crack identification accuracy and the false detection rate of non-crack cases under this threshold are statistically analyzed. The threshold value is repeatedly fine-tuned until the accuracy and false detection rate both reach the clinically preset acceptable standard, thus determining the final comprehensive score judgment threshold. When judging cracks, if the comprehensive score of the target tooth is higher than the final threshold, the target tooth is judged to have shallow microcracks or hidden cracks. At the same time, the scores of the micro, meso and macro scales are compared. The scale with the highest score is used as the core judgment criterion. If the micro scale score is the highest, it is marked as a shallow microcrack. If the meso scale score is the highest, it is marked as a hidden crack. If the macro scale score is the highest, the scores of the micro and meso scales are combined for auxiliary judgment. If the comprehensive score of the target tooth is lower than the final threshold, the target tooth is judged to not have shallow microcracks or hidden cracks. The final output includes the recognition result including whether a crack exists, the specific crack type, and the score details of each scale.

[0048] It should be noted that the crack determination rules in this application represent the focused features of clinically diagnosed shallow microcracks and occult cracks in historical cases, reflecting the corresponding correlation between features at different scales and the two types of cracks; the feature recognition score represents the quantified score of the target tooth crack features belonging to the crack at different scales, reflecting the degree of matching between the crack features at each scale of the target tooth and the abnormal crack features; the recognition result represents the identification result of whether the target tooth has shallow microcracks or occult cracks, as well as the specific crack type, reflecting the crack detection status of the target tooth, and can provide accurate data support for clinical tooth crack diagnosis.

[0049] In another aspect, in some embodiments, this application provides a visual recognition system for tooth cracks, referring to... Figure 4The figure is a schematic diagram of the structure of a visual recognition system for tooth cracks according to some embodiments of this application. The visual recognition system 400 for tooth cracks includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: Acquisition module 401, in this application, is mainly used to acquire multi-channel visual images of the entire target tooth body; Processing module 402, in this application, is used to preprocess the multi-channel visual image to obtain a preprocessed multi-channel visual image; It should be noted that the processing module 402 in this application is also used to extract multi-scale visual features of the target tooth from the preprocessed multi-channel visual image, and to perform tooth texture feature filtering on the multi-scale visual features to obtain a subset of texture features of the target tooth. In addition, it should be noted that the processing module 402 in this application is also used to obtain historical tooth crack recognition results, and to perform tooth crack focusing processing on the texture feature subset based on the tooth crack recognition features in the historical tooth crack recognition results to obtain the focused feature subset of the target tooth crack. The execution module 403 in this application is mainly used to identify shallow microcracks and hidden cracks in the target tooth based on the focused feature subset, so as to output the identification result of whether the target tooth has cracks.

[0050] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described visual recognition method for tooth cracks.

[0051] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device for implementing a visual recognition method for tooth cracks according to some embodiments of this application. The visual recognition method for tooth cracks in the above embodiments can be achieved through... Figure 5 The computer device 500 shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0052] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0053] The communication bus 502 can be used to transmit information between the aforementioned components.

[0054] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0055] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The method used in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0056] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0057] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0058] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0059] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for visual recognition of tooth cracks.

[0060] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0061] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for visually recognizing tooth cracks, characterized in that, Includes the following steps: Acquire multi-channel visual images of the entire target tooth structure; The multi-channel visual image is preprocessed to obtain a preprocessed multi-channel visual image; Multi-scale visual features of the target tooth are extracted from the preprocessed multi-channel visual image, and tooth texture features are filtered from the multi-scale visual features to obtain a subset of texture features of the target tooth. Obtain historical tooth crack identification results, and perform tooth crack focusing processing on the texture feature subset based on the tooth crack identification features in the historical tooth crack identification results to obtain the focused feature subset of the target tooth crack; Based on the focused feature subset, crack identification is performed on the shallow microcracks and hidden cracks of the target tooth to output the identification result of whether the target tooth has cracks.

2. The method as described in claim 1, characterized in that, Preprocessing the multi-channel visual image to obtain the preprocessed multi-channel visual image specifically includes: The multi-channel visual image is subjected to de-reflection processing to obtain a de-reflection processed multi-channel visual image; The de-reflection processed multi-channel visual image is enhanced to obtain an enhanced multi-channel visual image. The enhanced multi-channel visual image is then denoised to obtain a preprocessed multi-channel visual image.

3. The method as described in claim 1, characterized in that, Extracting multi-scale visual features of the target tooth from the preprocessed multi-channel visual image specifically includes: Multi-scale feature extraction is performed on the preprocessed multi-channel visual image to obtain image features at each scale; The multi-scale visual features of the target tooth are determined based on image features at various scales.

4. The method as described in claim 1, characterized in that, The tooth texture feature subset obtained by filtering the multi-scale visual features to obtain the texture features of the target tooth specifically includes: Obtain a texture reference for normal teeth; Based on the texture reference benchmark, the texture features of the multi-scale visual features are compared and filtered to obtain the texture features at each scale. By integrating the texture features at various scales, a subset of texture features of the target tooth is obtained.

5. The method as described in claim 1, characterized in that, The historical tooth crack identification results include clinically validated basic case information, multi-scale abnormal texture feature parameters of the corresponding cases, correlation coefficients between features and cracks, and the final clinical crack determination results.

6. The method as described in claim 1, characterized in that, Based on the tooth crack identification features in the historical tooth crack identification results, the texture feature subset is subjected to tooth crack focusing processing to obtain the focused feature subset of the target tooth crack, which specifically includes: Determine the tooth crack identification features of the historical tooth crack identification results; The tooth crack identification features are compared one by one with the texture feature subset to filter out the relevant texture features that focus on tooth cracks. The focused feature subset of the target tooth crack is determined by identifying all relevant texture features.

7. The method as described in claim 1, characterized in that, Based on the focused feature subset, crack identification is performed on the shallow microcracks and hidden cracks of the target tooth to output the identification result of whether the target tooth has cracks. Specifically, this includes: Determine the crack identification rules for shallow microcracks and hidden cracks in the target tooth structure; Based on the crack determination rule, feature recognition is performed on the focused feature subset to obtain feature recognition scores at each scale; The presence or absence of cracks in the target tooth is determined by feature recognition scores at various scales.

8. A visual recognition system for tooth cracks, characterized in that, include: The acquisition module is used to acquire multi-channel visual images of the entire target tooth structure; The processing module is used to preprocess the multi-channel visual image to obtain a preprocessed multi-channel visual image. The processing module is further configured to extract multi-scale visual features of the target tooth from the preprocessed multi-channel visual image, and to perform tooth texture feature filtering on the multi-scale visual features to obtain a subset of texture features of the target tooth. The processing module is also used to obtain historical tooth crack recognition results, and to perform tooth crack focusing processing on the texture feature subset based on the tooth crack recognition features in the historical tooth crack recognition results to obtain the focused feature subset of the target tooth crack. The execution module is used to identify shallow microcracks and hidden cracks in the target tooth based on the focused feature subset, and to output the identification result of whether the target tooth has cracks.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the visual recognition method for tooth cracks as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the visual recognition method for tooth cracks as described in any one of claims 1 to 7.

Images